# PromptEye Help — complete knowledge base # Prompt Intelligence Source: /help/prompt-intelligence/ --- title: Prompt Intelligence description: Choose and organize the questions PromptEye checks in AI answers, then understand what the results say about your brand. --- Prompt Intelligence helps you decide which questions are worth tracking, organize them around customer needs, and see whether your brand appears when AI models answer them. A prompt is the question PromptEye checks with the selected AI models. For example, “Which project management tool works best for a small design agency?” is a prompt; “project management tool” is only a search phrase. Use this page to build a useful set of customer questions, spot gaps in the topics you monitor, and open the underlying AI answers when a number needs investigation. Prompt Intelligence measures your brand's presence in AI answers. It does not measure visits to your website; use [Traffic](/help/ai-traffic/) for website visits. ## Before you start Choose the project you want to work on from the project selector. Each project has its own prompts, groups, and results. If you cannot add or change prompts, your project access may be read-only; ask a project owner or editor to make the change. PromptEye asks the AI models every monitored prompt regularly: once a day on most plans, and every two days on some plans. A new prompt has results after its first check. Prompts are checked for the market (country) chosen when the project was created; the market cannot be changed later, so create a separate project for another market. The prompt limit comes from the plan of the workspace the project belongs to, and it covers all projects in that workspace. When you add a prompt, the add screen shows how many slots are left. If the limit is reached, the app says so and suggests upgrading the plan. Archiving or deleting a monitored prompt frees its slot. Some evaluations and visibility results may be pending while PromptEye prepares them. The page can also show the last available visibility results while an update is in progress. If an evaluation shows an error rather than pending, open the prompt and check again later; contact support if it stays in error. ## Read the page from top to bottom ### Model and period Use the controls at the top of the page to choose the AI model and the period for visibility results. You can choose a custom date range of up to 60 days. Changing these controls updates the answer-based results. Change markers show how the result differs from the comparison period; they are hidden when either period has fewer than 7 measured days. A model marked as unavailable requires an add-on. The period and comparison options work as on [Visibility](/help/visibility/#set-the-model-and-period). These controls do not change which prompts are monitored. They also do not change the funnel coverage table, which summarizes the active prompt set. If a funnel gap remains after changing the date range, it means the set itself has no prompt assigned to that stage. ### Summary tiles The summary tiles help you decide where to look first. Hover or focus each tile's information icon for its exact calculation. - **Business priority** is a recommendation about a prompt's potential, based on its buying intent and estimated volume. It is not a guarantee of leads or sales. You can set your own 1–5 priority for an individual prompt; a manual choice replaces the recommendation until you switch back to the recommended value. - **Prompt Health** summarizes how well the prompt set balances fit with your company, estimated volume, and buying intent. A low result suggests checking whether the prompt is relevant to your offer and reflects a real customer question. - **AI Traffic** is an estimate of the monthly potential associated with the tracked questions, not a count of actual AI users or website visits. This tile is shown when the feature is available for your account. See [How to read prompt potential](#how-to-read-prompt-potential) for what the estimate can and cannot tell you. - **Company fit** is an AI assessment of how closely the prompts match the project's company profile, category, and offer. If it looks wrong, review the company information used by the project and the wording of the prompt. - **Prompts in the set** counts monitored prompts and groups, and shows how many prompts are still missing a complete evaluation. “Pending” means the evaluation is incomplete; it does not by itself mean the prompt failed. AI Traffic is an optional tile, so the number of tiles can vary by account. ### How to read prompt potential The AI Traffic value helps you compare the potential of questions in your monitored set. On this page, the summary tile gives an overall view and a typical per-prompt figure; each prompt row has its own estimate. Summary totals skip prompts in the **Error** state and combine estimates from the other prompts. An orange warning beside a total means the estimate is missing for some prompts; hover over it for details. If no usable estimate remains, the total is unavailable. The same rule applies to totals in Visibility. The estimate does not count people who actually asked an AI assistant that exact question. It does not predict how often a model mentions your brand, whether your page was cited, or how many people visited your website. Use **Visibility** and **Citations** for answer results, and [Traffic](/help/ai-traffic/) for visits to your site. For example, a broad question such as “Which project management app is best?” may show a different estimate from “Which project management app works for a 12-person design studio?” The estimate alone does not explain why two questions differ, and broader wording is not automatically better. Compare the figures as one input when choosing what to monitor, but keep specific, high-intent questions even if their estimate is lower. A smaller estimate does not mean the question is unimportant to your customers. For an individual prompt, the value may be shown as a number, **<50**, **0**, **No data**, **Error**, **Outdated**, or a dash. **0** means a zero estimate was saved; **No data** means there was not enough information to show an estimate, for example because none of the search phrasings of the prompt has a reported search volume; **Error** means the check failed; **Outdated** means the last check failed and an earlier value is still shown. A dash can mean no estimate has been saved yet. These states are different: a missing or zero estimate does not mean the prompt itself has no value. If the details control appears next to a value, open it to review the status. Where **Recalculate** is available, it asks PromptEye to check that prompt again; it does not create another prompt or use another prompt slot. The size of the estimate does not change how often the prompt is checked; that depends only on the plan. The estimate does not use an extra prompt slot. Adding a prompt does use a slot, whether its estimate is ready or not. You can add and monitor a relevant question without waiting for a potential estimate; a missing estimate does not block adding it or make it unusable. Plan limits apply to the prompts you monitor, not to whether each prompt has an estimate. Changing a prompt's wording changes the question PromptEye checks from its next check. Answers collected for the old wording stay in the same prompt's history, so the prompt's results then mix two questions. If you want a clean comparison, add the new wording as a separate prompt and archive the old one. After a wording change, PromptEye evaluates the intent, company fit, and AI Traffic again, so those values may show pending for a while; a stage you set yourself in the same edit is kept. You cannot change the wording to a question that is already monitored in the project. The edit form also lets you change its group, tags, and funnel stage, which changes where it appears in those views. If you disagree with the automated business priority, the manual priority setting changes the priority used for that prompt; it does not change the AI Traffic estimate. When business priority is still pending, its edit control may not yet be available. Use the estimate together with the funnel, not as a replacement for it. The funnel tells you which customer-decision stages your prompts cover; the estimate helps compare the potential of questions you might add to fill a gap. It does not assign a prompt to a funnel stage or tell you what customers ask at that stage. If a question has high estimated potential but low visibility, use **Full analysis in Visibility** to inspect the answers for that question. That combination is a reason to investigate, not proof that customers cannot find your brand or that you are losing sales. Low potential with high visibility means your brand appears in the checked answers for a question whose estimate is smaller; decide whether to keep it based on its relevance to customers as well as its estimate. The estimate is not a sales forecast or a measure of your total market. It is not broken down by audience segment, and it cannot confirm that it represents a particular niche or every country and language you care about. Check that the prompt wording fits your intended customer and review the project's market in **Project settings**. ### Funnel coverage by group Customers ask different questions as they move from learning about a problem to choosing a provider. PromptEye assigns each prompt an intent stage: **Awareness**, **Consideration**, **Comparison**, or **Decision**. The table has one row per group and one column per stage. A number is the number of active prompts in that group assigned to that stage; a dash means the group has no prompt there. For example, if a “Pricing” group has prompts in Awareness and Comparison but a dash under Decision, add a question a customer might ask just before choosing, such as “Which [type of product] has the best support for a 20-person team?” A missing stage is a gap in what you monitor, not proof that your brand is invisible at that stage. The summary above the table tells you how many groups cover all four stages and names the least-covered stage. Click a group row to open that group's details. The funnel is based on active prompts and their assigned intent; the model and date controls do not filter it. Intent is usually evaluated automatically, but you can change a prompt's stage yourself from its **Edit** action. That changes the stage count for its group. ### Groups and tags A **group** represents a customer need or topic, such as Pricing, Comparisons, or Use cases. Prompt Intelligence uses groups to show funnel coverage and to summarize results for related questions. Put questions together when you want to compare them as part of the same customer decision. An ungrouped prompt appears in a shared “No group” row, which makes its place in your analysis less clear. **Tags** are extra labels that can cross-cut groups, such as “e-commerce” or “enterprise.” They help you find prompts by subject without changing their group or funnel role. A prompt with several tags appears in each matching bucket when you group the table by tags; this is only another way to view it, not extra monitored prompts. Clicking a group in the funnel or table opens a detail panel. From there, open the full group page to see its summary, funnel, suggestions, and prompts. Group summaries combine the prompts currently in that group; open individual prompts when you need to understand differences that a group-level summary cannot show. A group is useful for understanding a need across several questions; a tag is useful for finding related questions across different needs. Groups are shared across the app. The same groups organize the group tables on [Visibility](/help/visibility/) and [AI Ads](/help/ai-ads/), and articles in [Content](/help/content/) can be linked to a group. Renaming a group or moving prompts changes those views too. If you delete a group, its prompts are not deleted; they move to the shared “No group” row, and content linked to that group loses its group link. ### Group suggestions On a group page, **Generate suggestions** proposes new prompts for that group. A **Gap** suggestion fills a funnel stage the group does not cover yet. A **Replicate** suggestion is close in theme to the group's best-performing prompts. Each suggestion shows its estimated AI Traffic, Prompt Health, priority, and why it is proposed now. Choose **Accept and add** to add it to the group, **Edit** to change the wording first, or **Reject**. An accepted suggestion becomes a monitored prompt and uses a slot like any other prompt. If nothing is generated, the page explains why: earlier suggestions are still waiting for your decision, the group has no gap and too few strong prompts to build on, or the workspace plan does not cover suggestions right now. Decide on the waiting suggestions, or come back when the group has new results. ### Prompt table: find and compare questions The table starts with prompts grouped by their assigned group. On the project page, you can switch between grouping by group and grouping by tags. Search can match prompt text, keyword, group, or tag/category. The tag, priority, and Health filters narrow which rows are shown; they do not change the underlying prompts. Clear filters or return to **All** when a prompt seems to be missing. The **Monitored** and **Archive** views answer different questions. Monitored contains prompts currently being checked and used in the active summary. Archive contains prompts you have paused. Archived prompts are not checked and do not count toward the plan's prompt limit, but the prompt remains in the Archive view. Restoring the same prompt resumes checks and requires a free prompt slot in the workspace. There are no results for the days a prompt was archived. The table's main result columns are: - **Visibility**: the share of checked AI answers in the selected period and model where your brand appears. - **Citations**: the share of checked answers that cite a page on your domain. - **Average position**: your brand's average place in the answer's list of brands; 1 is the highest position. - **Business priority** and **Health**: evaluations of the prompt's potential and quality. A pending label means there is not enough completed evaluation yet. Change markers show a difference from comparison data when that data is available; a dash means a value is unavailable for the prompt and selection. Visibility and citations describe the answers PromptEye checked; they are not website traffic metrics. Click a prompt row to inspect its evaluation and open the full analysis in Visibility, where you can review answer details and sources. ### Move, archive, restore, or delete prompts Select one or more rows to move prompts to another group, archive them, restore archived prompts, or delete them. Moving prompts changes how they are organized and which group summaries and funnel counts include them. It does not create a new prompt or change the wording. Archiving pauses the prompt's checks and removes it from active summaries and the prompt count. The prompt stays in Archive so you can restore the same prompt later. Restoring resumes monitoring and is blocked if the workspace has no free prompt slot; the app then shows the plan limit. Deleting removes the selected prompt from the project. The app asks you to confirm, and a deleted prompt cannot be restored. If you may want the prompt or its history later, archive it instead. ## Inspect one prompt Click a prompt row to open its detail panel. It shows the prompt's business priority, Prompt Health, estimated AI Traffic (when enabled), and company fit. The company-fit explanation gives context for the evaluation. The group funnel section marks the prompt's assigned stage and shows how many prompts in its group cover each stage. Use the prompt's **Edit** action to change that stage if the automatic assignment does not fit; this changes the group's funnel coverage. The visibility summary shows the prompt's visibility, average position, and citation share for the selected model and period. Use **Full analysis in Visibility** to see the underlying answer analysis and sources. A visit from a model to your website belongs in Traffic; an appearance or citation in a generated answer belongs here. If you have edit access, you can set a manual business priority from 1 to 5 or return to PromptEye's recommendation. A manual value is saved for the project, so everyone working on it sees it. It drives sorting, filters, and suggestions; the PromptEye recommendation stays visible and you can switch back to it at any time. ## Add prompts that reflect real customer questions Choose **Add → New prompt** for one question or **Add → New group** to create a group first. Adding prompts requires write access and an available prompt slot. A single prompt must have text and a group; tags are optional. Use the question a customer would ask an AI assistant, not a short keyword phrase. For example: - Better: “Which AI visibility tool should a small online store choose?” - Less useful: “AI visibility tools ranking 2026” One entry should cover one question. A prompt such as “What does it cost, how do I install it, and is it worth it?” combines several intents and makes its results harder to interpret. Split it into separate questions and group them if they belong to the same customer need. Choose **Many prompts** to paste one question per line or upload a CSV, XLS, XLSX, or TXT file. The file can have columns for the prompt, the group, and tags, with or without headers. Use **Download the template** for a ready file. If you don't have a list yet, **Copy the LLM prompt** copies instructions you can paste into ChatGPT, Claude, or Gemini; paste the result back into the list. Review the preview before saving: check the prompt text, group, and optional tags. If the file's columns were read incorrectly, use **Change the column mapping**. Prompts that are already monitored, duplicates within the list, and prompts longer than 160 characters are marked and skipped. A group name from the file that does not exist yet is marked **will be created**, and the group is created when you save. Every prompt to be added needs a group. If the list is longer than the free slots in the workspace, the extra rows are left out and the app tells you how many. PromptEye evaluates buying intent, fit with the company profile, estimated volume, Prompt Health, and business priority as data becomes available. These are estimates, not facts about your customers. Use them to decide which questions deserve attention, and edit the prompt or company profile if the explanation does not fit your business. You can correct the assigned intent stage yourself from the prompt's **Edit** action. ## If something looks wrong - **The page is empty:** select a project in the sidebar. If a project is selected, add prompts and wait for the first checks to finish. - **A result shows pending:** the evaluation or visibility update is incomplete. The page may show the last available visibility results during an update; pending is not the same as an error. - **A prompt is missing:** check whether you are viewing Monitored or Archive, clear the search and filters, and check its group or tags. - **You cannot add, move, archive, or delete:** check whether your project access allows editing, or ask someone who can edit the project. Adding and restoring also depend on free prompt slots. The limit is shared by all projects in the workspace, so prompts in other projects can use it up. - **A number seems surprising:** check the selected model and period, then open the prompt's full Visibility analysis to inspect the answers behind the summary. --- # Visibility Source: /help/visibility/ --- title: Visibility description: Read your brand's presence in monitored AI answers, compare competitors, and trace changes to prompts, citations, and sources. --- Visibility helps you answer a practical question: when AI models