AI visibility: how to check what ChatGPT, Gemini and Perplexity say about your brand

What AI visibility means and what you can actually measure

People look up ai visibility when they notice that customers no longer type a query and open ten links. They ask ChatGPT, Gemini or Perplexity for a recommendation and read one composed answer. AI visibility is the plain question behind that: when an assistant is asked about your category, does it name your brand, does it describe it correctly and does it point to your site? It is not a ranking position, because an answer is a paragraph with a few sources, not a list.

No public dashboard reports your share of answers across all assistants, and nobody outside the vendors sees everything that people ask. What you can measure is narrower and still useful. Write a fixed set of questions, put it to each system, record what comes back, and ask the same questions next month. The difference between the two records is your trend..

Four systems that find and quote a business in different ways

Google describes its AI features in its Search documentation. AI Overviews and AI Mode use a technique Google calls query fan-out: several related searches are run to build one response. Google says this lets the features show a wider and more diverse set of links than a classic search, and that the two features may use different models, so their links can differ. To appear as a supporting link, a page must be indexed and eligible to be shown in Search with a snippet.

OpenAI documents three separate crawlers. OAI-SearchBot helps surface websites in ChatGPT search features, GPTBot relates to training of foundation models, and ChatGPT-User acts when a person asks ChatGPT to visit a page. OpenAI states that sites which opt out of OAI-SearchBot will not appear in ChatGPT search answers, although they can still appear as navigational links. Perplexity publishes a similar split: PerplexityBot surfaces and links websites in Perplexity search results and is not used for foundation model training, while Perplexity-User visits a page when a user’s question calls for it and may include a link.

Microsoft’s Bing Webmaster Tools adds a fourth view. Its AI Performance report covers Microsoft Copilot, AI-generated summaries in Bing and select partner integrations. The practical lesson is that four systems have four sets of rules. Checking one assistant and assuming the rest behave alike gives a false picture, so the method below puts every question to every system.

Check the doors first: robots.txt, indexing and crawler rules

Before measuring anything, make sure nothing blocks the site. Google advises keeping crawling allowed in robots.txt and on any CDN or hosting layer, and says there are no extra requirements for AI Overviews and AI Mode: no special markup, no machine-readable AI file. OpenAI recommends allowing OAI-SearchBot in robots.txt, and allowing requests from its published IP ranges, if you want to appear in ChatGPT search. It adds that a robots.txt change affecting search can take about 24 hours to take effect. Perplexity gives the same advice for PerplexityBot.

The settings are independent. OpenAI notes that a site can allow OAI-SearchBot and still disallow GPTBot, so a company can choose to be quoted in answers without agreeing to training use. What should not happen is an accident: one wrong line in robots.txt or a firewall rule at the hosting layer can hide a good site from an assistant, and looking for it costs nothing.

A fixed list of buyer questions beats random prompting

Opening an assistant and asking what it knows about your company is a start, not a measurement. Write a list of fifteen to twenty questions in the words your customers use. Mix the types: category questions such as who offers a service in your city for a particular kind of client, comparison questions, problem questions that begin with how do I, questions about price and trust, and a few that contain your brand name.

Take the wording from real sources: enquiries you received, emails to support, the search queries listed in Search Console. Write each question down exactly and do not edit it later, because a list that changes every month cannot show change. Keep one language per list and make a second list for a second market.

What to write down for every answer

Keep one row per question, per system, per date. Save the full text of the answer or a screenshot. Then note whether your brand is named, how it is described, whether your site is linked and which pages, which other brands appear, and which outside sources the answer cites.

The accuracy column matters more than the mention column. An assistant that names you with an old address, a price you dropped or a service you stopped offering has given you a task. Mark every statement as right, wrong or missing. Wrong facts usually trace back to a page somewhere, and the sources listed in the answer show where to look first: an outdated directory, an old announcement, a profile nobody updated.

Because an answer is generated each time it is asked, run the questions that matter most more than once and write down the pattern, not a single lucky or unlucky reply. Note the date and the country or language you used, since those can change what comes back.

Free reports that show part of the picture

Google states that sites appearing in its AI features are counted in overall Search traffic, and that this traffic is reported in the Search Console Performance report under the Web search type. In practice that means no separate AI line to read. Search Console remains useful for another reason: the queries that already bring visitors are the best source for the buyer questions in your list.

Bing’s AI Performance report shows total citations, the average number of unique pages cited per day, a sample of grounding queries, citation counts for individual pages and a trend over time. Microsoft is explicit that it measures how often content is cited and says nothing about ranking, authority or placement inside an answer. Both reports describe your site’s side. Neither shows what ChatGPT or Perplexity said about you, so the manual list stays.

Monthly tracking and what to fix first

Repeat the whole list once a month, on a similar day, with the same wording, and put the new rows next to the old ones. Look at three things: mentions gained or lost, wrong statements corrected or newly appeared, and changes in the sources cited.

Fix in a sensible order. First, access: crawl rules and indexing errors. Second, facts at the source: your own pages and the business profiles. Google advises keeping Business Profile and Merchant Center information current. Third, missing answers: if no page of yours answers a question plainly, write it in text, since Google recommends keeping important content in text form and structured data consistent with the visible page. Google also notes that indexing and serving are not guaranteed, so treat every fix as a reason a mention becomes possible, not certain.

Do it yourself or order the check, and what to prepare

Doing it yourself is realistic. Twenty questions across four systems make eighty answers to read and record, which takes an afternoon the first time. The repeat every month is the tedious part, and it is where checks usually stop. If the monthly repeat is what you will skip, hand it over.

VITON13 Studio offers both steps as packages. The AI visibility check puts twenty buyer questions to ChatGPT, Gemini, Perplexity and Google AI Overviews, records how your brand appears in each answer and returns a report with the gaps and what causes them. Before you pay you see three of the twenty answers in full as a free draft. The check is a fixed $13 and takes two to three working days. Monthly AI visibility monitoring tracks brand mentions for a fixed list of questions, sends a short monthly report and fixes the issues it finds. It costs $33 a month and can be cancelled with 30 days’ notice, and the free draft is the list of questions we would track. Bring your site address, your markets and languages, and five questions real customers asked you.