- Write 10 to 20 questions a buyer would really ask, such as “best bookkeeping service for a small bakery in Leeds”, not your brand name.
- Ask each one in ChatGPT, Perplexity, Gemini, Claude and Copilot, in a clean session: logged out or in a temporary chat, with memory and personalisation off.
- For every answer, record four things: were you named, in what position, which competitors were named, and which sites were cited.
- Repeat the same questions every week. AI answers vary from run to run, so one answer is an anecdote and a weekly series is a trend.
- Note whether the answer searched the web. A search-backed answer reflects your pages today; an answer from model memory changes slowly.
The short answer: ask, the same way, on a schedule
The only reliable way to know whether ChatGPT mentions your brand is to ask it the questions your customers ask and keep the answers. No AI assistant gives site owners a report of how often it names them, and Google reports AI Overviews and AI Mode traffic inside the normal Search Console totals rather than as a separate list (Google Search Central). So the method is simple and a little tedious: a fixed set of questions, asked in a clean session, logged in a sheet, repeated weekly.
Searching for your own brand name (“what is Acme Bookkeeping?”) tells you very little. Customers who already know your name are not the ones you are trying to reach. The useful test is whether you appear when someone asks for a recommendation in your category without naming you.
Step by step: the manual method
-
Write 10 to 20 buyer questions
Use the words a customer would type, not your marketing language. Mix three kinds: category questions (“best CRM for a 5-person agency”), problem questions (“how do I stop missing client follow-ups?”) and comparison questions (“alternatives to [the market leader] for small teams”). Add your city or region if you sell locally. Your sales calls, support inbox and Search Console queries are the best source of wording.
-
Pick the assistants your buyers use
For most businesses that means ChatGPT, Perplexity, Google Gemini, Claude and Microsoft Copilot. If your customers are mostly on one of them, start there; five assistants times twenty questions is a hundred answers a week, which is a lot to read by hand.
-
Use a clean session every time
Personal history changes answers. Open a private browser window, use the assistant logged out where it allows that, or use its temporary or incognito chat mode, and turn off memory and personalisation where you can. Do not mention your company earlier in the chat. The goal is to see roughly what a stranger sees.
-
Paste the question exactly as written
Copy it from your sheet so the wording never drifts. Ask once per assistant per week, and if you have time, ask a second time in a fresh chat to see how much the answer moves.
-
Log the answer in a sheet
Record the fields in the table below, and save the full answer text or a screenshot. You will want the exact wording later, especially if the assistant describes you incorrectly.
-
Repeat weekly, on the same day
Keep the question list fixed for at least a couple of months. If you change the questions, start a new tab, otherwise you cannot tell whether a change came from the assistant or from you.
A tracking sheet you can copy
| Column | What to write | Example |
|---|---|---|
| Date | The day you asked | 2026-10-27 |
| Assistant | Which product, and the model or mode if it shows one | Perplexity |
| Searched the web? | Did the answer show sources or a “searching” step? | Yes |
| Question | Copied exactly from your list | best bookkeeping service for a small bakery |
| Mentioned | Yes or no: your brand or domain named or cited | Yes |
| Position | Your order among the brands named (1 = first) | 3 |
| Competitors named | Every other brand in the answer, in order | Brand A, Brand B |
| Cited sites | Domains the answer linked to or listed as sources | a review site, a trade directory |
| Accuracy note | Anything wrong or out of date about you | Old price; says no weekend support |
Web search vs model memory: why it matters
An AI answer comes from one of two places, and you should read them differently. Some answers are written from what the model learned in training. Others search the web for the question and summarise the pages they find, usually with links. Which one you got changes what a mention, or a missing mention, tells you.
| Assistant | How it can use the live web | What a mention tells you |
|---|---|---|
| Perplexity | Searches Searches the web in real time and adds numbered citations to each answer | Whether your pages, or pages about you, are strong enough to be picked today |
| ChatGPT | Sometimes Its search features use OpenAI's search crawler, OAI-SearchBot; other answers come from the model | With sources: current web evidence. Without: what the model already “knows” |
| Gemini | Sometimes Can be grounded with content from the Google Search index at the time of the question | The same split as ChatGPT |
| Claude | Sometimes Uses web search when enabled or when it judges it helpful, with citations | The same split as ChatGPT |
| Copilot | Sometimes Microsoft's assistant, formerly Bing Chat, can include and link to pages Bing has crawled; Bing's page-level tags control that use | With links: current web evidence from Bing. Without: model memory |
The practical rule: if an answer shows sources, record the cited domains, because they are the pages the assistant trusted for that question. If an answer shows no sources, treat it as a measure of brand memory. That changes slowly, over model updates, and is shaped by how widely and consistently your business is described across the web.
Why do answers change from one day to the next? AI assistants do not return a fixed result. The same question can produce a different list on a second try, and results can also differ with location, language, account history, the model version and whether a search ran. That is normal. It is exactly why a weekly series beats a single screenshot.
