Documentation overview

AI Visibility (GEO)

Be found and cited in AI answers. See whether the AI answers your buyers read name you and cite your site, whether AI search bots and agents can reach and use your site, and what to do so they find and read you.

Open it from Studios → AI Visibility. It is GEO and the AI Tracker as one tool: one setup, one Run, and one page with five tabs – What to do, Answers, Site & agents, Setup and History. A finished check opens on What to do. Old links to the AI Tracker or to GEO open the same page.

For one brand, AI Visibility answers four questions:

  1. Do ChatGPT search, Gemini, Claude and Google's AI Mode and AI Overviews name you and cite you?
  2. Can AI search bots reach your pages and read them?
  3. Can AI agents that act for a person use your site: find the price, contact or book, and sign up?
  4. What should you change?

No outside tool can see how an engine chooses its answers, and Google says so itself. GEO reads a sample of real answers and your site as the bots and agents see it, and tells you what helps them find and read you. It never promises a place in an answer.

Set it up

The Setup tab holds everything the check reads.

  • Site comes from your brand profile.
  • Pages to read – Top 10 or Top 25 of your most visited pages, from your site's own page list, or Choose to pick them. Every page is one you have; none are guessed.
  • Who to ask about – the brand, its website and what a buyer calls it, from your profile. Competitors are counted for your share of voice.
  • Questions – suggested ones are filled in; edit, remove or add your own, or Find questions from what buyers search (with the monthly demand behind each). Each shows whether it names you. The same questions are asked every time, so your runs read one trend.
  • Asked from is your brand's market. Google AI Mode is not offered in every country; where it isn't, its column says so.
  • Where to ask – the places the questions are asked:
    • ChatGPT search, Gemini, Claude and Google AI Mode and AI Overviews are ticked. Each gives the answer and the sites it cites.
    • Claude is read as "Claude with web search, asked through its API". No outside read of the Claude app exists, so the column says so.
    • Perplexity is unticked. It sends about 6% of the visits AI assistants send to websites (Statcounter, Sep 2026). Tick it to add its answers.
    • Untick every place for a check of your site on its own.
    • Copilot can't be asked from outside. Its citations come from your Bing report (see Add what only you can see below).
  • AI agents → Tasks to try – your price, contact or booking, and sign-up pages, found from your home page's links. Each is read the two ways agents read a page: as HTML without scripts, and in a browser.
  • Also ask from memory – the models' own answers without searching, at 1 credit an answer. None is ticked.
  • Add past AI answers that mention you reads earlier answers that named you. It is offered when your market includes the United States in English, and its price shows beside it.

Run the check starts it, with its price beside the button. In a few seconds you see whether AI search bots and agents can reach and read your site; the answers follow within a couple of minutes. You can leave the page; the check carries on and is saved under History and in your Library.

When the brand named is another company, or only the models from memory are ticked, the button reads Ask the assistants and asks the answers alone: the site and agents check is for your own brand's site.

What it costs

A check is priced on what it reads. The button shows the price for what you ticked: about 20 credits for the default check. Claude is the largest part, because every Claude answer runs its own web search. Before the check starts, the most it can cost is held; when it ends you pay for what it read, and the rest comes back to your credits. A check that fails is refunded in full.

  • A check of your site on its own (every place unticked) is 3 credits.
  • Opening a check again, on the page, in the Library or in chat, is 0 credits.
  • Check again on a file fix, or on an agent fix our fetches can read, is 0 credits. On an agent fix that needs the browser (named controls, layout shift, a CAPTCHA) it is 3 credits.
  • The weekly check is priced the same way; its price shows beside the switch.

What to do

The What to do tab opens with one line about your brand – for example "Named in 5 of 32 AI answers about your category. 3 things first" – then the figures: named in answers, your site cited, AI search bots, readable as HTML, AI agents and visits from AI. A figure that could not be read says "Not read", never 0. Export saves the report as a file; Check again opens the setup as this check had it.

Then one list of what to do, in order (see below), Where you stand, and Keep watching.

Where you stand

One line each for In the answers, Access, Readable, AI agents, Answer-ready, Structured data, llms.txt and Measurement. Each says Fix now, Watch, Clean or Not read, in one sentence, and what it was read from. There is no score and no grade.