respond to the questions you monitor, how often do they name your brand, which alternatives appear, and which sources support those answers? Use it to decide what to investigate next, then open the answer itself before treating a change as a business conclusion. Visibility measures the answers PromptEye has checked for one project. It is not a measure of all AI conversations, audience reach, website visits, leads, or sales. **Visibility** means your brand is named in an answer. **Citations** means an answer cites your website or another source. A model can cite your site without naming your brand: that is a citation, but it does not count as visibility. For visits to your site from AI services, see [Traffic](/help/ai-traffic/). To choose and organize monitored customer questions, see [Prompt Intelligence](/help/prompt-intelligence/). ## Before you interpret a result Choose the project whose audience and monitored prompts you want to assess. Each project has its own domain, brand definition, prompts, competitor configuration, and answer history. Results describe this monitored set and the selected model and period; they do not automatically represent every market, language, or customer question. The project needs completed answer checks before the page can show meaningful measures. A newly created project may have no results yet. Prompt groups can also include prompts whose first check is pending; pending prompts may appear in the group table without completed metrics. Loading skeletons mean the page is still fetching or calculating. They are not zero values. A dash means the relevant figure is unavailable for that selection; it does not mean the brand had zero visibility. If the page says it could not calculate the figures, use **Try again**. After a change to the brand names or competitor groups, PromptEye recalculates past results, and the figures may be unavailable for a while. Check again later. The app does not show how long this takes. ## Set the model and period The controls at the top of Visibility apply to the selected model and date range. They change which checked answers feed the summary, competitor comparison, group metrics, sources, and answer-level review. They do not change the project’s monitored prompts. - **All models** combines the available model results. Use it for a broad view of the monitored set. - **One model** isolates that model. Use it to find differences in how providers answer the same questions. - **A custom selection** compares a chosen set of active models. Selecting one model resolves to that model; selecting several creates a custom selection. - **A preset or custom date range** changes the answer history being summarized. The presets go from the last 7 days to **Since start**. A custom range picked in the calendar can cover up to 60 days, and the earliest date you can pick is the start of monitoring. A preset is greyed out when the project's history is shorter than that preset, for example **Last year** in a project monitored for two months. PromptEye remembers your model choice in your browser, so it stays selected when you open another project. The period is also remembered, separately for each screen: a year picked on Visibility does not change the period on AI Ads or Prompt Intelligence. ### Which models you can select ChatGPT, Perplexity, DeepSeek, and Gemini are monitored on every plan. Claude, Grok, AI Overview, Copilot, Google AI Mode, Llama, and ChatGPT Web are add-ons. Models that are not active for the project appear in the list as available but cannot be selected. The plan and add-ons of the workspace the project belongs to decide which models are active, not the plan of the person looking at it. A team member sees the team's models, even if their own account has other add-ons. If you move the project to another workspace in **Project settings**, the new workspace's plan and add-ons apply. A model starts collecting answers after its add-on is turned on, so there are no earlier results for it. When an add-on ends, that model is no longer active for the project and you can no longer select it in the filter. Change indicators (+/−) compare the selected period with a comparison period. They are differences between two measured periods, not forecasts. In the period picker, **Comparison** lets you choose the baseline: - **Previous period** – the same number of days just before the selected range. - **Previous period (same day of week)** – the same length, aligned to the same weekdays, which helps when your market has weekly patterns. - **Previous year (same period)** – the same dates one year earlier. - **Custom range** – any second period you pick in the calendar. The earlier of the two periods becomes the comparison. When you select a whole calendar month, quarter, or year, PromptEye compares it with the previous month, quarter, or year. Use **Switch to days** to compare an equal number of days instead. Changes are hidden when either period has fewer than 7 measured days. This is common in a new project, or with a short custom range; the picker then shows **changes hidden**. If the comparison period starts before monitoring began, the picker marks it as **partial data**, so read those changes with care. Position changes use an improvement-positive direction: a lower average position number is better, so moving from position 4 to 2 is shown as a positive improvement. For visibility, Reach Index, and citation share, a larger percentage is an increase, not proof that business outcomes improved. For a fair comparison, use the same project, prompt scope, model selection, and comparable date ranges. Adding or editing prompts, changing the project’s brand or domain configuration, or changing how competitor names are grouped can alter what later results include. Read the [Answers tab guide](/help/visibility/answers/) when you need to inspect the sampled results behind a percentage. ## Read Overview Overview puts the main signals together so you can decide where to investigate. The figures are calculated from observed answers and may be rounded. ### Summary metrics - **Visibility** is the share of readable, analyzed AI answers in the selected scope that name your brand. One answer means one monitored prompt answered by one model. A citation to your website alone does not count. Answers that could not be read reliably are excluded, not treated as answers without your brand. Treat this as presence in the checked answers, not as the share of all AI answers in the market. When comparing all models, only models that returned an answer contribute observations; a model error is not counted as a brand-free answer. - **Reach Index** weights visibility by each AI model's market share. It helps distinguish presence in a more widely used model from the same presence in a less widely used model. It is still a modeled comparative index, not actual exposure, traffic, or a guarantee of how many people saw an answer. - **AI Traffic** is an estimate associated with the monitored prompt set, displayed for the last 30 days and all prompts combined. It is not a count of visits from AI or a count of people who asked those exact questions. Check [Traffic](/help/ai-traffic/) for recorded website visits. - **Average position** is your brand’s average place among brand mentions in answers that name it. Position 1 is the highest. Answers that do not name your brand do not supply a position. A lower number is a better place; use its change marker with that direction in mind. The sample size shown under a competitor comparison (for example, “N answers”) is the number of answer observations used for that selection. One prompt checked by three models can contribute three observations when all three answered. Read the count with the percentage: a result based on a small sample can move sharply after only a few new answers. Visibility and position answer different questions. Visibility asks **how many checked answers name us?** Position asks **where do we appear when named?** A high position with low visibility means the brand is prominent in fewer checked answers; a high visibility with a weaker position means it appears often but may be listed behind alternatives. Neither alone explains why, so inspect the prompt and answer examples. ### Visibility against competitors This comparison shows your brand and the competitors present in the checked answers. A competitor’s **Visibility** is the share of answers that name that brand. **Reach** weights visibility by each model's market share. **Position** is the average place where the brand is mentioned. **Citations** is a separate share of source citations. These measures use the selected answer period and model scope. The table and chart focus on the selected competitors, while **Show all** opens a fuller list when there are more rows. The chart can show gaps for days without measured answers; a gap is not a zero. In **You vs. selected**, the chart and comparison list follow your tracked selections. **You vs. leaders** shows your brand alongside the most visible brands for the current selection. The mode and tracked list are shared with the competitor comparison on Citations. Selecting a competitor adds it to the tracked set; hiding one removes it from that comparison. These display choices do not change the answers already collected or the visibility calculation. **Manage competitors** changes how competitor names are interpreted for the project. Grouping alternate spellings under one company can combine names that refer to the same brand; excluding a brand removes it from competitor analysis. These changes affect competitor comparisons and answer labels, not the wording of monitored prompts. PromptEye may suggest name matches under **Merge proposals**, but they are not applied until someone chooses **Merge**. Review proposed matches before accepting: combining two different companies makes their results harder to interpret. **Check uncertain with AI** adds an assessment next to uncertain proposals, saying whether they look like the same company or separate companies, with a reason. It is advice only; nothing is merged until you choose **Merge**. If you choose **Dismiss**, that pairing is not suggested again. A project with read-only access cannot change this configuration. When you create a group, you can also add its **Domains**. Citations of those addresses then count for that group in the Citations comparison. Without a domain, PromptEye looks up the domain by the company name, which may miss or mismatch it. To undo a group, use **Ungroup**; its spellings return as separate rows. To bring back an excluded company, open the **Excluded** tab and choose **Restore**. Changes are saved and take effect immediately across the whole Visibility module. The tracked comparison is limited to ten competitors. Hiding one removes it from the chart and table and frees a slot; use **Show competitor** to add it back later. Hiding does not erase its past answers. **Show competitor** lists only brands that have been named in the checked answers. You cannot add a company that has never appeared; it becomes available once an answer names it. The tracked list and the **You vs. selected** / **You vs. leaders** choice are saved for you in this project. Other people with edit access keep their own selection. People with read-only access see the project owner's selection; they can change it on screen, but their changes are not saved. When several spellings are grouped, the company counts once: an answer containing more than one spelling is not added multiple times to the group. That overlap is why separate spelling rows may not sum to the grouped result. Excluding a brand is broader than hiding it: it leaves that brand out of the module’s competitor tables, charts, and Share of Voice. The project owner or an editor with write access can change these settings. No brand rows in the selected period/model means no readable answer in that selection named a brand. It does not establish that competitors were absent from all AI answers. If answers exist but their brand mentions could not be interpreted, the page reports those answers as not counted; they are excluded from percentages rather than treated as brand-free answers. ### Visibility per model The model chart separates overall brand visibility from each model’s line and may also show Reach Index. It helps locate where the overall figure comes from: for example, an overall figure can hide strong visibility in one model and weak visibility in another. Use the header controls to change the models and period; use the chart legend to hide a line temporarily while reading the chart. Hiding a line is a display-only choice. Only measured days have values. If there is not enough history yet, the chart says so rather than filling missing days with zero. A project created today can show a waiting state until the first usable data is available. Compare lines only across periods where both models have observations. ### Prompt groups, changes, and sources **Visibility per group** summarizes prompts that belong to each group. It can show average visibility, Reach Index, AI Traffic, average position, and a funnel coverage diagnosis. A group row opens group details; expand it to see its prompts, and open a prompt for its focused analysis. A pending placeholder means at least one prompt is still awaiting a metric; a dash means the value is unavailable. The funnel describes which purchase-intent stages the group’s prompts cover: complete means all four stages have prompts, broken means a stage is missing, and narrowed means the group intentionally covers only the displayed range. This diagnoses prompt coverage; it does not say whether the brand appears in those answers. In a prompt’s focused panel, **Business priority** combines purchase intent with AI Traffic, while **Prompt Health** also considers how well the prompt fits the business and the quality of its intent. These help you prioritize the prompt set; they do not show whether the brand is visible. The panel’s **Visibility score** is an overall assessment based on brand visibility and the competitor gap. If the panel says the prompt is pending, wait for usable measurements before acting on the assessment. **Monitored since** marks when monitoring of that prompt began, so it explains why a newer prompt has less history than older prompts in the same project. **Biggest visibility changes** surfaces up to eight prompts with the largest measured visibility changes in the selected period. Use a row to open that prompt’s analysis and check the actual answers. A prompt needs measurements in both periods to have a change. If the project has only its first measurement, the card explains that a second check is needed. A large change is a useful investigation lead, not evidence by itself of a market-wide shift. **Cited sources** on Overview ranks the domains cited in answers across the prompt set and shows each domain’s share of source citations. Your project domain is highlighted when it appears. This is not the share of answers that named your brand, and a citation does not imply endorsement. Use the separate [Citations tab](/help/visibility/citations/) to review citation trends, competing brands, and source details. **Share of Voice** divides brand mentions across all prompt groups in the selected answer set. It answers how the observed brand mentions are distributed among brands, rather than what percentage of answers name your brand. A brand can have high visibility and lower Share of Voice when many other brands are mentioned too. The chart’s “your share” is the project brand’s portion of this mention pool. Use the full list to see the ranked brands. Export downloads the rows currently supplied to that card. For example, if your brand appears in 4 of 10 answers, Visibility is 40%. If those answers contain 25 brand mentions in total and your brand accounts for 3 of them, its Share of Voice is 12%. One answer can name several brands, so the two percentages are not expected to match. Citation quality and sentiment cards provide context for analyzed citations and answer tone when those analyses are available. They can help prioritize a closer read, but they do not replace the answer text or establish a customer’s opinion. If a card is loading or has no analysis yet, do not interpret that as a neutral or poor result. ## Turn a signal into a next step Use the dashboard as a triage path rather than as a score to optimize in isolation: 1. Check that the project, model, and period match the question you are asking. 2. Identify the prompt group or individual prompt that drives the result. Look at both visibility and position, and compare models where useful. 3. Open [Answers](/help/visibility/answers/) for that prompt. Check whether the brand is named, merely cited, absent, or not counted, and read the answer and sources. 4. Compare the cited domains on [Citations](/help/visibility/citations/). This can show whether answers rely on your domain, competitor domains, or other sources. 5. Decide whether the issue is a prompt-coverage gap, a competitor naming/grouping issue, an answer-level positioning issue, or a source opportunity. Visibility itself does not prescribe a content change or prove that publishing a page will change a result. For example, if a prompt has high estimated AI Traffic but low visibility, first confirm that the prompt is relevant to the customers you want, then inspect its checked answers and sources. The estimate is a way to prioritize a question; it does not show how many people actually asked it. For the estimate and prompt-set controls, see [Prompt Intelligence](/help/prompt-intelligence/). ## Export a view Where an export menu appears, it downloads the rows prepared for that card or table in the offered formats (CSV, Excel, or Markdown). Exports are snapshots of the displayed result set at the time you export; they do not change monitoring, prompts, competitor setup, or saved answer history. A card may export more rows than are currently visible on screen if its export data is prepared from the full result list. Check the exported column headers and current filters before using a file as a record of a specific comparison. ## Common questions ### Does a source citation count as visibility? No. Visibility requires the answer to name the brand. If the answer cites a page on your domain without naming your brand, it appears as a citation and has no brand position. This distinction is visible in Answers. ### Why do Visibility and Share of Voice disagree? Visibility counts answers that name your brand. Share of Voice counts your brand’s portion of all brand mentions in those answers. Since an answer can name several brands, these use different denominators. A 40% Visibility result and 12% Share of Voice can both be correct. ### What is included in the Reach Index? Reach Index weights brand Visibility by each model's market share, so presence in more widely used models counts more. Treat it as a weighted comparison rather than a live market-share measurement. It does not measure actual people who saw an answer. For a meaningful comparison, keep the project, prompts, models, and dates consistent. ### Why is average position blank when Visibility is above zero? Position is available only when there are readable answers that name the brand and its place among the detected brands can be determined. A dash means the selected scope has no usable position value; it is not position zero. When the value is present, 1 is the highest place and a lower number is better. ### What does “monitored since” mean on a prompt? It marks when monitoring for that prompt began. It helps explain why an older project can have less history for a recently added prompt. Results from before that date do not exist for the prompt, so compare only periods it was actually monitored. ### What does a prompt’s Visibility score or verdict mean? The prompt panel gives an overall assessment based on that prompt’s brand visibility and the gap to competitors. It is a quick way to