Does ChatGPT mention my business?

You can find out in about ten minutes, and it is worth doing, because a growing share of the questions your buyers ask now get answered before anyone reaches a list of links. Open ChatGPT, Perplexity and Google's AI results, ask the questions a customer would ask about your category, and write down whether you are named and whether the link is right.

Why does this matter now?

Pew Research tracked the actual browsing of 900 US adults through March 2025. In that month 58 percent of them ran at least one Google search that produced an AI summary, and 18 percent of all searches in the study generated one. This is not a future problem.

The effect on clicks is the part that should get your attention. People clicked a search result in 8 percent of visits where a summary appeared, against 15 percent where it did not. They clicked a link inside the summary itself in 1 percent of visits. And they were more likely to stop browsing entirely: 26 percent of pages with a summary ended the session, against 16 percent without.

Read those three together and the conclusion is uncomfortable. Being cited in the answer is worth far less traffic than ranking used to be. It is still worth having, because being absent from the answer is worth nothing at all, and because the person who does click has already been told you are credible.

How do I check properly?

Ask the way a buyer asks, not the way a marketer does. A buyer types a situation, with a budget or a constraint in it. A marketer types a keyword. The two produce completely different answers and only one of them tells you anything.

Write down ten to fifteen prompts across the whole journey, from first discovery through to a direct comparison. Keep branded prompts in a separate list from unbranded ones, because your own name drags you into answers you did not earn and will flatter you into thinking the problem is solved.

Run each prompt in a fresh session. Assistants carry context, and a conversation that has already discussed you will keep discussing you.

Record two things per prompt, not one: whether you were named, and whether the link attached to your name was correct. A citation pointing at your homepage instead of the page that answered the question is a half win worth fixing.

What prompts should I test?

Most people freeze at this step, so here is the shape. Say you sell bookkeeping to restaurants. Your unbranded list runs from the vaguest version of the problem to the most specific, because different stages of that journey pull completely different sources.

At the top: who handles the books for a small restaurant, and what does a restaurant bookkeeper actually do. In the middle: how much should a restaurant pay for bookkeeping, and whether to hire someone in house or outsource it. At the bottom: the best bookkeeping service for restaurants, and how two named competitors compare.

Then a separate branded list: what your company does, whether it is any good, and how it compares to a named rival. Kept apart, because the branded answers will look flattering and mean almost nothing.

Ten to fifteen prompts is enough to see a pattern and few enough that you will actually rerun them next month. A list of forty gets built once and abandoned.

Why do assistants cite some sources and not others?

An assistant answering a broad question does not run one search. It breaks the question into smaller ones and gathers sources for each fragment. Pages that surface across several fragments get cited. A page that matches the original question perfectly but appears only once often does not.

That inverts a piece of advice you have probably been given. One long article covering a topic loosely tends to lose to a set of pages that each answer one question completely, because the set intersects more fragments than the single page does.

It also rewards answering early. If your answer arrives in paragraph nine, after the context and the history and the anecdote, a retrieval system may never reach it, and a reader skimming the citation certainly will not.

What do I fix if a competitor shows up instead?

Start by reading the pages that were actually cited. Not to copy them, but to see their shape. They usually answer the exact question in the first few lines, use plain headings phrased the way people ask, and cover the neighbouring questions on the same page rather than spreading them across five.

Then move your own answer to the top, turn your headings into questions, and cover the nearby questions instead of repeating the main one across several pages. None of that is new writing. It is rearranging what you already published, which is why it is the cheapest work available to you.

Then check they can reach you at all. A robots file that blocks AI crawlers ends the conversation before it starts, and plenty of sites block them by accident rather than by decision. It takes five minutes to check and it gates everything else on this list.

What if there is nothing to cite?

Sometimes the honest answer is that you are not cited because there is nothing to cite. Ahrefs looked at roughly 14 billion pages and found 96.55 percent get no traffic from Google at all, with no search demand, no links and intent mismatch as the three causes. An assistant is drawing from a narrower pool than a search engine, not a wider one. If you are invisible in one you are usually invisible in both, and the fix is pages worth quoting rather than formatting.

How do I track this over time?

A single check is a snapshot, and these answers move week to week. Keep the prompt list, run it on a schedule, and record which prompts name you rather than trying to score the whole thing at once.

What you are looking for is the count rising over months, on unbranded prompts. Any single answer is noise. The trend is the only thing here that means anything.

Common questions

Why do I get a different answer every time I ask?

Because these systems sample rather than look up, and because they retrieve fresh sources per question. That is why a single check is close to meaningless. Run the same prompt list on a schedule and count how many prompts name you across the set, since the trend is stable even though any one answer is not.

Does asking about my own business make it more likely to be mentioned?

No. Asking does not teach the model anything about you, and a conversation that has already discussed you will keep discussing you within that session only. This is why branded and unbranded prompts have to be kept in separate lists, and why every prompt should be run in a fresh session.

How often should I rerun the prompt list?

Monthly is enough for most small teams, and it is a cadence you will actually keep. Weekly produces noise you will read too much into. Keep the same prompts every time, because changing the list is the fastest way to lose the only thing you are measuring, which is direction over months.

Key takeaways

You can check in about ten minutes by asking the questions a buyer would ask and recording whether you are named and whether the link is right.

Pew Research found 58 percent of people saw at least one AI summary in a month, and 18 percent of all searches produced one.

Keep branded prompts separate from unbranded ones, because your own name pulls you into answers you did not earn.

Assistants break a question into fragments, so a set of pages each answering one question completely tends to beat one long page covering the topic loosely.

Related reading

The formatting side of this has a name and a short answer: What is AEO and does my small team need it?. If you rank but nobody clicks, that is a different problem with a different fix: My blog ranks but no one clicks. Why?. And on whether the file everyone is talking about actually helps, see Do I need an llms.txt file?.

Sources

Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 2025. Browsing data from 900 US adults, March 2025.

Ahrefs, 96.55 percent of content gets no traffic from Google, December 2023.