The list

The fixes, numbered in order, then the pages to make that the answers point to, each once. Each says why, what to change, where, and the evidence behind it, and ends in an action:

  • Write it and Copy brief – a page brief to write from: the page to make for a question where you are missing, or an answer-first rewrite of a page you already have (the key fact at the top, headings that say what follows, comparisons as tables).
  • Draft the review request and Draft the pitch – for the sites the answers cite. GEO reads each cited page and says whether it names you. The advice is real listings, real reviews and real answers given as yourselves, never paid or fake mentions.
  • Open in SEO Meta & Schema – structured data fixes open there with the page and the additions filled in. GEO writes no markup itself.
  • Add as a task – any item goes to Projects as a task.
  • Ask about this – opens the chat beside the check, on that item, so you can ask why it matters, or have it write or plan the change.

Site & agents

The Site & agents tab holds what the check read from your site.

AI agents

Agents come in two kinds. Fetchers read a page's HTML without running scripts: ChatGPT, Claude, Perplexity and Google when they open a link, and the page readers in OpenClaw and Hermes Agent. Browser agents drive a real browser and look like an ordinary visitor: ChatGPT Atlas, Claude in Chrome, Comet, Copilot in Edge, and the browsers in OpenClaw and Hermes Agent.

For each task (find the price, contact or book, sign up) the report shows Without scripts and In a browser: Open, or what stands in the way – an empty page, a CAPTCHA, a login, a control without a name, a layout that shifts. A task is open when the page answers, the facts are in what the agent reads, every control on the way has a name, and no CAPTCHA or login stands in the way. It is a check of the path, not a run of any one agent.

Who we checked as lists each agent with how it reaches you, what it does with robots.txt, our test and the source. Each test is a fetch from our servers with that agent's user agent; the real agents come from their own addresses. Agents in a normal browser can't be told apart from people, and the row says so.

Your files

robots.txt, llms.txt, Structured data and IndexNow, each read as it is.

  • When a file is fine, its tab says Nothing to change and shows what was read. Lines to add appear only when something needs changing.
  • robots.txt offers Keep as it is or Decline AI training. Declining training does not remove you from ChatGPT, Claude, Google or Apple search. A bot is named in its own group only when it must be treated differently from the * group. Copy the lines or Download robots.txt with them.
  • llms.txt – when you have one, GEO reads its outline. When you have none, it shows "Not there · optional" and builds one from your real pages for you to publish. No AI engine documents using it to pick answers, and the tab says so.
  • IndexNow – when you publish a changed page, Tell search engines about it in one request instead of waiting for a crawl. Without a key file, GEO gives you the key file to put on your site.
  • When a firewall blocks an AI bot or agent, the report lists the user agents and the official address lists to allow together.

After you publish a change, Check again reads it from your site.

Answers

The Answers tab is the measurement: Mentions, Cited and Share of voice for each place buyers ask, then visibility, position, every question against every place, the sites the answers cite and whether each cited page names you, and what the answers get wrong about you. See AI Visibility: the answers. The AI Overview is counted on its own row, because Google does not show one for every question.

Visits AI assistants sent

Visits from ChatGPT, Gemini, Perplexity, Copilot and Claude in the last 30 days, by landing page, from your site analytics. Many apps hide where a visit came from, so the count is a floor. Agents in a normal browser count as people.

Everything we checked

Every check with its result and its source, each labelled Official, Best practice, Measured (a published third-party measurement), Measured by us, From your Site Audit, Proposal or Unverified. Kept out of search, the sitemap and titles come from your latest Site Audit; with none, the row offers Run a Site Audit.

Keep watching

Check every week, under What to do, is the brand's one weekly check, on the same questions, so the brand gets one alert a week, not two. You are told when an AI search bot or agent is newly blocked or gets a challenge page, robots.txt stops answering, a page loses its text in the HTML, a CAPTCHA or login appears on the way to a task, a task page loses its link from your home page, llms.txt turns into a web page, or you drop out of an answer you were in.

Add what only you can see

The last block of the report takes three files no crawler can read:

  • Google AI impressions per page – the export of Search Console's generative AI report.
  • Copilot citations – the export of Bing Webmaster Tools' AI Performance report.
  • Which AI bots and agents fetch your pages – a server or CDN log. Each hit is checked against the addresses OpenAI, Anthropic, Perplexity, Google and Common Crawl publish.

Only the totals are kept: never a raw log line and never an IP address. Adding a file costs 0 credits.

History

Every check and every earlier run, newest first: full checks, checks of the site on its own, and runs that asked the answers alone. Each opens on What to do. A check can be deleted, with its Library entry.

In chat and with your AI team

Chat at the top of the page opens the chat beside the check you have open; it knows the check. Ask chat whether AI answers name you, whether AI bots can reach your site, or whether an agent can sign someone up, and it runs the same check; a card under the reply shows where you stand and opens What to do, and the page opens the check the chat ran. Ask it for an llms.txt or robots.txt lines and it writes the file from your site for 0 credits. The SEO Strategist uses the same check.