AI Visibility Check: What ChatGPT and Gemini Say About Your Brand

To run an AI visibility check by hand and see what AI assistants say about your brand, open ChatGPT, Gemini, Perplexity and Google in clean sessions and ask ten questions a buyer would ask, without your brand name. Repeat each question three to five times. Log every answer in a spreadsheet: named, recommended, cited, wrong. Then count the share of answers that include you.

Step 1: Which questions should you ask AI assistants?

Do not start with "What do you know about [brand]?". That question only shows what an assistant says once it is handed your name. The useful test is whether your brand appears when nobody mentions it.

Write ten category questions in your customers' words, for example:

• "Best [product type] for [use case]"
• "[Competitor] alternatives for a small team"

Then add two or three brand questions ("How much does [brand] cost?") and count them separately: category questions show whether you get recommended, brand questions whether the facts are right.

Keep the wording fixed, or you cannot compare runs. A starter set by buying stage is in which prompts to track.

Step 2: Which AI assistants should you check?

Check the assistants your customers use, plus Google. Statcounter's AI chatbot chart for September 2026 gives ChatGPT 80.48% worldwide, Gemini 10.91%, Perplexity 5.85%, Microsoft Copilot 1.79% and Claude 0.95%.

In its press release of May 7, 2026, Statcounter describes the metric as each chatbot's share of referrals to websites, measured on more than 1 million sites. It counts visits sent to websites, not users or conversations. Grok is absent because it passes no referral data, the release says. Google AI Overviews are not on the chart either.

Assistants also differ in where answers come from. The table summarizes vendor documentation read on October 1, 2026.

When an answer carries source links, log them: they show which pages the assistant drew on.

Step 3: How do you get clean, unpersonalized answers?

Assistants can adapt answers to the person asking. To see what a new customer sees, take yourself out of the test.

• ChatGPT: use a logged-out private window, or a temporary chat set to Unpersonalized. OpenAI's temporary chat help page says that option uses neither memory nor custom instructions, while the Personalized option can use both.
• Gemini: use a temporary chat (it needs a signed-in personal Google account). Google's help page lists personalized responses among the things a temporary chat cannot give.
• Google AI Overviews: search in a private window, signed out. Google Search Help lists the countries and languages where overviews are available to all users.
• Every assistant: start a new chat for each run, so earlier answers do not steer later ones. Note your country and language: OpenAI's search help page says ChatGPT may estimate your location from your IP address.

A Profound study published September 9, 2026 ran clothing, furniture and credit card prompts on ChatGPT, Claude and Gemini over 14 days (August 11–24, 2026; 71,147 responses). Each prompt began with a one-line persona, such as an age or occupation. Answers for different personas shared about 1 in 4 recommended brands. Answers for the same persona shared about 2 in 5.

Two caveats: this is one vendor's data, and the persona was typed into the prompt, not stored in account memory. Still, the list varied even when the persona stayed the same.

Step 4: How do you track brand mentions in AI answers?

Log one row per answer. The rows below are illustrative, not real data.

• Separate named, recommended and cited. Named means the brand appears anywhere in the answer. Recommended means the assistant advises choosing you for the asker's case. Cited means your site is among the sources. Full definitions: what AI visibility means.
• Save the full answer text in a second sheet.
• Write down every factual error: a wrong price, a missing feature, a mix-up with a similarly named company.

Step 5: How do you read the results?

Count a share, not a position. For each assistant, divide the answers that name your brand by all answers to your category prompts.

In a SparkToro and Gumshoe study published January 28, 2026, 600 volunteers ran 12 prompts a combined 2,961 times on ChatGPT, Claude and Google's AI Overviews (AI Mode when no overview appeared). For ChatGPT and Google's AI, the chance that any two responses held the same list of brands was under 1 in 100. Claude was only slightly more consistent. The same list in the same order turned up about once in 1,000 runs. Author Rand Fishkin concluded that a visibility percentage across many repeated prompts is a reasonable metric and a ranking position is not.

One screenshot proves nothing either way.

A worked example (arithmetic, not data): ten prompts run five times give 50 answers per assistant. If your brand is in 15 of them, the mention rate is 30%, and the 95% confidence interval (Wilson method, treating answers as independent) runs from roughly 19% to 44%. That sample separates "never", "sometimes" and "most of the time". It cannot tell 30% from 40%.

A paper by Schulte, Bleeker and Kaufmann, posted on arXiv on April 8, 2026, ran eight German-language prompts in each of four Swiss industries on ChatGPT, Gemini, Google AI Mode and Perplexity: daily from January 24 to March 20, 2026, then up to ten times in a row on March 21–25. In the repeat runs, the standard error of a brand's detection rate fell below 0.10 at seven runs (a 95% interval of about ±16 percentage points). The authors recommend at least seven runs per prompt per day. That is one country and a few industries, so treat seven as a floor, not a law. More in why AI answers change every time.

Where does a manual AI visibility check stop working?

The manual method answers two questions well: do assistants know you exist, and are the facts right? It breaks down in four places.

• Volume. Ten prompts, four assistants and five runs make 200 answers to read and log.
• Sample size. SparkToro's Fishkin estimates it usually takes at least 60 to 100 runs of a prompt to learn an assistant's set of recommendations.
• Personalization and location. You test from one device and one country. A buyer elsewhere may get a different answer.
• Time. One check is a snapshot. The arXiv authors recommend a rolling window of two to four weeks.

Run your first AI visibility check

Start with the manual method: ten prompts and a spreadsheet. Or run the free one-off check on the Brandometer home page.

Brandometer automates the repetition. It sends each prompt to 9 AI engines repeatedly in clean sessions and reports the mention rate with confidence intervals, whether you were recommended or merely named, which competitors appeared instead and which sources were cited. It measures and suggests hypotheses; it cannot change what an assistant says. See the methodology.

To try it, create an account. New accounts get 200 bonus credits after email confirmation. There is no subscription: credits are prepaid, never expire, and you pay only when you measure.

FAQ

Is there a free AI visibility checker?

Yes. Ahrefs and Backlinko, among others, had free checkers online on October 1, 2026, and Brandometer has a free one-off check on the home page. Treat any free check as a snapshot of a few answers on one day, not a trend.

Why did ChatGPT mention us yesterday but not today?

That alone does not show a change. Assistants generate each answer anew: in the SparkToro and Gumshoe study (January 2026), two ChatGPT answers named the same list of brands less than 1 time in 100. Judge by the share across many runs, not by one reply.

Do I need to check every AI assistant?

No. Start with ChatGPT, Gemini and Google AI Overviews, then add assistants that your customers mention or that send you referral traffic.