How to check if ChatGPT recommends your brand

Why one test is not enough

Most people ask ChatGPT "what are the best [category] companies?" once. If their brand is on the list, they relax; if it is not, they worry. Both reactions are premature, because AI assistants do not give fixed answers. In a study by SparkToro and Gumshoe with 600 volunteers and 2,961 runs, there was less than a 1-in-100 chance that ChatGPT or Google's AI returned the same list of brands twice for the same prompt. The brands, their order and the length of the list all changed from run to run. The study is not peer reviewed, but what it describes will be familiar to anyone who has checked by hand.

So the useful question is not "what position am I in on ChatGPT?" It is "how often am I mentioned for the questions my customers ask, and how often are my competitors mentioned?" We explain the thinking behind this in What is generative engine optimization?

ChatGPT is a sensible place to start. OpenAI's chatbot reached 900 million weekly active users in February 2026, according to TechCrunch. If you sell in Turkey, Statcounter's June 2026 data puts ChatGPT at 76.2% of AI chatbot usage share there, with Gemini second at 15.89%. That is a monthly snapshot of usage share, not a user count, but it shows where to begin.

A step-by-step manual check

1. Write the questions your customers would ask

Work with conversational questions, not a keyword list. People do not type "dental clinic Istanbul" into an assistant; they ask "Can you recommend a reliable dental clinic in Istanbul for implants?" It helps to split your list into three groups:

• Category questions: "Which accounting software would you recommend for a small online shop?" Your brand is not named and the assistant builds its recommendation from scratch. This is the most valuable group.
• Problem questions: "What can I use to stop typing invoices in by hand?" The customer has a need but may not know the category name yet.
• Comparison questions: "[Competitor A] or [Competitor B]?" or "What is a good alternative to [Competitor A]?" These show whether you come up when competitors are named.

Do not put your own brand name in the questions. "Is [my brand] any good?" shows whether the assistant knows you, not whether it recommends you. You can keep such questions as a separate group, but leave them out of your visibility numbers. Starting with 10 to 20 questions keeps the work manageable while giving you a picture that does not hinge on a single prompt.

2. Ask in a clean session

If you ask from your own account, in a session where you have already talked about your brand, you cannot be sure the answer is not shaped by your history. To get closer to what a customer sees:

• Open a private (incognito) browser window and, if possible, ask without logging in.
• If you have to log in, use a temporary chat that does not use memory or personalisation.
• Start a new chat for every question. Questions asked in a row in the same chat influence each other.
• If several people on the team run checks, make sure they all follow the same rules.

3. Match the language and location

Test in the language your customers use. Do not assume the answer to an English question matches the answer to the same question in Turkish, German or Spanish; the brands and sources that come up can differ by language. If you are a local business, put the city or district in the question ("in Kadıköy", "in Manchester"). Because your location may affect the answer, checking from the same place and the same kind of device each time keeps results comparable.

4. Ask each question several times, on different days

Given how much answers vary, we suggest asking each question at least three times, spread over different days if you can. Appearing or not appearing in one run proves little. Showing up in all three runs is a strong signal; missing from all three may point to a real gap.

5. Record every run the same way

A simple spreadsheet is enough. What matters is recording each run with the same fields, so you can compare weeks later:

Do not skip the sources column. If the assistant keeps pointing to the same comparison site or forum when it recommends a competitor, that is the most concrete clue about where to focus. Patterns differ by platform: Profound's analysis of citations from August 2024 to June 2025 found that Wikipedia accounted for 47.9% of citations among ChatGPT's top 10 sources. In an Ahrefs study of 26,283 cited URLs across 750 prompts, recently updated "best X" list articles were the most-cited page type, at 43.8%.

6. Sum it up in two numbers

1. Visibility rate: answers that mention your brand ÷ total answers. For example, 10 questions asked three times each gives you 30 answers; if your brand appears in 6 of them, your rate is 20%.
2. Share of voice: your mentions ÷ the total mentions of you and the competitors you track. It tells you more than the absolute rate, because it compares you directly with your competitors.

Also make a gap list: the questions where competitors appear and you never do. This list shows most clearly where to start work.

The limits of checking by hand

The method above is free and good enough for a first impression. When you try to turn it into regular measurement, you run into a few problems:

• The volume grows fast. 15 questions, three runs and one engine already means 45 answers. Do the same for Gemini, Perplexity, Claude and Google AI Overviews and you are reading and logging hundreds of answers each period.
• Engines do not agree. Semrush's 2026 AI Visibility Index analysed 126 million US prompts and more than 1,200 brands; only 36 of them stayed in the top 100 most-mentioned on every platform in every month. Your result in ChatGPT does not stand in for your result in Gemini.
• Consistency is hard. When different people check on different days with different settings, results stop being comparable. It is also easy to miss a spelling variant of your brand or a competitor buried in the middle of a long answer.
• Built-in reports do not show competitors. The generative AI performance reports that Google Search Console opened to a subset of sites in June 2026 show impressions only, with no clicks, click-through rate or queries. They show neither the answer text nor competitors, so they cannot tell you who is being recommended instead of you.

Automating the check

A free quick check

If you want to see where you stand before building a spreadsheet, try the free AI visibility check. You enter your brand, website and category, and optionally your city. It asks an AI engine 3 questions your customers might ask and, in about 30 seconds, shows whether your brand comes up, which brands are recommended instead and which sources are cited. No sign-up or card is needed. It is a snapshot from a single engine, so it does not replace the repeated measurement described above, but it is a good starting point.

Regular, multi-engine tracking

KaiRank automates this kind of check. Every week it asks your list of questions across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews (how many engines are included depends on the plan) and reports whether your brand is mentioned, its position, your share of voice against competitors, the sources cited and the questions where competitors appear and you do not. Scans run weekly, not daily, so the aim is to follow the trend over several weeks rather than every daily swing; a single weekly scan is still a snapshot and should be read that way. Plans and limits are on the pricing page.

Quick checklist

• 10 to 20 questions written the way customers ask them (category, problem, comparison)
• Your own brand name is not in the questions
• A private window or a session free of your history; every question in a new chat
• Questions in your customers' language, with the city or district if you are a local business
• Each question asked at least three times, on different days
• Every run in the same table: date, engine, question, mentioned or not, position, competitors, sources
• The result: visibility rate, share of voice and a gap list
• OAI-SearchBot is not blocked in robots.txt
• The same measurement repeated at regular intervals, read as a trend