decide which prompt deserves attention; open the full analysis and read its answers before changing content or drawing a business conclusion. A prompt can still be waiting for enough data, in which case the assessment is pending. ### What does AI Traffic mean here? It is an estimate associated with the monitored questions, not visits to your website and not a count of people who asked those exact questions in an AI assistant. The Overview card combines usable estimates for the monitored set over its displayed last-30-days window. Prompts whose estimate is in **Error** are skipped; an orange warning beside the total explains that the estimate is missing for some prompts. The same rule applies to group and Prompt Intelligence totals. The Traffic area reports observed visits to your site from AI services; use that when the question is whether AI referrals reached your website. ### Why can the AI Traffic estimate show “No data,” “Error,” or “Outdated”? These labels describe the estimate’s calculation state. **No data** means no usable result was saved; **Error** means the latest calculation failed; **Outdated** means the latest attempt failed and the previous value is still shown. Summary totals skip prompts in **No data** or **Error** when another prompt has a usable estimate, and include the saved value for **Outdated** prompts. A missing result does not erase estimates calculated for the rest of the prompt set. If no usable estimate remains, the summary shows **No data**. **Recalculate** asks PromptEye to try again for that prompt. It is available only on some plans, so you may not see this option on your account. Recalculating does not add monitoring results or guarantee that a new estimate will be available. The details panel can also list phrases that were not used and a short reason, such as being outside the prompt’s topic or an outlier. ### Why does “Biggest visibility changes” show only eight prompts or nothing yet? The card shows up to the eight prompts with the largest measured changes for the selected period. A prompt needs measurements in both comparison periods; after the first measurement there is no change to calculate, so it can appear empty until another check has completed. ### What do Complete, Broken, and Narrowed mean in Funnel? They describe the purchase-intent stages represented by prompts in a group. **Complete** covers all four stages; **Broken** has a gap in the range; **Narrowed** covers only the displayed range. This helps you decide whether the prompt set reflects the customer journey you want to monitor. It does not rate the answers or say whether your brand is visible. ### Why are some answers marked “not counted”? The answer was received, but PromptEye could not reliably read the brand mentions in it. It stays visible in Answers for context and is excluded from percentage calculations. It is neither a confirmed mention nor a confirmed absence. If many answers are excluded, interpret the percentages using the sample count shown in the comparison. ### What does “We could not calculate visibility” mean? The figures for the selected range failed to load. This is not a result of zero. Use **Try again**. If the page says **Visibility is being recalculated**, a change to the project's brand names or competitor groups is being applied to past results; the figures return when it finishes. The app does not show how long this takes. ### Why can a result change when I select another model? Each model produces a different set of checked answers. Changing the model changes the observations included in the measure; it does not change the prompt text. Compare models over the same dates and prompt set when looking for differences. ### Why is a change or position missing? A change requires comparable measurements in both periods. Position also requires answers that name the brand. If either input is unavailable for the selected model and period, the app leaves the figure blank rather than inventing a value. ### Does a higher Visibility score mean more people saw my brand? It means a larger share of the checked answers named your brand. It does not count audience views or visits. Use [Traffic](/help/ai-traffic/) for observed website visits from AI services. ### AI answers use another name for my brand. Why is it not counted? Visibility counts the brand names set for the project. If answers use an abbreviation, an old name, or a different spelling, add it under **Brand names** in **Project settings**. Visibility updates right away, including past results. In the same place, add other domains that belong to you under **Domains**, such as a blog or a country site, so their citations count as yours. Only people with edit access can change these settings. ### Can I share Visibility with a client who has no PromptEye account? Yes. In **Project settings**, open **Sharing** and turn on **Public link**. Anyone with the link can open a read-only view of the project's Overview, Citations, and Answers without signing in; they cannot change settings or take actions. Use **Preview** to see what they see. Turning the link off stops it working immediately. Turning it on again creates a new address, and the old one does not start working again. Only people with edit access can turn the link on or off. ### Why can't I change competitors or the tracked list? You have read-only access to the project. Ask the project owner or someone with edit access to change groups and exclusions. You can still change the comparison on screen, but your choice is not saved. ### What if there are no answers or the table is empty? First check the project, model, period, and any answer filters. A new project may still be waiting for its first measurement. An empty competitor panel can also mean no readable answer in that selection mentioned a brand. In Answers, the prompt, search, presence, and sentiment filters narrow the rows; clear them to broaden the list. If the page says it could not calculate visibility, use **Try again** and check again later. For answer-level filters, summaries, source links, partial scanning, and exports, see [Review individual answers](/help/visibility/answers/). For source-domain and citation-share interpretation, see [Citations and sources](/help/visibility/citations/). --- # Review individual answers Source: /help/visibility/answers/ --- title: Review individual answers description: Filter and inspect checked AI answers to understand brand mentions, position, sentiment, citations, and sources. --- The **Answers** tab shows individual model responses behind Visibility results. Use it to find out why a score moved: did the model name your brand, mention competitors, place your brand high or low, cite your domain, or rely on another source? A summary percentage is easier to act on when you can check the examples it came from. The page follows the project, model, and period controls in the Visibility header. It can also open in a prompt or group focus so the table only covers that focus. A search for a brand or phrase only filters the answer list; it does not edit the prompt or change saved results. The table shows 25 answers at a time. Changing the search, model, period, or prompt focus takes you back to the first page. ## Narrow the list - **Brand: All / Only with brand / Without brand** filters according to whether your brand is named in the answer. “Only with brand” is useful for checking positioning; “Without brand” helps you see what an answer says when the brand is missing. A source citation without a brand mention belongs to “Without brand” because citation alone does not satisfy presence. - **Sentiment: All / Positive / Neutral / Negative** filters the sentiment assigned to the answer by PromptEye’s automated answer analysis. It classifies the tone toward your brand in the answer, not a customer survey and not proof that a reader felt that way. Some rows have no sentiment available when the analysis could not provide a result. - **Search answers** matches answer text and can also find prompt text, model name, brand names, and source URL/title/domain. Clear the search and return filters to **All** if expected rows are missing. These filters work together. For example, choosing **Without brand**, **Negative**, and searching for a competitor shows only matching checked answers that meet all three conditions. If no rows appear, broaden the filters before concluding that no such answers were collected. ## Read a row - **Date and prompt** identify when the checked answer was collected and which monitored question it answered. Multiple model responses for the same prompt and date are grouped under a shared heading; the count tells you how many answer rows are in that group. - **Model** identifies the AI model that produced the response. Use it to compare answers from different providers without assuming one response represents all models. - **Your brand** indicates whether your brand is named, cited as a source without being named, or absent. **Not counted** means the answer was received but its brand mentions could not be read reliably; it is excluded from the metric denominator rather than treated as an absence. - **Mentioned brands** shows recognized brands in the answer. The table displays up to five marks, with a `+N` indicator when there are more; expand the row to read the answer itself. - **Position** is your brand’s position among the brands listed in an answer. A dash means there is no position, for example because the brand was not named, was only present as an address, or the answer could not be counted. Lower numbers are higher placement; position 1 is best. - **Citation quality** labels the role of your brand in the answer when it can be classified: recommendation, expert citation, comparison item, or incidental mention. It describes the answer’s treatment of the brand, not an independent rating of the page’s quality. - **Sentiment** shows positive, neutral, or negative answer tone when available. Treat it as a prompt for review, not a direct measure of customer sentiment. - **Sources** previews up to three source domains. Expand the row to see the full answer and source links. ## Expand an answer Select a row or its expand button to read the stored answer, a short presence summary, and its cited sources. The summary distinguishes three situations: the answer names your brand (with a position when one is available), cites your site without naming the brand, or does not include the brand. This is the clearest way to separate mentions from citations when the Overview score seems surprising. Source domains that appear in the answer text are linked to their source URLs, and the source pills open the URLs in a new tab. Some models display an **Unverified sources** note. The link records what came back with the answer; it is not a guarantee that a page is live, authoritative, or endorsed. Read the answer around the citation before treating it as support for a claim. For source-domain trends across the selected prompt set, use [Citations and sources](/help/visibility/citations/). For Google AI Overview, a prompt may not trigger an overview response. In that case the expanded row says there is no AI Overview answer for that prompt. This is different from a readable answer that simply omits your brand. ## Missing, partial, and not-counted results An empty table can mean the chosen project/model/period has no matching answer rows, the filters are too narrow, or the prompt focus has no answer in the selected scope. Reset the filters and broaden the period/model before treating an empty result as evidence of absence. With narrow filters, PromptEye searches the history in parts. If the table says **Only part of the history was scanned — nothing matched these filters yet**, use **Keep searching** to look through older answers. The message does not mean there are no matching answers in the rest of the history. Changing a filter or the period restarts the scan with the new selection. Rows marked **Not counted** are intentionally excluded from Visibility percentages because PromptEye could not reliably read which brands they mention. They remain visible for transparency, but do not treat them as either a brand mention or a confirmed non-mention. The app may also display missing AI Overview results for a prompt, which likewise should not be read as an answer that omitted the brand. If a result is absent, check the selected model and dates, prompt/group focus, and filters. If the answer list remains empty while Overview shows data, the two surfaces may be showing different scopes or filters; align those controls and review the prompt. The app does not promise that every summary card has a matching row in the current answer page if its scope differs. ## Export answers Use the export menu above the table to download the rows currently loaded on the visible answer page in the offered formats. It does not export every older answer in the project automatically. Move through pages or continue a partial scan, then export the rows you need. The file is a point-in-time snapshot; it does not change prompts, filters saved elsewhere, or answer history. Check its contents and the table’s scan state before treating it as a complete project record. --- # Citations and sources Source: /help/visibility/citations/ --- title: Citations and sources description: Understand how Visibility reports source citations, cited domains, citation share, and brand mentions. --- The **Citations** tab helps you see which brands and domains appear as sources in answers to your monitored prompts. Use it when you want to investigate whether models rely on your website, competitor websites, or other domains. A citation is evidence that the answer included a source link; it does not prove that the answer named or recommended the brand. For brand presence in answer text, see [Visibility Overview](/help/visibility/). For the underlying answer text, source links, and whether a brand was named or only cited, see [Review individual answers](/help/visibility/answers/). ## Citation share by brand The first card, **Citations**, compares the project brand and competitors. Its percentage is the share of answers that included sources in which the brand’s domain was cited. The count in parentheses is the number of those answers where the domain citation was detected. This is not a count of every individual link: if one answer cites your domain more than once, it still contributes one answer to the count. A citation is counted separately from a brand mention: an answer may cite your site but not name your company, or name your company without citing your site. The card also shows Visibility, Reach, and average position for context. These columns retain their usual meanings: Visibility is the share of answers naming a brand; Reach weights visibility by each model's market share; position is the average place in answers that name the brand, with 1 highest. The citation chart’s timeline shows the share of citations by day where daily history is available. Missing days are gaps in measured data, not zero citations. Use the model and period controls at the top of the page to change the answers behind the comparison. The compact date control beside the source table changes the source table’s reporting period independently. When comparing the chart and source table, align those periods first. If the available history is too short to draw daily citation trends, the card explains that and provides the period total in the table. ### Select which competitors to compare **You vs. selected** displays your brand and the competitors you have chosen to track. **You vs. leaders** displays your brand with the most visible brands for the current answer selection. The selection is shared with the competitor comparison on Overview, so changing modes or pinning/hiding brands in either location updates the other. Pinning a competitor uses one of the tracked slots (up to ten competitors, excluding your own brand); it affects the comparison display, not the underlying answer collection. If you are in leaders mode when pinning, the view switches back to selected so the new choice appears. The table can show up to ten rows before **Show all** opens the fuller list. Export saves the prepared competitor rows for this view; it does not alter tracked selections or historical data. ## Cited sources The source table lists domains that models used in answers for the project prompts. The selected model determines the per-model occurrence percentage. The range control offers 7 days, 14 days, 1 month, 3 months, 6 months, 1 year, or all available history. This date range is separate from the main answer-period selector; the selected model still applies. Align both controls before comparing this domain list with the chart above. Neither control changes the monitored prompts. The table’s **Citation share** uses the pool of source citations as its denominator: it shows how much of that pool belongs to a domain. The brand comparison above answers a different question: in how many answers containing sources was a brand domain cited? One answer can contain multiple cited domains, so these percentages need not match. The table counts a domain once per answer when showing occurrence, even if the same answer contains multiple links from that domain. The table can show domain metrics such as **DR** and estimated monthly visits, plus sponsored publication offer details where available. They describe the domain or publication itself; they are not the domain’s share of citations, and they do not tell you that a publication caused a particular answer. If a metric is unavailable for a domain, its absence is not a zero. Your project domain and the other domains added under **Domains** in **Project settings** are recognized as yours. If a competitor's citations seem to be missing, add its domain to its group in **Manage competitors**. Use the domain list to choose what to inspect, not as a quality ranking by itself. A competitor source appearing often may be a clue to review the page or evidence models are using. A source on your domain appearing often means the page was cited; it does not establish that the answer accurately described your product, named your brand, or sent a visitor to your site. Open answer rows in [Answers](/help/visibility/answers/) to read the context. Website visits are reported separately in [Traffic](/help/ai-traffic/). ## How to interpret differences - **More source answers cite your domain while Visibility stays flat:** your site is being used as a source more often in the checked answers, but the answer still may not name your brand. Read sample answers to see the context. - **Visibility rises while citations stay flat:** more checked answers name the brand, but the source citation measure did not rise with it. Inspect whether mentions are supported by sources, and which domains are cited. - **Your domain is missing:** check the selected model and source-table period, and confirm the project’s domain/alternative-domain configuration. The table reports observed citation domains; it does not say whether all potentially relevant URLs were accessible to a model. - **A citation value is blank or a dash:** there is no value for that selection. Do not treat it as zero unless the interface displays an explicit 0% or 0 count. ## Sources are not verified recommendations Source URLs come from model answer data. The Answers detail includes a note that sources are unverified for models where this applies. A source link shows what was returned with the answer; it does not prove that the page is live, authoritative, accurate, or endorsed by the model. Open the source and the answer context before deciding that it supports a claim or represents a content opportunity. --- # Brand analysis Source: /help/brand-analysis/ --- title: Brand analysis description: Use Brand analysis to understand the topics, competitors, and answer patterns behind your brand’s visibility in AI responses. --- Brand analysis helps you move from **“How often does AI mention us?”