AI visibility tracker for ChatGPT, Claude, Gemini, Perplexity and Google
ApexMention's AI visibility tracker asks your buyers' questions in ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode, each question once a week. It records whether every answer names you, in what place, who it names instead and which pages it cites. Then the same product writes for the questions you're missing and pitches the sites the answers rely on.
What it measures
AI search visibility comes down to one fact per answer: did the assistant name you? The tracker records that for every answer, together with every brand the answer names, in order, and every page it cites. The numbers in your report are counted from those records.
| Measure | What it counts |
|---|---|
| Named in answers | The share of answers that name you |
| Share of voice | Your mentions as a share of all brand mentions. Each brand counts once per answer |
| Average place | Where you come among the brands an answer names, averaged over the answers that name you. 1 is first |
| Win rate | The share of all answers that name you first |
| Sources | Every page the answers cite, counted by page and by site |
| Source types | Each cited site sorted into one of 8 types (your site, competitor, review site, forum, video, social, reference and publisher), each with what to do about it |
| Cited, not named | Answers that cite one of your own pages and still recommend someone else, taken from the newest answer to each question on each assistant |
An answer names you when your brand name, your domain or another name you have added for the business appears in it as a whole word. A small AI model reads each answer and lists the brands it names in order, which is where place and share of voice come from. That model can misread an unusual name. Adding the spelling as an alias corrects the answers already recorded as well as new ones.
Failed checks are left out of every rate, and so are questions where Google showed no AI Overview. The report compares the period you pick with the one before it. Questions rotate through the week, so each point on a trend line covers the seven days ending that day.
How each assistant is asked
Four of the six are read the way a person meets them, from the app or Google's page. Claude and Perplexity are asked through their companies' APIs, and an API's answer can differ from what the app shows.
| Assistant | Where the answer comes from | Asked from |
|---|---|---|
| ChatGPT | ChatGPT's app as a logged-out user sees it, read through a scraping provider | Your site's country |
| Gemini | Gemini's app as a logged-out user sees it, read through a scraping provider | Your site's country |
| Google AI Mode | AI Mode as a logged-out user sees it, read through a scraping provider | Your site's country and language |
| Google AI Overviews | The AI Overview on Google's results page for the question, read through a search data provider | Your site's country and language |
| Claude | Anthropic's API, with its web search tool on | No location is set |
| Perplexity | Perplexity's API, with its Sonar model | No location is set |
ChatGPT
If the app can't be read on a given check, the check falls back to OpenAI's API with web search on, and the answer records which of the two it came from. We also keep the web searches ChatGPT ran before answering.
OpenAI's documentation of its crawlers says sites that block OAI-SearchBot are left out of ChatGPT's search answers, though they can still appear as navigation links. If ChatGPT never cites your pages, check your robots.txt with the AI crawler checker before anything else.
Gemini
Gemini is read the same way as ChatGPT, with the same fallback to Google's own API. Google documents how AI Overviews and AI Mode choose their links. We found no equivalent page for the Gemini app, so we don't describe one.
Google AI Mode
Google's page on its AI features says AI Mode and AI Overviews may use different models and techniques, so the answers and links they show for the same question will differ. That is why they are two rows in your report and not one. The same page says both may split one question into several related searches, which Google calls query fan-out, before answering.
Google AI Overviews
Google says it shows an AI Overview only when its systems judge that it adds to the normal results, so many questions get none. We record those as having no AI Overview and count them as neither a mention nor a miss, so they don't pull your rate down. To appear as a link in an AI Overview, the same page says, a page has to be indexed and eligible to show with a snippet in Search, and there is no further technical requirement.
Claude
Claude gets two web searches per answer at most, to keep each check cheap. Anthropic's documentation for its web search tool says Claude decides from the question whether to search, and answers from what it already knows when the question doesn't need current information. So some Claude answers cite no pages at all, and any brand they name comes from the model's training rather than from a page it read.
