How do I know if ChatGPT mentions my brand?
Ask it yourself, in a fresh chat with no memory, using the question a buyer would ask rather than your company name. Then repeat it in Claude, Gemini and Perplexity, because they disagree with each other. Record the model, the date and the answer word for word.
Published · Last updated
By Stiven Hernandez, founder, UPAI · Updated 2026-09-25
Key takeaways
The short version of this page, in the order the sections below develop it. Each point is stated here in full so it can be read without the rest of the page, and the section that argues it gives the measurement and the source behind it.
- Your Mention Rate: the share of your buyers' real questions where an assistant names your company.
- Because they do not agree.
- Five steps, about ten minutes in total.
- Your company named in the answer the reader sees.
Only 67% of the S&P 100 companies that any assistant named were named by all four. JPMorgan Chase appeared in Gemini and in none of the others; IBM appeared in ChatGPT alone.
Our own measurement: 398 usable answers across four assistants, eight industries, September 2026. See the study
What exactly are you measuring?
Your Mention Rate: the share of your buyers' real questions where an assistant names your company. It counts mentions, not traffic, and it is the one number a analytics report cannot give you. Define it before you measure it, or you will compare two different things next month. A Mention Rate needs three fixed parts: the question list, the assistants asked, and the number of runs per question. Change any of them and the series breaks.
Why does one assistant not tell you enough?
One assistant does not tell you enough because the assistants do not agree. We asked four assistants the same unbranded buying questions about the S&P 100. Of the companies any of them named, only 67% were named by all four. Seven of the largest public companies in the United States were visible in exactly one assistant and invisible in the other three. A company measuring itself in ChatGPT alone would conclude the opposite of a company measuring in Gemini.
This is why a single screenshot proves nothing, in either direction. Absence in one place is not absence; presence in one place is not coverage.
How do you check it properly?
Checking it properly takes five steps and about ten minutes in total. Each one removes something that would make the answer about you rather than about your category: memory that recalls your earlier questions, a prompt containing your own name, one assistant standing in for four, and a single run standing in for a pattern. Do them in order and what comes back is a reading you can compare against the same reading next month, which is the only version worth recording.
- 01
Open a fresh conversation. Memory, custom instructions and earlier chats change what comes back. A logged-out window is cleanest.
- 02
Ask the buying question, not your name. Ask what a buyer types — 'best X for Y', 'who should I hire for Z'. Asking about your company guarantees it appears.
- 03
Ask all four. ChatGPT, Claude, Gemini and Perplexity. Perplexity browses the live web; the others answer largely from what they have learned.
- 04
Repeat each question. These systems are stochastic. The same question returns different companies on different runs, so a single run is an anecdote.
- 05
Write down the model and the date. Answers drift as the sources drift. A result without a model ID and a date cannot be compared to anything later.
What counts as being mentioned?
Your company named in the answer the reader actually sees. Being used as a source without being named is a different outcome and more common than most teams expect, so keep the two apart when you record results. A citation with no mention means the model read your page and did not think the company behind it was worth naming, which is a different problem with a different fix. Record both columns from the first run: reconstructing the distinction later, from screenshots, is not possible.
| Column | Why it matters | Example |
|---|---|---|
| Question | The unit that moves; changing wording breaks the series | Best inventory software for mid-market retail |
| Assistant and model ID | They disagree, and each updates separately | gemini-2.5-flash-lite |
| Date | Answers drift as the sources drift | 2026-10-16 |
| Named or not | Your Mention Rate for that question | Not named |
| Sources cited | Where the answer came from, and what to fix | g2.com, reddit.com |
Why will your analytics never show this?
Your analytics will never show this because almost nobody clicks. Pew Research Center metered what 900 US adults actually did: when an AI summary appeared, people clicked a link inside it in just 1% of visits. Across 68,879 Google searches, a result was clicked 8% of the time when a summary appeared, against 15% when none did.
One percent means the reader takes the names and leaves. No referrer, no session, no line in your reporting. A traffic report cannot show you a buyer who never became a visit, which is why the thing worth counting is mentions.
"Google users who encountered an AI summary also rarely clicked on a link in the summary itself. This occurred in just 1% of all visits."
Questions people ask next
These are the follow-ups we are asked most often once the answer above has landed. The same rule applies to each: where a figure appears, its source and year sit in the same sentence, and where we do not know something we say so rather than estimate it.
- Is there a tool that does this for me?
- Several, priced from roughly $29 to $499 a month. They earn their keep once you know what you are measuring and want it tracked over time. Before that, four browser tabs and ten minutes tell you the same thing.
- How often should I check?
- Answers move when the sources move, not on a schedule. Weekly is enough to see a real change. Daily mostly measures the models' own randomness and will have you reacting to noise.
- Does asking about my own company work?
- No. Ask an assistant about your company and it will talk about your company. The only question that tells you anything is the one a buyer asks, which never contains your name.
Related
Other questions in this library that cover the ground next to this one. Each is answered the same way — the direct answer first, then the measurement behind it — and each carries its own sources, so you can check any of them on its own page without taking this one on trust.
Sources
Every figure on this page comes from one of the sources below, each linked so you can open it and check the number in its original context rather than against our summary. Where a figure is our own measurement, its method and sample size are stated with it.