** to **“In which customer topics are we strong, where does another brand lead, and what should we investigate next?”** It brings together topic gaps, the role and tone of brand mentions, and a review of potentially risky wording in AI answers. Analysis follows the market selected in Project Settings. For a project set to **Global**, the analysis uses the market label **Global** and writes generated findings in English. For a country-specific project, generated findings use that country's official language. Use it alongside [Visibility](/help/visibility/). Visibility shows measured presence and citations across prompts and models. Brand analysis gives you a deeper, periodically refreshed view of the themes in those answers. It is an analytical aid: use its signals to choose what to inspect or improve, not as proof that every customer sees the same answer. ## Run an analysis when the evidence is ready Brand analysis is a separate run that saves a snapshot. It does not run by itself or update with each monitoring result; a new analysis starts only when someone with edit access selects **Run analysis**. PromptEye then uses the latest monitoring results for the project’s active prompts, from all monitored models. Archived prompts are not included, and the model and period selected on Visibility do not affect the run. The page updates when the run finishes, and the new run appears in Analysis history. The estimate below the button is a guide based on the number of active prompts, not a completion guarantee. While a run is in progress, you cannot start another one. When the project has no active prompts, has no monitoring results yet, or the latest run already includes the current results, the button is unavailable. The last case shows **Up to date**: you can run a new analysis after the prompts' next checks have produced new results (once a day on most plans, every two days on some). If a run stays **In progress** for more than about 15 minutes, you can start a new one. Each run uses part of the Brand analysis allowance your workspace gets for each billing period. The allowance is counted in prompts: a run uses as many units as the project has active prompts, and the plan includes a set number of runs per billing period for its prompt limit. A project with more active prompts therefore uses more of the allowance per run. The allowance is shared by all projects in the workspace, so runs in other projects reduce what is left. Failed and invalid-response runs do not count. If the limit prevents a run, wait for the next billing period or change the plan. Runs can show **In progress**, **Ready**, **Error**, or **Invalid response**. Ready means the saved analysis can be reviewed. Error means the run failed; Invalid response means the returned result could not be accepted. For either failure, use **Retry** when it appears. A retry starts another run and uses the plan allowance if it succeeds. If a run remains in progress, give it time to finish; the estimate is only approximate. The timeline shows the ten most recent runs. Select a point to review that run’s saved results. Runs cannot be deleted, and the page has no export. A saved run does not change later: if you change brand names or competitor groups, only the next run reflects it. If you have not run an analysis, the cards and topic table have no completed analysis to show. A new run is most useful after the project has accumulated fresh responses, or after you have made changes and want to compare a later sample with the previous one. ## Start with the summary, then choose a topic to investigate The summary cards show detected topics, distinct leaders, and context gaps in the selected ready run. **Previous** is the value from the preceding ready run, when one exists. In the current view, **Detected topics** and **Context gaps** count the same topic rows; they are not two independent scores. The counts help you notice whether the analysis found a different set of themes or leaders, but do not by themselves say whether the business improved. The Context gap table groups related monitored prompts into customer topics. For example, several questions about onboarding, integrations, and team permissions might appear under one software-evaluation topic. “Grouped from N monitored prompts” points to the linked prompt results used as evidence for that topic. PromptEye keeps only a small set of these links in each row, so N is not necessarily the total number of related prompts. A topic label is a useful summary, not a new prompt you need to monitor. **Best brand** is the leading brand identified for that topic in the analyzed answers. The evidence beneath it shows how many model observations included that brand and the average position where position data was available. For example, “3/5 models, avg pos 2” means the brand appeared in three of five usable observations and averaged position two among the observations where its position could be determined. A missing position does not mean the brand was absent; it means there was no position value to show. **Gap** is an analysis score from 0 to 100 that summarizes the distance between your brand and the topic leader. Zero represents no meaningful gap in the analysis; 100 represents complete absence of your brand while a competitor leads. It is not a literal percentage of missing content or a direct count of pages you need to create. A higher score is a signal to investigate the topic, not a forecast of lost revenue. The app groups scores as Critical (about 66 and above), Moderate (33–66), and Low (below 33). Scores near a boundary can change category with a small change in the analyzed sample. The gap is not simply the percentage of answers that omit your brand. It compares your brand with the topic leader, taking into account how often and how high each brand appears and which sources the answers use. A brand can be mentioned but still have a substantial gap if the competitor appears more often or ranks higher. Conversely, a low gap does not prove that your brand owns every part of a topic. Use **Why the leader wins** and **Why we are missing it** as hypotheses to check against the examples in Visibility. They are generated explanations based on the analyzed responses. They are not a verified audit of your website or a statement of the competitor’s actual strategy. When the evidence is thin, treat the explanation as a prompt for further review rather than a confirmed cause. For example, if a competitor leads on “compliance reporting” and the explanation points to clearer coverage of audit workflows, inspect the source answers and your own product pages. You might decide to clarify the workflow in your content, improve the prompt set, or conclude that the topic is not commercially relevant. The analysis does not make that business decision for you. ## Read the trend as a sequence of snapshots **Competitive gaps over time** groups the topics in each completed run into the same three gap bands. It helps answer whether the balance of detected gaps is moving in a useful direction across runs. **Analysis history** lets you select a prior run and inspect its saved table and cards. The sentiment trend connects available scores across runs. If a run has no sentiment result, it is left out of the line rather than shown as zero; with fewer than two usable scores, the chart can show a dot without a line. These are snapshots of separate analyses, not a continuous daily trend. The underlying answers can vary between runs even if you have not changed your site. Compare several runs and look for a repeated pattern before treating a one-run increase or decrease as a business change. If a topic appears in one run and not another, that can reflect changes in the available answers or how related prompts were grouped; it does not by itself prove that a competitor changed strategy. ## See the role and tone of brand mentions The **Citation quality** and **Brand mention sentiment** cards are shown in the [Visibility overview](/help/visibility/), alongside Visibility results. They are connected to Brand analysis because their classifications are saved with an analysis run. Citation quality describes the role your brand plays in responses where it appears: a recommendation, an expert source, a comparison element, one of several listed sources, an incidental mention, or an anti-recommendation. This is different from Visibility: Visibility asks whether the brand appears; Citation quality asks how the answer uses the mention. The distribution is calculated only from responses where your brand appeared, and the card’s footnote names the analysis run it belongs to. If it says the result was computed from only part of the mentions, some mentions could not be classified; read the percentages with that smaller classified sample in mind. The Visibility overview shows the latest ready analysis for these cards, so changing the page’s model or date filters does not turn them into a daily or date-filtered trend. If the readable answers contain no brand mentions, the card says so. If there is no completed result to display, it asks you to run an analysis. These role labels are classifications, not endorsements or a measure of whether a recommendation is commercially justified. For example, “recommendation” means the response explicitly favors the brand; it does not establish that the recommendation is accurate or that the user acted on it. Open the underlying answer in Visibility before deciding what to change. **Brand mention sentiment** is a separate measure of whether the language about your brand is positive, neutral, or negative. It applies only to responses that mention your brand. Review the examples and attributes behind the summary: a neutral factual description is not a negative review, and a positive classification is not evidence of customer satisfaction. Changes versus the previous run compare analysis snapshots, not customer sentiment over a continuous period. ## Use the compliance review as a prompt to inspect wording The **Compliance audit** highlights language in AI responses that may overpromise, make unrealistic claims, or sound unprofessional when attributed to your brand. The score runs from 1 (severe violations) to 10 (fully professional), and the flagged phrases table shows the phrase, risk type, risk level, and answer context. This is a review of language found in the analyzed AI responses. It is not a legal compliance audit, a review of every page on your website, or a conclusion that your company itself made the claim. A flagged phrase is a reason to inspect the original answer and its source. If the card reports no flagged phrases, that means none were identified in this run’s reviewed sample; it is not a guarantee that no problematic wording exists elsewhere. ## A practical workflow 1. Make sure the project has active prompts and their first monitoring results. 2. Run Brand analysis and wait for the status to become Ready. 3. Start with the largest gaps and check whether the topic matters to your customers. 4. Open the relevant answers in Visibility to see the actual wording, brands, and sources behind the signal. 5. Turn a supported finding into a concrete action, such as clarifying a product capability or creating a useful explanation for a missing customer question. 6. Run another analysis after new monitoring results are available. Compare repeated patterns across runs rather than relying on one score. If a chart or card has no data, first check whether the selected run is Ready and whether that metric had a usable sample. For example, Citation quality and brand sentiment need responses that mention your brand; topic gaps need readable monitoring results. An empty card can therefore mean “no qualifying evidence in this run,” not necessarily “the feature failed.” Use the status and the card’s own note to distinguish these cases. --- # AI Ads Source: /help/ai-ads/ --- title: AI Ads description: Understand which advertisers and ad creatives appear beside your tracked AI prompts, how the Ads metrics work, and what to investigate next. --- The **Ads** page helps you see where paid placements are beginning to appear around the questions you monitor. Use it to understand which advertisers are entering your category, which prompts already carry ads, and how their messages compare with your own presence. This is a monitoring view. It does not place an ad or start a campaign. It records ads found alongside answers for your tracked prompts. At present, ChatGPT is the only active model in this view; other listed models are planned and cannot be selected yet. The separate OpenAI Ads account and campaign tools in PromptEye are a different feature. ## Start with the scope of the data The page's numbers depend on the selected date range and the prompts in the current project. An ad appearing next to one of your prompts means paid competition was observed for that question in the selected period. It does not tell you how often every person saw the ad, whether they clicked it, or how much the advertiser spent. Ads are recorded from the point when ad tracking was switched on. PromptEye cannot reconstruct measurements from before that point, and prompts you add later have ad data only from their first checks. A short history can show an early signal, but trends need a few weeks of data. Ads are captured from ChatGPT answers to your monitored prompts. The model you choose on Visibility does not affect this page. If Visibility shows data but this page has never shown a single ad, contact PromptEye support to check whether ad tracking is switched on for your account. The model picker labels models as **Active** or **Planned**. Active means the page currently collects data for that model. Planned means it is not available for ad monitoring yet; the screen says planned models will join without extra cost once their providers launch ads. Today, ChatGPT is active. Gemini, Google AI Overview, Perplexity, Copilot, and Claude are listed as planned. ## Choose a period that answers your question The date control at the top changes the time window used throughout the overview and its drill-downs. It opens on the last month. Choose a recent period to see what is happening now, or a longer period to understand whether advertisers keep returning to the same prompts. You can also set a custom range, up to 60 days. This page keeps its own period: changing it does not change the period on Visibility, and the other way round. By default, changes are measured against the previous period of the same length. For example, a 14-day range is compared with the 14 days immediately before it. In the period picker you can choose another baseline, such as the same period a year earlier or a custom range; see [Set the model and period](/help/visibility/#set-the-model-and-period) for the options. A change is a difference in the measured rate or count; it does not by itself explain why the market changed. Changes are hidden when either period has fewer than 7 measured days, which is common shortly after ad tracking starts. If you see a notice that the period holds more observations than can be read at once, the figures are based on only part of the data. Narrow the date range before using the numbers to make a precise comparison. The notice applies to the displayed figures, including changes that depend on the comparison period. ## Read the summary cards ### Monetised prompts A prompt is **monetised** when at least one ad appeared beside an answer to that prompt during the selected period. The card shows the number of those prompts and their share of all tracked prompts. It compares the count with the previous period when a baseline is available. This is a prompt-level measure: a prompt with one observed ad and a prompt with ads seen repeatedly both count as monetised here. Use **Ad frequency** in the group table or the prompt details to look at how often ads appeared in measurements. ### Advertisers An advertiser is usually identified by the destination domain of its ad. When that domain is unavailable, PromptEye uses the advertiser name to group observations. Multiple campaigns with the same destination domain count as one advertiser. If your own brand or domain appears among the observed advertisers, the card says **you are among them**; the matching advertiser is also marked **(you)** in the relevant detail. PromptEye recognizes your ads by the project's domain and brand name, including the other domains and brand names added in **Project settings**. If your ads lead to a domain that is not listed there, add it so they are marked as yours. The change beside this card represents advertisers that are new relative to the previous period. A company may be new to the selected comparison window without being new to the wider market: the page only knows what has been observed in your monitoring history. ### Active creatives and retired creatives A **creative** is one identifiable version of an ad, with its own headline, copy, destination, and possibly an image. Several versions from one advertiser count as separate creatives. The card counts creatives seen in the selected period as active. It also shows how many were classified as retired during that period. Its change is **new creatives minus retired creatives**. A creative is marked retired when it has not appeared for at least seven days before the latest observation in the selected period. Because this depends on the selected period, the same creative can be active in one period and retired in another. This label is a signal of recent inactivity in your measurements; it does not establish that the advertiser permanently stopped running it elsewhere. ### New advertisers This card highlights companies whose ads were not present in the preceding period of the same length. Use it as a prompt to investigate a possible new entrant, not as proof that the company has just started advertising anywhere. If there is no usable comparison period, the page cannot reliably label an advertiser new. ## Use the advertiser chart to compare paid presence The advertiser table and daily chart use **Ad. Visibility**. For an advertiser, this is the share of collected answers in which at least one of that advertiser's ads appeared. The chart shows the measure by day; the table summarizes it across the selected period. A company can have a higher Ad. Visibility than another without having a larger share of all ad impressions, because one answer may contain more than one ad. **Prompt coverage** answers a different question: on how many of your tracked prompts did the advertiser appear? The percentage is relative to the full tracked prompt set, with the prompt count shown beside it. An advertiser can have high Ad. Visibility on a small number of prompts but low prompt coverage overall. Select an advertiser to inspect its detail. **Advertising since** is the earliest observation available for that advertiser in the selected data; **new in period** means it was absent from the previous comparison period. Neither label proves when the company began advertising outside your monitoring history. The trend chart compares its daily Ad. Visibility with your brand when your ads were observed, the average for all advertisers, and the average for the ten highest-ranked advertisers. These averages provide context; they are not a target or a prediction of what your brand should reach. The detail also breaks presence down by prompt group and by individual prompt. **Group coverage** shows how many prompts in a group carry that advertiser. **Average Ad. Visibility** is averaged across the group's prompts. At prompt level, the advertiser's rate is the share of that prompt's measurements in which its ad appeared. Your own rate is shown separately, or the page says **you don't advertise** for that prompt when no matching ad was observed. ## Compare each advertiser's share of ads **Share of ads** redistributes the advertisers' Ad. Visibility values so their shares add up to 100%. It helps answer, “Among the advertisers we