Perplexity
Perplexity's help centre page on how it works says its app searches the web for each question and gives every answer numbered citations; we ask its API, which can differ from the app. Its crawler documentation for PerplexityBot says PerplexityBot is the crawler that brings sites into Perplexity's results and is not used to train models, so it is the one to let in.
Why each question is checked weekly
Every tracked question is asked on every assistant once a week. The checks are spread over the week, a few questions a day, so there are always recent answers. A new site's first check asks every question at once so the report starts complete, and you can start a check by hand at any time.
There are two reasons for weekly. Each check is a paid request to each assistant, so asking every question every day would cost seven times as much. And asking the same question more often moves the reading less than you would expect. Profound ran two copies of the same tracking for 14 days in June 2026, on 753 prompts across seven AI platforms: one asked each prompt once a day, the other ten times a day. Visibility came out at 78.7% and 80.4% (Profound, 8 July 2026). Its author, the economist Jennifer Zou, also found that most of the day-to-day movement left over comes from the platforms themselves changing, which asking more often cannot remove.
That test compared one run a day with ten, not one a week with one a day, so it does not show that weekly is enough. We have not measured that ourselves yet. If you need to see a change within a day of making it, a daily tracker such as Peec AI or Profound is the better tool.
What happens with the results
An AI visibility tool on its own tells you where you are missing. In ApexMention the results feed three kinds of work, each started with one click, and the following weeks' checks show whether it worked.
A question you're missing becomes an article
One click turns a tracked question into a planned article. It takes the next free place in the schedule, and the article that was there moves back. The article stays tied to the question, so you can see whether the answers changed after it was published.
Pages the answers cite become pitches
The Win citations list collects the pages cited in answers that leave you out. A page on a competitor's own site is one to answer with an article of yours. A publisher, review site or forum is one to get mentioned on. For a publisher, ApexMention looks up the author and a work email, drafts a short note and sends it from your own Gmail once you approve it: at most 15 a week per site, with one follow-up a week later. The page is read again every week. When it names or links you, the pitch is marked as won and the question is asked of every assistant again.
Pages that are read but don't win get rewritten
When an assistant cites one of your published articles and recommends someone else, the cited, not named list shows the question, the page and who was named instead. One click rewrites the article to say plainly who you suit and why, with one fact to back it, and the update waits for your review. For pages ApexMention didn't write, you get the same instruction to copy.
Questions
- Is an AI visibility tracker the same as an LLM visibility tracker?
- Yes, people use both names for the same kind of tool. One distinction is worth knowing: what gets measured is the assistant with its web search on, not the language model on its own. A model asked without search can only name brands from its training. With search on it reads pages first, which is why the cited pages are where most of the work is.
- How is this different from the free AI visibility checker?
- The free checker asks three questions once, through each company's API with web search on, and keeps nothing. The tracker asks your own list of questions every week on six assistants, reads ChatGPT, Gemini and AI Mode from their apps, and keeps every answer so you can see a trend.
- How many questions does it track?
- It starts with 15 written from your site: best-of lists, comparisons and problems your buyers try to solve. None contains your brand name, because the point is to see whether assistants bring you up unprompted. You can add your own and remove any.
- Can I read the answers themselves?
- Yes. The newest answer from each assistant to each question is kept in full, with the brands it named and the pages it cited, so you can read them side by side.
Sources
Each one was opened and read, most recently on 16 November 2026; where a figure was read on an earlier day, the text gives that date. Figures in the text say what was counted, how many and when.
- 1.Documentation of its crawlers · developers.openai.com
- 2.Page on its AI features · developers.google.com
- 3.Documentation for its web search tool · platform.claude.com
- 4.Help centre page on how it works · perplexity.ai
- 5.Crawler documentation for PerplexityBot · docs.perplexity.ai
- 6.Profound, 8 July 2026 · tryprofound.com