How to read your results
After four or more weeks you have enough rows to see patterns. Look at four numbers per assistant, and keep them separate rather than blending everything into one score.
- Mention rate. Answers that named you divided by answers you collected, per assistant. Count only your fixed question set.
- Typical position. Being named fifth in a list of eight is very different from being named first. Track the median, not the best week.
- Who replaces you. List the competitors that appear when you do not. If the same two names keep showing up, study what is written about them.
- Which sites are cited. Count how often each domain appears as a source. A short list of review sites, directories and comparison articles usually dominates a category.
Also read the accuracy notes. An assistant that names you with the wrong price, a closed location or a discontinued product is a problem worth fixing before you chase more mentions.
What to fix when you are missing
Match what you see to the likely cause. None of these are tricks; they are the same things that help people find and trust you in normal search.
| What you see | Likely cause | What to do |
|---|---|---|
| Missing in search-backed answers only | The assistant cannot reach your pages, or you have no page that answers the question | Check robots.txt allows the search crawlers (see our AI crawler guide), then publish a page that answers the question directly |
| Missing everywhere | Little independent coverage of your business in your category | Get accurately listed on the directories and review sites the answers cite, and earn mentions in genuine comparisons |
| Named, but low in the list | Competitors are described more often, or more clearly | State plainly who you serve, what you cost and how you differ, on your own site and in your profiles |
| Named with wrong facts | Old or conflicting information about you online | Update your site, profiles and directory listings so the same facts appear everywhere |
| Only in answers without sources | Model memory knows you, but your current pages are not picked | Improve the page for that question: answer near the top, prices and limits in text, a comparison table, sources |
Do not try to game it. Google's own guidance says seeking inauthentic “mentions” across the web is not as helpful as it might seem, and that you do not need to rewrite content just for AI systems (Google Search Central). Planting your name in forums or buying placement in “best of” lists also costs you trust with the people reading them.
For Google specifically, there is nothing extra to submit: Google says a page must be indexed and eligible to show with a snippet in Search, and that there are no additional technical requirements to appear in AI Overviews or AI Mode. The broader approach is in our generative engine optimization guide.
When to automate the checks
The manual method works well for a handful of questions. It gets hard to keep up once you track more than about ten questions across several assistants, need the same clean conditions every week, or want to compare competitors over months. That is when a scheduled tool earns its place.
MrKeyword's AI Visibility tracker does the same job as the sheet above: it asks ChatGPT, Gemini, Claude and Perplexity your questions every week and records whether you were named, your position, the competitors named and the domains cited, with the sentence that mentioned you. It uses the engines' APIs, so ChatGPT, Gemini and Claude answers there measure model memory and Perplexity measures live web evidence. It does not see what a particular person was shown, does not track Copilot, and cannot make an assistant recommend you.
Track three questions freeThe Free plan checks 3 AI visibility questions every week across every connected engine. No card needed.
Start freeQuestions and answers
Check if ChatGPT mentions your brand questions
How do I know if ChatGPT recommends my business?
Ask ChatGPT the questions a customer would ask, without naming your business, in a temporary or logged-out chat. Record whether you were named, in what position, and which sources were cited. Repeat the same questions weekly, because single answers vary.
Why does ChatGPT give a different answer every time?
AI assistants generate each answer fresh, so lists and wording change between runs. Answers can also differ by location, account history, model version and whether a web search ran. Track a fixed question set over several weeks to see the real trend.
Should I search for my own brand name?
It is worth doing once to check the facts an assistant gives about you. But for visibility, ask category and problem questions that do not name you, because that is how new customers discover businesses.
Does blocking AI crawlers stop me appearing in ChatGPT?
Blocking OpenAI's search crawler, OAI-SearchBot, stops your pages being shown in ChatGPT search answers, according to OpenAI. Blocking GPTBot only affects training. Check your robots.txt before assuming an assistant is ignoring you.
How often should I check AI answers?
Weekly is a good rhythm. It is frequent enough to notice a change after you update a page, and slow enough that you can keep the conditions the same by hand.
Can I track AI mentions for free?
Yes. The manual method in this guide costs nothing but time. If you want it automated, MrKeyword's Free plan tracks 3 questions every week across ChatGPT, Gemini, Claude and Perplexity.
Sources
- How does Perplexity work? — Perplexity Help Center, on real-time web search and numbered citations
- Overview of OpenAI crawlers — OpenAI, on OAI-SearchBot and ChatGPT's search features
- Enabling and using web search — Claude Help Center, on when Claude searches and its citations
- Google's common crawlers: Google-Extended — Google, on grounding Gemini Apps with content from the Google Search index
- Our vision to bring Microsoft Copilot to everyone — Microsoft Bing blog, on Bing Chat becoming Copilot
- New options to control usage of content in Bing Chat — Bing Webmaster blog, on the NOCACHE and NOARCHIVE tags
- AI features and your website — Google Search Central, on AI Overviews, AI Mode and Search Console reporting
- Optimizing for generative AI features on Google Search — Google Search Central, on tactics that do not help