observed, how is paid presence divided?” It does not mean that each company appeared in exactly that percentage of answers. Click a company in the chart legend to hide it. The remaining shares are recalculated across the companies still shown. This is useful for comparing a smaller set, but the resulting percentages describe only that visible subset. Restore the company in the legend before treating the chart as the full market view. ## Find prompts where advertising has appeared The **Monetised prompts** table lists prompts with at least one ad in the selected period. It is sorted by the number of advertisers competing on each prompt and shows whether your own ad was observed there. Select a prompt through the detail views to inspect the advertisers and creatives connected with it. If a row says **Prompt no longer tracked**, the ad was observed for a prompt that has since been deleted from the project. The historical observation can still help explain why an advertiser appears in the selected period, but the prompt is no longer an active monitoring target. The **Free prompts** view lists prompts for which no ad was observed during the selected period. This can help you identify questions where paid placements have not yet appeared in your measurements. “Free” means no ad was observed for that prompt in this data window; it does not guarantee that no one is advertising on the topic elsewhere or that an ad placement is available to buy. ## Read prompt groups as category coverage Prompt groups are the same groups you see on Visibility. Ads does not provide a separate way to edit them; create groups and move prompts in [Prompt Intelligence](/help/prompt-intelligence/). Group changes also change how Ads organizes its group-level coverage. They do not change the prompt wording or past ad observations. Prompts without a group do not appear in **Ads per prompt group** or in the **Free prompts** list, although they still count in the summary cards. Assign every prompt to a group if you want the group views to cover your whole prompt set. In **Ads per prompt group**, **Monetised** shows how many prompts in the group had at least one observed ad. For example, “3 of 5” means three of the group's five prompts carried an ad in the selected period. **Ad frequency** measures how often an ad appeared in the measurements of the group's monetised prompts. A value around 50% means an ad appeared in roughly half of those measurements. It is not the percentage of people who saw an ad. If no prompt in a group was monetised, frequency is not meaningful and the table shows a dash. Select a group to see which advertisers and prompts make up that result. **Group coverage** on an advertiser's detail shows both the number and share of prompts in each group where that advertiser appeared. This can reveal whether paid activity is concentrated around one type of customer question or spread across the category. ## Review the creative library The creative table and **Competitor creative library** let you inspect the ad versions behind the summary metrics. The table can be sorted, searched, filtered by advertiser, and paged. It includes the number of your monitored prompts where each creative appeared, its highest Ad. Visibility on any one prompt, and its share of observed ad impressions. **Max Ad. Visibility** is a creative's strongest prompt-level rate, not its average across all prompts. **Prompts** is the count of monitored prompts where that creative appeared. **Share of ads** shows its portion of all observed ad impressions for the selected period. Open the preview to inspect the creative as it was captured in ChatGPT. The preview includes the advertiser, headline, body copy when available, destination, and image when available. Missing copy or an image means that element was not available in the captured creative; it does not mean the original ad never contained it. Use **All**, **New**, and **Retired** to narrow the library, then search by advertiser or ad copy. A new creative is one first observed within the selected period. The retired status uses the same seven-day inactivity rule described above; it does not confirm that the advertiser ended the campaign. Some destination links include campaign parameters such as campaign, medium, or segment. These are labels attached to the destination URL. If none are present, the page says **no campaign parameters** and identifies that version by its stable creative identifier instead. The absence of campaign parameters does not prevent you from reviewing the creative. ## Use Insights to decide what to inspect next Insights summarize a few useful signals from the selected period: - **Opportunity** points to prompts with no observed ads and can open the free-prompts list. Treat it as a possible opening to investigate, not a guarantee that buying an ad is possible or inexpensive. - **Watch out** flags a competitor ad beside a prompt that contains your brand name or one of the brand names added in **Project settings**. Open the creative to see which advertiser's destination is shown next to that brand-related question. - **New player** highlights an advertiser first observed in the selected comparison period. Open its details to see its prompt and group coverage. An insight is a shortcut into the evidence, not a complete explanation of the market. Check the prompt, period, and creative before deciding whether it calls for an advertising or content response. ## Common questions ### Will I pay more when other AI assistants start showing ads? No. The model picker says planned models join at no extra cost once their providers launch ads. ### Is Ad. Visibility the same as brand Visibility? No. Brand Visibility on the [Visibility](/help/visibility/) page is the share of answers that name your brand in the text. Ad. Visibility is the share of answers that carried an advertiser's paid placement. A brand can have high Visibility and no ads, or run ads on prompts where the answer does not name it. ### Can I export the Ads figures? The Ads page does not offer an export at the moment. To share a view, send the link to an advertiser, creative, or group detail. The person opening it sees the period selected on their own screen, so ask them to set the same period. ### Why does a link show "Not found in this period"? The advertiser, creative, or group had no data in the period currently selected. Widen the period or go back to the overview. ## When no ads or figures appear **No ads in this period** means none of the tracked prompts returned an ad in the selected model and period. It is evidence about this monitored set only. Check that the date range includes days after tracking began and that the prompts still represent the customer questions you care about. Do not interpret an empty result as proof that the entire category has no advertising. If the page says **The figures could not be loaded**, use **Try again**. A loading error is different from a successful result with no ads: until the retry succeeds, the page has no figures to interpret. If a shared link opens to **Not found in this period**, widen the date range or return to the overview. The selected period may not contain the item, or the link may refer to an item that is no longer available in that view. ## A practical review workflow 1. Choose a period long enough to include several measurements and confirm the model is active. 2. Check the Monetised prompts and Advertisers cards to see whether paid placements are emerging. 3. Compare Ad. Visibility with prompt coverage. This separates strong presence on a few prompts from broader category coverage. 4. Open the groups with the most activity, then inspect the prompts and creatives behind them. 5. Check Free prompts and the Insights cards for questions worth monitoring more closely. 6. If a partial-data notice appears, narrow the date range and review the comparison again before sharing or acting on the figures. This workflow can help decide where to investigate further. Ads measures observed placements around your tracked prompts; it does not measure clicks, conversions, spend, or the full advertising activity of a market. --- # Use PromptEye with Claude and other tools Source: /help/developer/ --- title: Use PromptEye with Claude and other tools description: Connect PromptEye to Claude or your own tools, and explore what you can do with your data. --- The **Developer** page helps you use your PromptEye data outside the app. For example, you can ask Claude about your projects in a conversation, or connect PromptEye to another tool your team uses. We are still developing these connections and adding to what they can do. The available options may grow over time. ## Ask Claude about your PromptEye data PromptEye MCP connects PromptEye to Claude Desktop. Once connected, you can ask questions in everyday language instead of looking through the app yourself. To get started, open the **Developer** page in PromptEye and choose **Download for Claude**. Install the downloaded add-on in Claude Desktop. Then create an API key with **API access**, and enter the key and the API address from PromptEye in the add-on settings. The full key appears only once, so copy it when it is created. If you lose it, create a new one and replace the old key in Claude. Claude Desktop is not the only option: on the **Integrations** page, **Open repository** shows how to connect PromptEye MCP to other assistants that support MCP. When you are ready, start by asking Claude: > What can PromptEye MCP help me with? Which tools can I use? Claude can tell you which PromptEye tools are available in your setup. Then you can ask it to use one. For example: - “Show me the projects I have in PromptEye.” - “What is happening with my brand’s visibility in [project name]?” - “Which competitors are showing up in [project name]?” - “What prompts are we tracking for [project name]?” - “What sources are appearing in the AI results for [project name]?” Claude can also look up this Help Center, so you can ask how a feature works, for example “What does Reach Index mean in PromptEye?”. What Claude can answer depends on the tools available in your MCP version and the data in your PromptEye account. If you are not sure where to start, ask Claude what it can look up. ## Connect another tool with the API The API is a way for other apps to share information with PromptEye. It can help your team bring PromptEye data into its own tools, or connect a workflow you already use. The **API** card shows the address needed for a connection and links to the API documentation, which explains what is currently available. You do not need to use the API to work in PromptEye. If you are setting up a connection yourself, the API documentation is the place to check whether it supports what you have in mind. We are continuing to develop the API and its integrations. ## Your connection keys The **API Keys** card is where you create and remove keys used by Claude and other connected tools. Give each key a name that helps you remember where it is used, such as “Claude Desktop”. Each key has one or both permissions. For Claude, Looker Studio, and your own tools, choose **API access**, which allows reading and changing your PromptEye data through the API. **LLM tracking ingest** only allows sending bot visits from your website to [Traffic](/help/ai-traffic/setup/); it does not let anyone look up your data. Use a separate key for each purpose, so that a key placed on your web server cannot read your data. A key belongs to the person who created it and works with the projects that person can access. If their access to a project is removed, the key no longer reaches that project. You will only see the full key when you create it. Save it somewhere safe and share it only with the tool you are connecting. If you delete a key, anything using it will stop connecting until you add a replacement key to that tool. --- # Connect PromptEye to your other tools Source: /help/integrations/ --- title: Connect PromptEye to your other tools description: Bring Google and Bing data into PromptEye, or use PromptEye reports in Looker Studio. --- Integrations connect PromptEye with tools you already use. Google connections bring Search Console and Analytics data into projects. Bing Webmaster Tools can add search queries as context when you create topics and prompts. Looker Studio lets you build reports from PromptEye data. ## Google Search Console and Analytics Connect your Google account once, then link a Search Console site or Analytics 4 property to a PromptEye project. PromptEye uses read-only access: it can bring data into PromptEye, but it does not change your Google account or need your Google password. ### Connect a project 1. Open **Integrations** and connect your Google account. 2. Choose the PromptEye project and the matching Google property. 3. Select **Assign property**. You need access to the Google property and permission to edit the PromptEye project. If a project is read-only for you, ask its owner or editor to connect it. Search Console brings search performance, such as queries and pages people find through Google; it appears in [Traffic → SEO Crawlers](/help/ai-traffic/seo-crawlers/) and informs Content recommendations. Analytics 4 helps measure identified visits referred by AI services on [Traffic → Overview](/help/ai-traffic/overview/). The project connections list shows which properties are linked, when they last synced, and when the next sync is due. The first sync starts automatically, and after that data is synced once a day. The Google account connection belongs to the person who connected it. The data it brings into a project is visible to everyone who can open that project. If that person disconnects their Google account, the projects they linked stop receiving new data until someone links them again. ### If a property is missing or syncing stops Make sure the Google account you connected can access the property, and that the property's site or domain matches the project. If authorization expires, reconnect Google. A sync error means new data may not be arriving; check the message shown next to the connection. ### Disconnecting Google You can disconnect one project or the whole Google account. Disconnecting a project stops new Search Console and Analytics syncs for that project. Disconnecting the account removes Google's authorization and disconnects both services from every linked project. Data already synchronized remains in PromptEye, but no new Google data will arrive until you reconnect and assign the properties again. ## Bing Webmaster Tools Connect Bing Webmaster Tools to the workspace you are currently working in. The connection is shared by the whole workspace, and the card shows who connected it. PromptEye can then use the top queries Bing reports for project domains listed in that Bing account as context when generating topics and prompts. If PromptEye says no verified domain matches, check that the domain in the project is verified in the connected Bing account. If the connection has stopped working, reconnect it to restore access. Disconnecting Bing stops PromptEye from using those Bing queries as context; it does not disconnect Google or remove your PromptEye projects. ## PromptEye MCP The **PromptEye MCP** card connects PromptEye to an AI assistant. For Claude Desktop, choose **Download for Claude** and install the package. For other assistants that support MCP (Model Context Protocol, an open standard for connecting assistants to tools), use **Open repository** and follow its instructions. Both need an API key with **API access**. See [Use PromptEye with Claude and other tools](/help/developer/). ## Looker Studio Use the Looker Studio card to open the PromptEye connector and build reports from PromptEye data, including projects, visibility, competitors, prompts, sources, and mentions. During setup, you need an API key with **API access**. Create and manage that key on the [Developer page](/help/developer/). If you are unsure whether the connector includes a metric you need, check the connector setup and API documentation linked from the Developer page. Available data can depend on the connector and the projects your account can access. --- # Manage your PromptEye account and workspace Source: /help/settings/ --- title: Manage your PromptEye account and workspace description: Find your profile, company details, billing, usage limits, report branding, theme, and workspace access settings. --- Use **Settings** to manage your account, choose what appears in public reports, review plan usage, and control access to a workspace. The options are grouped into cards and tabs so you can change one part without searching through the rest of the app. ## Profile Your display name changes the name shown in your account menu. Leave it blank to show the part of your email address before `@` instead. Changing your account email changes the address you use for PromptEye. Enter your current password and confirm the new address using the code sent to it. Until you enter a valid code, your current email stays in place. If the code expires or you run out of attempts, start again to request a new one. The **Notifications email** field controls where public report notifications are sent: when a report is sent to a recipient and when a recipient asks to be contacted. Add more than one address, separated by commas, if teammates should receive them too. Leave it empty to send notifications only to your account email. It is the same setting as on the **Integration** tab of [Public Reports](/help/public-reports/info/notifications/). The profile also shows your plan and prompt usage for a quick check. For more workspace usage details, open **Limits**. ## Company Company name, address, country, and tax number are saved as the billing identity for your personal workspace. PromptEye can use these details when you set up its billing. A team workspace is billed separately, so changing your personal details will not change the company's details on a team invoice; enter those when setting up that workspace's billing. The company logo appears on public report pages shared with clients. Replacing it changes the logo on those report pages, including existing links; removing it makes them appear without your logo. ## Billing Choose which workspace you want to pay for or manage. Changing the selection changes which workspace's plan, usage, and billing portal you open. Changing the plan or subscription affects that selected workspace only. **Manage billing** opens the payment provider's portal for the selected workspace. Use it to manage that workspace's subscription or payment details. A team workspace has its own subscription, separate from your personal workspace. ## Limits See how much of the current workspace's plan allowance has been used, such as projects, tracked prompts, or other plan features. Where a limit is numeric, the card shows usage against that limit. Some plans use older quota rules and do not show feature limits here; this does not mean the workspace has no limits. The usage display helps explain why an action may be blocked when a workspace reaches a plan limit. Limits shown as **Not included**, **Custom**, **Unlimited**, or **All** follow the plan's terms. If no feature limits are shown, that plan uses a different quota model; it does not mean everything is unlimited. ## Branding Set the fonts and colors used in public report pages shared with clients. Saving changes how reports are presented, including reports people already open from a link. These choices do not change how PromptEye looks for you; use **Theme** for that. The logo is managed in **Company** and also appears on public report pages. ## Theme Choose two accent colors for your PromptEye account. Saving changes the sidebar and matching highlights you see in the app. It does not change the reports your clients see; use **Branding** and **Company** for those. ## Access Manage who can reach the selected workspace and projects. Invite people by email and choose a role: - **Owner** and **Admin** manage the workspace, its members, and its projects. - **Member** works in every project of the workspace. - **Client (read-only)** sees every project in the workspace and can change nothing. Use **Invite to selected projects** when someone only needs some projects. They reach only the projects you pick; a project added later is not added to their list automatically. Changing or removing access changes what they can see in PromptEye right away. An invitation must be accepted with the email address it was sent to, and it expires after a while; pending invitations show their expiry date, and you can revoke one or send a new one. A workspace must always have at least one owner, so the last owner cannot be removed or leave until another owner is added. ## Subscription and payment problems If a card payment fails, everything keeps working until the date shown in the banner; update the payment method before then. If an add-on payment fails, the add-on keeps working while the payment is retried. When a workspace's subscription ends, the workspace becomes read-only: nothing can be created or edited, but the data is kept and everything returns as soon as payment resumes. Only an owner can resume payment. On the Free plan, PromptEye does not collect or analyze data for your prompts; upgrade to start collecting. ## Project settings Each project has its own **Project settings**, opened from the project menu in the sidebar: - **Brand measurement information** shows the project name, market, and creation date. These are set when the project is created and cannot be changed. To monitor another market, create a separate project. - **Descriptive data** lets you add labels, such as Client or Priority, to organize the project list. Labels do not affect measurement. - **Workspace** shows where the project lives and lets you move it to another workspace. Its prompts move with it and count against the new workspace's plan, and the new plan decides which models are monitored. Past costs stay with the workspace that paid for them. - **Sharing** turns on a public, read-only link to the project's Visibility results for people without an account. Turning it off stops the link at once; turning it on again creates a new address. - **Alternative identifiers** add other brand names and domains that belong to the project. Visibility and citation results update immediately, including past results. - **Favicon** sets the icon that identifies the project in the sidebar. You can find it automatically, paste a link, or upload a square image up to 200 KB. You need edit access to the project to change these settings. You can delete a workspace only when you are its only member, it is not your last workspace, and it has no active subscription. Deleting it permanently deletes its projects and their prompts, results, and related data. This cannot be undone. Move or preserve anything you still need before confirming. --- # Traffic Source: /help/ai-traffic/ --- title: Traffic description: Understand AI visits, search crawlers, website request logs, and the integrations behind PromptEye Traffic. --- Traffic brings together five views of what automated services do on your website. Open it from the product menu, then choose a tab for the question you are trying to answer. | Tab | Use it when you want to… | | --- | --- | | [Overview](./overview/) | See the latest AI visit, traffic sources, changes over time, and whether AI visits lead to sessions in Google Analytics. | | [AI Traffic](./ai-traffic/) | Find which AI services fetch your pages, what they reach, and whether requests run into errors. | | [SEO Crawlers](./seo-crawlers/) | Review search engine and SEO tool activity separately from AI services. | | [Logs](./logs/) | Trace individual identified bot requests by page, user agent, status, and time. | | [Settings and setup](./setup/) | Install tracking or connect Google Analytics, Search Console, and a sitemap. | The project selector determines which website the reports describe. The date and service filters are shared by the report tabs; changing them changes the numbers and lists you are looking at. Traffic keeps its own period, separate from Visibility and other screens. A custom date range can cover up to 60 days. The chart zoom is a closer view of the period already loaded, so it will not retrieve older events. ## How the data fits together PromptEye records visits from bots it can identify after the tracking integration is installed on your site. That tells you about automated requests, not people viewing your site and not how often your brand appears in AI answers. Google Analytics can add a separate view of sessions that began from an identifiable AI referral. Search Console adds Google search clicks and impressions. These sources answer different questions, so their totals are not expected to match. The **AI Traffic** tab here counts recorded visits of AI services to your website. It is not the same as the **AI Traffic** estimate on Visibility and Prompt Intelligence, which estimates the demand behind your monitored questions. Traffic also cannot tell you which prompt or AI answer led to a visit; it shows which service fetched which page and when. If all the tabs are empty, first check **Settings and setup** for the selected project. A connected account alone does not install the website tracker: the tracker must be deployed, and an identified bot must visit the site before AI request data appears. ## A practical way to investigate For example, after publishing a product guide: 1. Check **AI Traffic** to see whether assistant or crawler requests reached the new path. 2. Open **Logs** if you need to inspect the exact request and HTTP response. 3. Check **SEO Crawlers** and **Overview** to compare search crawler activity and Google search performance. 4. If Google Analytics is connected, review whether identified AI referrals led to sessions or key events. That is a clue about visits from AI tools, not proof of which answer mentioned the page. For the details and limits of each report, use the tab guides in the sidebar. --- # AI Traffic Source: /help/ai-traffic/ai-traffic/ --- title: AI Traffic description: See which AI services visit your website, which pages they reach, and whether they encounter errors. --- AI Traffic shows visits from AI assistants, agents, and AI search crawlers to the selected project’s website. Use it to see whether AI services are finding your pages, which services visit most often, and whether your site is making those pages easy to reach. These are visits to your website; they do not show how often your brand appears in AI answers. For answer visibility and citations, use Prompt Intelligence. The data appears after AI traffic tracking is installed on the project’s website. If this is your first visit, start on **Traffic → Overview** and follow the setup checklist. Until the tracking code is installed and an identified bot visits the site, the AI Traffic page may be empty. The page refreshes as new visits are received. ## Choose a time period and AI service The filters at the top affect the figures and tables on this page. Choose a period to investigate a launch, compare recent activity, or focus on a particular date range. A custom range can cover up to 60 days. You can also select one AI service to isolate its visits; return to **All active models** to see the whole picture. Comparisons and percentage changes use the comparison period selected in the date filter. If there is not enough earlier data, PromptEye shows no comparison. The chart’s zoom controls only narrow the view inside the period already selected above; they do not load older data. ## Are AI services visiting more often? The summary cards show identified AI visits, distinct pages reached, ChatGPT fetches that may indicate citations, and the average visits per day. The average is based on the selected range, and the peak tells you which day had the most visits. Use these as signals of crawler and assistant activity, not as a measure of people visiting your site or of your position in AI answers. The **AI Traffic Breakdown** chart helps explain when visits happened and what kind of service made them. Switch between **By Type** and **By Company** to see activity grouped by service role or provider. The type view separates assistants, agents, and search crawlers. A higher line means more recorded visits in that group during the selected period. Use the date filter to change the reporting period, then use the chart zoom to inspect a shorter part of it. ## Which services and pages should I look at? **Recent AI visits** gives you a quick view of the latest identified requests, including the service, visit type, time, and page path. Select **View All** to open Logs when you need to inspect individual requests, response codes, or user agents. **Traffic by AI platform** ranks the identified bots in the selected period. Visits show request volume; unique pages show how many different page paths that bot reached. Traffic share is that bot’s portion of identified AI visits in the current view. Use this list to spot a service that has started visiting often or one that has not reached important pages. **Top pages reached by AI** ranks website paths by the number of recorded visits. The share is each page’s portion of identified AI visits, and the change indicator compares with the selected baseline when one is available. A frequently fetched page is not proof that it appeared in an answer, but it can help you decide which content to keep current or investigate further. ## Did bots reach the pages successfully? **AI crawl health** summarizes successful responses, redirects, client errors, and server errors from identified AI requests. The issue list helps you find paths that bots could not reach directly. Filter it to focus on redirects, 4xx errors, or 5xx errors, then export the filtered results as a CSV to share with whoever maintains your website. An individual error does not automatically mean that an important page is broken. Check the path, response code, redirect destination, number of hits, and last-seen time together. For example, a repeated 404 on a current product page deserves attention; a redirect from an old URL to its replacement may be expected. The assessment cards describe the broader pattern for the selected period, so use them as a pointer for investigation rather than a substitute for checking the affected pages. Some of those errors are not about your content at all. Automated scanners request paths such as environment files, administrator areas, or repository metadata, borrowing a well-known bot name as they go, and your site correctly answers 404. When this happens, a line under the summary counts how many of the requests in the period were this kind of probing, and a **Vulnerability scanning** filter narrows the issue table to the ones that ended in an error or a redirect. Expect the count to be larger than the number of rows: it includes probing requests that were answered normally, while the table only holds the issues listed above. The error counts and the assessment cards are unchanged; the line is there so a red error rate on a healthy site has an explanation attached. ## What do the ChatGPT citation signals mean? The citation section lists pages fetched by **ChatGPT-User** while ChatGPT was responding to user requests. A fetch is a strong signal that the page may have been considered as a source, but PromptEye cannot confirm that the page link appeared in the final answer. That is why the summary calls these estimated citations or potential mentions. The list goes by the name the request gave for itself, so anything imitating ChatGPT-User appears here too. If a path in this list looks nothing like a page an assistant would read, check it in Logs before treating it as a citation signal. Search the list by page path to find a particular page. The fetch count shows how often it was requested in the selected period, and the last-fetched time helps you see how recently that happened. Use **Export CSV** to download all rows matching the current search, including rows on later table pages. If the report says it is partial, the range exceeded the amount of traffic PromptEye can read for this report; the list contains the most recent requests, so older fetches may be missing. ## Common questions ### Why is the page empty? Confirm that AI traffic tracking is installed for this project and that you selected the project whose website has the tracking code. A configured integration can still show no visits until an identified AI bot reaches the site. Check the **Overview** checklist for setup status and **Logs** for recorded requests. ### Why do visits not match website analytics? This page counts identified automated requests, not human sessions. The tracker recognizes a maintained list of bot signatures, so unknown bots are not counted, and a bot is recorded under the name it gave for itself. PromptEye checks that name against the address the request came from and stores the result with each request; the figures on this page still include every identified request, whatever that check returned. Some recognized search-engine and SEO bots appear separately under **SEO Crawlers**. Analytics reports human sessions, and the two sources measure different things. Read [what the tracker sends and how it works](../setup/) before installing it. ### Does a ChatGPT fetch prove that ChatGPT cited my page? No. It means the ChatGPT browsing service fetched the page while preparing a response. It is a useful clue, but it does not confirm that the final response included your page as a link. ### Why is a comparison missing? PromptEye needs recorded data for both the selected period and its comparison period. If the project has not collected enough history yet, the current values still appear but there is no earlier baseline to compare against. ## Questions to investigate - Which pages did AI assistants fetch most often this month? - Did visits from a particular AI service increase after we published new content? - Are AI bots repeatedly reaching pages that return errors or redirects? - Which pages did ChatGPT-User fetch recently, and should we review them for accuracy? - Does a recent increase come from assistants, agents, or search crawlers? Use **Logs** when you need to trace a specific request. Use **SEO Crawlers** for search engine and SEO tool activity; that traffic is kept separate from AI traffic. For setup, data fields, privacy questions, and how to connect Google services, see [Traffic Settings and Setup](../setup/). --- # Traffic Logs Source: /help/ai-traffic/logs/ --- title: Traffic Logs description: Find an individual identified bot request and inspect its page, user agent, response, and time. --- Open **Logs** when a summary points to a problem and you need to inspect the individual requests behind it. The table shows identified bot requests, newest first, with provider, bot type, user agent, page path, HTTP status, and timestamp. Search by page path or user agent to narrow the rows. This is useful when checking whether a particular page is repeatedly requested, or when you want to find one bot by its recorded user-agent text. The search only narrows the loaded log results; it does not change the collection of events or the other report tabs. Use the status with the path and time to decide what to do next. A 2xx response means the request succeeded, a 3xx means it was redirected, a 4xx means the page was unavailable or access was refused, and a 5xx means the server failed to handle the request. A missing status can occur on older recorded requests that did not include one. **Export** downloads the rows currently shown for the selected project and period. Logs shows at most the 200 newest matching requests, so an export is not a complete archive when the site received more than 200 requests in that range. The live indicator means new signals are received as bots visit; it does not mean every automated request on the internet can be identified. If a request is absent, check that the tracker is installed on the right project and that the visitor matches a bot PromptEye recognizes. A row shows the name the visitor gave for itself, and that name is a claim anyone can write. PromptEye separately compares the address each request came from with the addresses its supposed operator publishes, and records whether the claim held up. That result is not shown in this table yet; it is available through the [PromptEye API](/help/developer/). So a row naming a well-known assistant tells you what the visitor said, not whether it was telling the truth. See [Settings and setup](../setup/) for what the three possible answers mean. For a response error, note the path and status and share them with whoever maintains the website. A 404 on an old URL may be expected; a 404 on a current product page may need a redirect or a fix. A 403 may come from access rules or a firewall. Logs can show the symptom but do not identify which server rule caused it. See [AI crawl health](../seo-crawlers/) to group repeated errors and redirects. Not every error here is about your content. Requests for paths such as `/.env.backup` or `/.git/config` are automated probing for a way into the site, and they return 404 because those files are not there — which is the correct answer. Crawl health counts how many of these you received, so you can tell a page that genuinely broke from a stranger trying door handles. --- # Traffic Overview Source: /help/ai-traffic/overview/ --- title: Traffic Overview description: Read the latest AI visit, platform share, traffic changes, and AI referrals in Google Analytics. --- Use **Overview** for a quick read on recent AI activity and what changed during the selected period. It is a useful starting point when you want to know whether AI services have begun visiting, which provider is most active, or where to investigate next. ## Setup checklist If the project still needs setup, a checklist appears above the report. It can guide you through website tracking, a sitemap, Google Search Console, and Google Analytics. You can dismiss unfinished checklist items; dismissal hides the reminder for this project in this browser and does not connect or disconnect anything. Read [Settings and setup](../setup/) for what each connection contributes. ## Last AI visit This panel shows the newest identified AI request, including its provider, page, time, and response. **Open** takes you to that page on your website; use it to check what the bot reached. A visit from an assistant may be labeled a likely citation signal. It means the assistant fetched that page while preparing a response, but PromptEye cannot confirm that the final answer linked to it. If no visit has been recorded, the panel stays empty. That does not mean setup failed: a bot must actually request a page before there is a visit to show. ## Platform share and Insights **AI Traffic Share by Platform** shows how the identified visits are divided among providers. A high share tells you which provider generated most of the recorded bot activity; it does not tell you which provider sent the most human visitors. **Insights** turns the selected period into a few prompts for investigation, such as a change in total visits, the leading platform, or the number of pages fetched by assistants. Use the period comparison selector to choose the baseline. If PromptEye has not collected enough history for both periods, a comparison is unavailable and the current period still appears on its own. ## AI visits over time The trend chart compares daily identified AI visits with the baseline you selected. You can turn the comparison off or use the same weekdays or a year earlier when available. This helps separate a real change from a difference in period length or weekday mix. A peak day marks the busiest day in the selected range; it is a useful place to open **Logs** and inspect what happened. ## AI referral sessions When Google Analytics is connected, Overview also shows sessions that Google Analytics attributes to identified AI referral sources, engaged sessions, engagement rate, average session duration, key events, sources, and landing pages. This helps answer whether AI-referred people arrived and interacted with the site after a bot visit. These are analytics sessions, not bot requests. Direct traffic cannot be attributed reliably, so the report includes only identified referrals. If Google Analytics is not connected, the page offers its connection flow here. Connecting it lets PromptEye read the selected property's analytics for this project; it does not install AI request tracking. --- # SEO Crawlers Source: /help/ai-traffic/seo-crawlers/ --- title: SEO Crawlers description: Separate search engine and SEO auditing bot activity from AI assistant and crawler visits. --- Use **SEO Crawlers** to see automated requests from search engines and SEO tools. PromptEye keeps these separate from AI Traffic because a search engine crawl and an AI assistant fetching a page have different purposes. A rise in one should not be read as a rise in the other. ## Google Search Console If Search Console is connected to the project, this tab includes Google search clicks, impressions, click-through rate (CTR), average position, daily trends, top queries, and top pages. These numbers describe Google Search performance, not crawler requests. For example, a bot may crawl a page without anyone clicking it in Google results. The query and page lists help connect search visibility to content: impressions with few clicks may be a reason to review the title or snippet, while a page with clicks can be worth comparing with its AI crawl activity. Search Console can omit some queries and data may not be available immediately for a new connection or date range. If Search Console is not connected, the page offers a connection flow. Choose the Google property that matches the project website; choosing a different property would make the report describe another site. See [Settings and setup](../setup/) for connecting it. ## Crawler activity and crawl health **Top SEO Crawlers** ranks the identified search engine and SEO audit bots by their requests in the selected period. Use it to check whether search engines are reaching the site or whether an audit tool is scanning it. These visits do not represent human visitors. **SEO crawl health** groups successful requests, redirects, 4xx errors, and 5xx errors. Repeated 4xx responses can point to missing or blocked pages; frequent redirects can make crawlers take extra steps; repeated server errors or slow responses can make crawling less reliable. Check the path, status, and last-seen time before deciding something needs fixing: a redirect to a replacement URL can be intentional. Before reading a high error rate as a problem with your site, check how much of it was automated probing. Scanners request files that exist on no site, such as environment files or cloud credentials, and take a well-known crawler name while doing it. A line under the summary counts every probing request in the period, and a **Vulnerability scanning** filter narrows the issue table to the ones that failed or redirected, so the count is usually the larger of the two. The totals and thresholds do not change; separating them tells you whether search engines are struggling to reach your pages or whether a stranger is simply trying doors that were never there. ## Sitemap coverage Connect a public sitemap for a page-by-page comparison between URLs it lists and the bot visits PromptEye has recorded. The report groups URLs as recently visited, stale, never visited, or a gap/new URL. Adjust the stale threshold to decide how many days without a visit counts as stale, and use status, bot, or URL filters to find a specific group. **Sync now** asks PromptEye to read the sitemap again, so newly added or removed URLs can be reflected. **Export CSV** downloads the complete sitemap report, including URLs not loaded into the current table page and regardless of the current table filters; apply filters in your spreadsheet if you need a narrower handoff. **Remove** removes this sitemap configuration and its coverage report for the project. You can connect it again later. The sitemap URL must be publicly accessible and use the same domain as the project. A private, blocked, invalid, or different-domain URL cannot be connected. A sitemap shows what URLs are listed; it does not force a bot to visit them or prove they were indexed. --- # Traffic Settings and Setup Source: /help/ai-traffic/setup/ --- title: Traffic Settings and Setup description: Install AI traffic tracking and connect Google Analytics, Search Console, or a sitemap for a project. --- Use **Settings** to install AI traffic tracking for a project. The setup screen creates or accepts the credentials and code that your website needs to send identified bot visits to PromptEye. Selecting a platform and copying the code does not install it; someone with access to the website must deploy the snippet, plugin, middleware, or server configuration. ## Install AI traffic tracking Choose the project whose website you want to measure, then select the platform that matches how that site is hosted: - **Cloudflare Workers** provides a Worker collector. - **Vercel / Next.js** provides middleware for a Next.js deployment. - **WordPress** provides a ready-to-install plugin ZIP and optional plugin code. - **Laravel / Asgard CMS** provides middleware to add to the web middleware group. - **Nginx (njs)** provides server-level tracking for an Nginx setup with the njs module. The generated configuration includes the selected project identifier and an API key. The project identifier groups events under that project, so using the wrong project sends visits to the wrong dashboard. Each installation reports to one project: for several sites, create a project for each and install the tracker with that project's configuration. The key must have the **LLM tracking ingest** permission. Keep it in server-side configuration and do not publish it in browser code or a public repository. You can paste an existing key with the required permission or generate one in the setup flow. A newly generated key is shown only once: copy and save it before leaving the screen. If it is lost, create another key and update the website configuration. Reusing a permitted key can help track multiple projects under the account, but each generated snippet still needs the correct project identifier. Keep production, staging, and other domains attached to the correct PromptEye projects. Each generated configuration contains a project ID. If staging uses the production ID, its test visits will appear in production reports. The key grants permission to submit tracking events for the account; it is not a substitute for checking the project ID in the installed configuration. ### Install the generated files The setup screen gives you a ready-to-copy snippet or package for the platform you chose. The person deploying it should: | Platform | What to do on the website side | | --- | --- | | Cloudflare Workers | Create a Worker, add the `INGEST_URL`, `API_KEY`, and `PROJECT_ID` variables, then attach a route for the site's domain. The Worker sees requests at the edge, including requests served before they reach the origin. | | Vercel / Next.js | Put the generated middleware at the app root as `middleware.ts`, set the three `LBT_…` environment variables in Vercel, and deploy. The generated matcher skips Next.js static/image assets and the favicon. | | WordPress | Download the ZIP, upload it in **Plugins → Add New Plugin → Upload Plugin**, then install and activate it. The plugin has an admin settings page for its key and project ID. | | Laravel / Asgard CMS | Copy the middleware file into the application, add it to the `web` middleware group, and deploy. | | Nginx (njs) | Enable the Nginx JavaScript module, add the generated tracker and internal ingest location, and reload the Nginx configuration. This option is for whoever maintains the web server. | Use one collector at the layer that sees the requests you want to count. If you install the same tracker at Cloudflare and again in WordPress or Next.js, one request may create two events. If a CDN serves cached pages without invoking the origin application, an origin-only integration may not see those requests; an edge integration may be a better fit. Confirm the request path through your own hosting setup before choosing. After deployment, return to Traffic. PromptEye starts showing data when an identified bot reaches the site. Clicking **I've installed it** returns you to Overview; it does not verify that the code is live or create visits by itself. If no data appears, check the deployed environment, project identifier, key permission, and Logs after a bot request occurs. ### What happens to a request? The collector runs in your site's server or edge layer. It does not fetch or import your existing web-server log files. It checks each request's user-agent (the text that identifies the visiting software) against the bot names included in the installed tracker. Unmatched requests do not produce a PromptEye event. That means ordinary visitors are not counted as AI traffic, even though their requests pass through the same website infrastructure. The setup is called AI Traffic, but the installed list also contains selected search-engine and SEO audit bots. PromptEye separates those recognized requests into **SEO Crawlers** so they do not inflate the AI totals. For a match, the tracker sends a small event to PromptEye. It does not send the page contents, form submissions, request body, or cookie header. The event can include: - the time, matched bot name, provider, and bot category; - the requested path and user-agent text; - the referring URL when the browser or bot supplied one; - the address the request came from. PromptEye uses it to confirm who the visitor really was, as described below. Tracking code copied before this was added does not send it on Vercel / Next.js; recopy the snippet from this screen if you want that check on a Vercel site; - the response status, redirect destination, and response time on Cloudflare Workers, WordPress, and Laravel setups. The generated Vercel / Next.js and Nginx collectors currently do not include these response fields; - the country and Cloudflare bot-verification details on Cloudflare Workers. The exact fields depend on the platform. The WordPress and Laravel versions use the request URI, which can include query parameters; the other supplied versions send the URL path without its query string. A referring URL may also contain extra URL details. Because IP addresses and URLs can identify or reveal information about visitors, do not describe this integration as collecting no personal data. Review the fields against your privacy notice and applicable requirements before deployment. The snippets do not add browser JavaScript or set tracking cookies, but that alone does not decide whether a notice or other legal step is required. The installed tracker contains a fixed list of known bot signatures; bots missing from that list are not reported until the tracker is updated. ### Confirming that a bot is who it claims to be A user-agent is a claim the caller writes itself, so anyone can send a request that says it is GPTBot. To tell a real visit apart from an imitation, PromptEye compares the address a request came from with the address ranges that bot's operator publishes for exactly this purpose. The comparison runs on PromptEye's side, so a correction reaches your site without you redeploying anything. Each recorded request ends up in one of three states: | State | What it means | | --- | --- | | Confirmed | The request came from an address the operator publishes, or Cloudflare identified it as that operator's verified bot. | | Not confirmed | The operator publishes a list of its addresses and this request did not come from one of them. Someone borrowed the name. | | Not checked | The question could not be answered, so PromptEye makes no claim either way. | Seven of the services PromptEye tracks publish an address list: Google, OpenAI, Anthropic, Perplexity, Microsoft, Apple, and DuckDuckGo. Others, including Meta, Amazon, and ByteDance, publish nothing that can be checked this way, so their requests stay unchecked — unless your site runs the Cloudflare Worker collector, which can pass on Cloudflare's own verified-bot signal. That list is maintained by PromptEye and cannot be extended per project. A request is also left unchecked when the installed tracking code sends no address, or when the operator's published list does not cover the kind of address the request came from. These are kept apart from "not confirmed" on purpose: calling them imitations would accuse real crawlers, and calling them genuine would overstate what was proven. Confirmed means "not pretending to be someone else". It does not mean the visit was wanted or harmless: the same confirmed address range carries a crawler building an index and an assistant fetching a page because a stranger asked it to. PromptEye reports; it never blocks a request or changes what your site returns. #### What this needs from your installation The check uses the address your tracking code sends, so an older installation can leave every request unchecked: - **Cloudflare Workers** — nothing to do. - **Vercel / Next.js** — code copied before this feature sends no address at all. Recopy the snippet from this screen and redeploy. - **WordPress, Laravel, Nginx** — these already send an address, but if your site sits behind Cloudflare it is the Cloudflare edge's address, not the visitor's. Recopy the snippet so the tracker also sends the header that identifies the real caller. If your site sits behind a different CDN or load balancer — Fastly, Akamai, CloudFront, or your own proxy — PromptEye cannot tell that the address belongs to the proxy, and a genuine bot can be recorded as **not confirmed**. Until that is supported, read results from those sites with that in mind, or install the collector at the edge layer instead of the origin. #### Where to find it The Traffic screens do not show this state next to each request yet. It is available through the [PromptEye API](/help/developer/). The crawl health screens described below use a different signal, the requested path, and not this check. Requests recorded before this check existed are marked as not checked, except those Cloudflare had already identified as verified bots. ### Requests that are not crawling at all PromptEye also looks at what a request asked for, which is a separate signal from the address check above and works even when the address cannot be judged. Some paths are only ever requested by software probing a site for a way in: environment files, repository metadata, cloud credentials, and administrator areas. Alongside them PromptEye counts requests for source maps, the files that would hand over your original front-end code. Requests for real pages, images, scripts and sitemaps are never labelled this way, including WordPress assets that happen to live under an administrative folder. Requests like these are labelled as vulnerability scanning. They are still counted in every figure. **AI crawl health** and **SEO crawl health** tell you how many of the requests in the period were this kind of probing rather than a crawler reaching for your content, and a **Vulnerability scanning** filter narrows the issue table to the ones that ended in an error or a redirect. The count covers every probing request in the period; the table only ever lists the issues it already shows, so a large count with few rows means most of those requests did not produce an error. The segment appears only when there is something to count. See [AI Traffic](../ai-traffic/) and [SEO Crawlers](../seo-crawlers/). ### Does tracking slow down the site? The supplied collectors send the event after, or alongside, the page response wherever the platform allows it, so visitors do not wait for it. If sending fails, the page is still served normally. The tracker still uses a little server or edge capacity, and on some WordPress and Laravel setups the event is sent at the end of the request, which can add a short delay there. Failed events are not sent again later. If PromptEye cannot be reached at that moment, that bot visit may be missing from the reports. ## Connect Google Analytics Google Analytics adds a human-visit view for sessions attributed to identified AI referrals: engagement, session duration, key events, sources, and landing pages. It does not measure bot requests and does not replace the website tracker. Choose **Connect Google** and authorize the Google account that can access the site's property. Select the Analytics property for this project and confirm the connection. The selected property controls which site's sessions PromptEye reads. If the account has access to multiple properties, confirm the site name before binding it. Connecting grants PromptEye access to read that property's analytics for the report; it does not add code to the website or change its Analytics settings. If the connection succeeds but no report appears, check that the selected property has data for the requested dates and that Google can still authorize access. To use a different property, disconnect it and connect again, choosing the right property. Disconnecting stops using that property for this project's referral report. The AI visit tracker and its bot logs are separate. You can reconnect a property later. ## Connect Google Search Console Search Console adds Google search performance: clicks, impressions, CTR, average position, queries, and pages. It is separate from AI bot traffic and from Google Analytics sessions. Choose **Connect Google** and authorize the Google account that can access the site in Search Console. Select the property matching the project domain and confirm the connection. The property you bind determines which search data appears in SEO Crawlers. If a site has both URL-prefix and domain properties, select the one that represents the URLs you want to report. Connecting grants PromptEye access to read that property's search performance; it does not change the site's indexing or Search Console settings. Data can be absent when Google has no results for the selected range or the connected account cannot access that property. Disconnecting Search Console stops displaying its search performance for the project; it does not uninstall the bot tracker or remove sitemap configuration. You can connect the property again when needed. ## Connect and manage a sitemap In **SEO Crawlers → Sitemap coverage**, connect a public HTTP or HTTPS sitemap on the same domain as the project. PromptEye reads its URLs and compares them with observed bot visits. A sitemap helps find listed pages that crawlers have not reached; it does not submit pages to a search engine or guarantee indexing. The connection can fail if the address is invalid, blocked, private, or belongs to a different domain. After connecting, wait for the first sync or choose **Sync now** to request another read. Removing the sitemap clears this project's sitemap coverage configuration and report; it does not remove the site's sitemap file or affect AI tracking and Google connections. ## Who can make changes? Some setup actions are disabled for read-only project access. A project editor or owner needs to perform the connection or installation setup. If you can view reports but cannot connect an integration, ask a project owner to make the change. ## Common questions ### Does PromptEye collect ordinary visitors' browsing history? The supplied tracker sends an event only when a request matches a bot signature. It does not send events for unmatched visitors, and it does not send page contents, form data, request bodies, or cookie headers. Some event fields can still be sensitive: all five supplied collectors now send the address a request came from, because that is what the identity check compares, and WordPress/Laravel paths can contain query parameters. The addresses are compared against lists PromptEye downloads from the operators; nothing about your visitors is sent to those operators. Check the exact fields above before deciding what to disclose to your visitors. ### Do I need to add JavaScript or a cookie banner? The supplied integrations run on the server or edge and do not add a browser tracking script or set tracking cookies. However, some versions send IP addresses and URLs to PromptEye. Whether your site needs a privacy notice or another step depends on your circumstances; do not assume that server-side tracking automatically removes those obligations. ### I do not have server access. Can I finish setup myself? You can select the platform and prepare the key and installation package, but someone with access to the hosting, web server, or application deployment must install it. You can send them this request: > Please install the PromptEye AI Traffic tracker for project **[project name/domain]** using the attached instructions. Keep the API key in server-side configuration, confirm the project ID is correct, and tell me when the deployment is live. Copying the snippet or clicking the confirmation button in PromptEye does not deploy it. ### Can I exclude paths such as `/admin`, `/api`, or `/checkout`? There is no path-exclusion control in PromptEye's setup screen. The supplied trackers match bot requests across their configured routes. Vercel's generated matcher omits Next.js static/image assets and `favicon.ico`; that is a built-in exception, not a setting. If you need other paths excluded, ask whoever maintains the integration to change its server-side matching rule before it sends events. ### How do I test without waiting for a real bot? On a staging site or a separate test project, you can send a request with a known user-agent, for example: ```sh curl -A 'GPTBot' https://staging.example.com/ ``` The collector will treat that matching text as a bot and record a test visit. Where the installed code sends your address, PromptEye records the visit as not confirmed, because it did not come from an OpenAI address — that is the check working, not a fault. On an installation that sends no usable address, or from a staging site behind a CDN PromptEye does not recognize, the same request comes back as not checked instead. Avoid running it against production unless you want the artificial visit to affect its counts; there is no delete-this-test-row action in Logs. Do not test by asking an AI assistant to open your page. The assistant really does fetch it, so the traffic you are trying to measure becomes traffic you created, and the more you test the more the report agrees with you. ### I lost the API key or it appeared in a public place. What now? The full key cannot be displayed again from its masked value. If it is only lost, generate a replacement and update the server-side configuration. If it may have been exposed, delete that key from **Settings → Developer** so it can no longer submit events, then generate a replacement and deploy it. Deleting a key affects every integration that uses it; update those installations before expecting their events to resume. Revoking a key stops future submissions with that key; it is not documented as deleting events already recorded. ### Why are the counts different, and is a spike an attack? AI Traffic counts identified bot requests, Logs shows up to the 200 newest matching requests for the selected period, and Google Analytics reports human sessions attributed to identified AI referrals. These are different events and should not have matching totals. A spike can be a crawler fetching many pages, or it can be someone borrowing a well-known name. Start with the paths: a burst of requests for environment files, administrator logins, or repository metadata is scanning, and crawl health now labels it as such. Then check the recorded confirmation state through the API. Confirmed requests came from the operator's own addresses; a spike of requests that are not confirmed is worth showing to whoever maintains your hosting, together with the server's own logs. ### How long are events kept, and how do I delete them? There is no action for deleting individual traffic events, including test visits. Removing the tracker stops new events after the code is no longer running; deleting an API key prevents that key from submitting future events. Neither action deletes events already recorded. Logs and its CSV export show at most the 200 newest matching requests for the selected period; this display limit is not a retention schedule. If you need a specific retention or deletion period, confirm it with your PromptEye contact before enabling tracking. ### Why do I see only some bots, or a bot I do not recognize? The generated integration includes a fixed list of known names. A bot not on that list will not be counted, and a caller can write any of the listed names into its own requests. PromptEye records a bot under the name it claimed and stores separately whether the address behind it confirms that claim, so a name you do not recognize is a starting point for a look rather than a conclusion. The report is a useful sample of identified activity, not a complete census or security alert system. ### Why are AI bots not reaching my pages? PromptEye can report only requests that reach the layer where the collector is installed and match a known bot name. Check whether your `robots.txt`, firewall/WAF, or hosting rules block the bot, whether a CDN serves the request without reaching the installed collector, and whether the sitemap includes the missing page. **Sitemap coverage** can show URLs that have never had a recorded bot visit, but it cannot tell you by itself whether a bot was blocked or why it did not visit. Your hosting or firewall logs can help identify that cause. --- # Reports overview Source: /help/public-reports/reports/ --- title: Reports overview description: Report list, filters, export, and lead status. --- The **Reports** tab shows public reports created for your agency account and contact requests submitted through those reports. A public report shares an assessment of a prospective client's brand visibility in AI answers and gives them a way to contact your agency. You can use it to start a conversation about their needs. See [how to read the score](/help/public-reports/reports/score/) and [how to generate a test report](/help/public-reports/reports/create-report/). The list is grouped by lead status: **New**, **In progress**, and **Handled**. Lead status is separate from report generation status. The cards at the top switch the view; **Urgent** shows reports with at least one contact request. The email column shows the address stored when the report was created. Search filters by brand name, domain, and email. You can sort the table by date, brand, domain, and score. The percentage appears when a report is ready; a dash appears while it is processing or if it has failed. The **CSV** button exports selected reports when a selection exists; otherwise it exports reports matching the search and tag filter across all status groups, with contact details and lead status. Select a row to open the details panel. There you can change lead status, review contact requests, open the report, control whether clients see its full version, resend it to the email address stored with the report, convert it to a project, or delete it. The row menu also lets you copy its link. If the monthly counter is shown, it displays how many reports have been used and the limit for the current billing period. The limit renews when a new billing period starts. Public reports do not use your projects' prompt slots. ## What the recipient receives When a report is ready, PromptEye emails its link to the address given in the request. The email is in German for reports requested in German, in Polish for reports requested in Polish or without a language, and in English for other languages. It uses the logo and primary color from **Branding**. At the same time, the addresses saved under notification emails on the **Integration** tab get a notice that a new report was generated; if that field is empty, the notice goes to your account email. The same addresses are notified when a recipient sends a contact request from the report. A report is checked with ChatGPT, Perplexity, DeepSeek, Gemini, and Google AI Overview. A report cannot be edited or regenerated. To get fresh results, request a new report; the same request within 30 days may return the earlier report (see [Generate a test report](/help/public-reports/reports/create-report/)). ## Organize reports with tags Tags let you categorize reports using your own labels and filter them into useful groups. Select report checkboxes, open **Bulk actions**, and choose **Assign tags**. Enter a label, press Enter, then select **Apply**. New tags are created when assigned; existing tags appear as suggestions. Assigning tags preserves the lead's other tags. Use **Remove tags** to remove selected labels. Tags are internal labels, not account invitations or access permissions. Each report can have up to 20 tags, with up to 50 characters per tag; matching ignores letter case. Open the **Tags** menu beside search to show reports with a selected tag. The button shows the active tag; choose **All tags** to clear it. The checkbox in each table header selects that entire group, including rows below the visible part of the screen. In the **Select** menu, **Select all reports across all groups** includes all groups matching the current search and tag filter, even if a status card hides some groups. Changing the search or tag filter clears the selection. The selected count shows how many reports an action will affect. ## Work with selected reports **Bulk actions** can change status, assign or remove tags, create projects, enable or disable the full version, or delete reports. Review the selected count and confirm the action. Full-version access applies to anyone opening those public report links. Deletion permanently removes the reports and stops their public links from working; projects already created from them remain. The single **CSV** button exports only selected reports when a selection exists and shows their count. With no selection it exports all reports matching search and tag filters. Both exports include email addresses and tags. Bulk actions appear only when reports are selected. Actions process reports individually: a failure does not undo earlier successes. Keep the page open until the result appears. Failed and skipped reports remain selected for another attempt; the result reports each count. --- # Generate a test report Source: /help/public-reports/reports/create-report/ --- title: Generate a test report description: Test form, generation time, and report reuse. --- The **Generate report** button in the Public Reports header opens a test form. It creates a report the same way as a form connected to your website through the **Integration** tab. Enter a brand and an email address. You can also provide a domain, then choose a country, language, and local, regional, or countrywide reach. Once the request is accepted, the form closes. A newly created report appears in the **Reports** tab; if an existing report is reused, its earlier entry remains on the list. An accepted request may still be processing, so the result can appear later. If the same agency requests a report with the same domain, brand, language, country, and reach within 30 days, the system can return an existing report, including one still processing or one that failed. For this comparison, leading and trailing spaces in the brand are ignored, as is letter case. A request without a domain does not reuse an earlier report. When a ready report is reused, its email is sent to the address in the new request. Reusing a report that is still processing or has failed does not schedule an email to that new address. If something goes wrong, check the report's status in the list. New reports count toward the account's limit for the billing period. When available, usage and the limit are shown in the **Reports** tab. A new request is rejected after the limit is reached. --- # Manage leads and reports Source: /help/public-reports/reports/manage-leads/ --- title: Manage leads and reports description: Contact requests, lead status, and report actions. --- Open a report from the list to review contact requests and set its lead status to **New**, **In progress**, or **Handled**. The **Urgent** marker appears on reports with at least one contact request. Select the latest contact to see its saved details. Depending on how the recipient contacted you, these may come from a message, a callback request, or a Calendly meeting. In the report panel, you can open its public link, resend the email to the address stored with the report, and control whether recipients see the full version. The report owner sees the full data. When the full version is turned off for recipients, the public view limits example prompts and competitors to three each. From the row menu, you can copy the link, open the report, convert a ready report to a project, or delete it. Conversion is available only when the report is ready and has not already been converted. Deletion requires confirmation and cannot be undone. --- # Understand the report score Source: /help/public-reports/reports/score/ --- title: Understand the report score description: What the percentage, model results, and competitor comparison mean. --- The report score shows how often the brand appeared in AI model answers to the prompts used in that report. The percentage is the share of analyzable answers in which the brand was detected. For example, 40% means the brand appeared in four out of ten such answers. Answers whose analysis failed are not counted as answers without a brand mention. If there are no analyzable answers, the report fails and no score is published. The report also shows results for individual models, example prompts and answers, and a comparison with competing brands. A competitor's score represents its share of mentions across the analyzed answers. A brand's position within an answer is separate from the percentage score. Use the score as a view of brand visibility for the selected prompts, models, language, country, and reach. It does not measure website traffic or overall search ranking. When comparing reports, check that they use the same settings. --- # Projects from reports Source: /help/public-reports/projects/ --- title: Projects from reports description: Convert a ready report to a project and manage converted projects. --- The **Projects** tab shows projects created by converting public reports. In the **Reports** tab, open a ready report's menu or details panel and select **Convert to project**. Choose a country in the form; you can also add an optional label. If the report has no domain, you must enter one before converting it. After a successful conversion, the app opens the new project's overview. The report's prompts become the project's monitored prompts, and they use the workspace's prompt slots. Conversion needs a paid plan or an active trial, and it fails if the workspace does not have enough free prompt slots for all of the report's prompts. If you belong to several workspaces, you can choose where the project is created; you need to be an owner, admin, or member there. A report can be converted only once, and the conversion cannot be undone. To remove the project, delete it. The project list includes a delete action with confirmation. If the list is empty, return to **Reports** and select a report to convert. For several reports, select them in **Reports** and choose **Bulk actions → Create projects**. Each project uses the country stored in its report. Reports still processing or in error, without a domain or supported country, or already converted are skipped. Use the individual conversion form to supply a missing domain or choose a country. Each successful project consumes the applicable prompt allowance; a later failure leaves earlier projects in place. --- # Connect a form to Public Reports Source: /help/public-reports/info/ --- title: Connect a form to Public Reports description: Agency ID, the generation endpoint, and integration examples. --- The **Integration** tab provides your agency ID and a sample report request for the person connecting your website or form. Copy your **Agency ID** and share it with the person building the integration. The same tab shows the `POST /generate-report` endpoint and examples in text, cURL, Axios, and Fetch formats. Separate buttons copy the example code and request type. The request must contain `agencyId`, `brand`, `email`, and `country`. Send `country` as a supported two-letter ISO country code, such as `pl` or `PL`; a country name such as `Poland` returns **400 Bad Request**. Optional `language` must be `en`, `pl`, `de`, or `es`. The endpoint stores codes in lowercase. You can also include `website`, `reach`, and `utm`. The in-app example shows the exact `reach` values and request shape. You may omit the domain, but you will need one later if you want to convert the report to a project. An accepted request appears in **Reports**. Repeating the same request with a domain may reuse a report from the past 30 days. If that report is still processing or has failed, the repeated request does not schedule an email to the new address; check the report's status in the app. Use `local`, `regional`, or `national` for reach; `national` means countrywide in the selected market. Enter a Calendly URL to add a **Meeting** tab to the report contact form. Save an empty URL to remove that meeting option. --- # Notifications and contact options Source: /help/public-reports/info/notifications/ --- title: Notifications and contact options description: Notification addresses and the Calendly link for public reports. --- In the **Integration** tab, you can save notification email addresses and a Calendly link for report recipients. Enter comma-separated addresses in the notification field and save your changes. The app removes empty entries and duplicates. Saved addresses are notified when a report is sent to a recipient and when a recipient sends a contact request. If the field is empty, these notifications go to your account email. This is the same **Notifications email** setting you see in **Settings**. Enter a meeting URL in the Calendly field and select **Save**. Once the link is saved, a **Meeting** tab can appear in the report's contact form. Saving an empty field removes the URL. The agency ID and request examples are further down the same tab. --- # Public report branding Source: /help/public-reports/branding/ --- title: Public report branding description: Logo, fonts, and colors for reports shared with recipients. --- The **Branding** tab lets you configure the appearance of public reports shared with recipients. You can upload a logo, enter font import URLs and family names for headings, body text, and labels, and choose primary and additional colors. Additional color fields cover borders, secondary text, errors, and success states. After changing fonts or colors, select **Save**. These settings belong to your account and apply to every public report, including links you have already sent: the next time someone opens a report, they see the current branding. The logo and primary color also appear in the email that sends the report to the recipient. The logo has a separate upload control. The same logo and report styling can also be managed in **Settings**; both places change the same settings. Select **Restore defaults** to clear all custom fonts and colors and immediately save the default report styling. This also discards unsaved font and color edits. Your shared account logo has its own remove control and is not removed by the style reset. You can enter new styling values and save them again at any time.