Why ChatGPT does not name your business, and what the citation data says to change

Because the pages it reads when somebody asks your buying question do not name you. That answer sounds glib until it is counted. On 15 October 2026 we asked ChatGPT and Gemini 102 buying questions each in the United States, the United Kingdom and India, kept 597 usable readings, and recorded 10,652 citations across 2,708 distinct domains. 85.6 per cent of those citations went to a vendor's or an agency's own website. 487 of the 597 readings returned any source at all. Being named, then, is mostly a question of whether your own pages are in the set the engine reads for that question, and sometimes a question of whether it read anything at all.

Four reasons come out of that data, and they are ordered here by how much of the gap each one explains.

One: the answer is assembled from somebody else's own site

With 85.6 per cent of citations landing on a vendor's or agency's own website, the competitor named in the answer is usually named because its own page was the source. Not a directory, not a journalist, not a review profile: its own page, written to answer the question a buyer asked. Reviews and listings came to 0.8 per cent of the 10,652, and news and large publishers to 2.5 per cent, which puts the common advice to chase coverage and directories in proportion.

What this changes: the page that gets you named is one that answers a buyer question directly, in its first paragraph, with a figure a reader can check and a date on it. Not a brochure page about your company.

Two: on a share of questions, nothing was read

487 of the 597 readings returned any source, so 110 of them produced an answer without citing one. On those, the engine was working from what it already held rather than from the live web, and no amount of new publishing shows up in the answer that day. This is the most misread result in the set, because it looks like a content problem and is actually a timing one.

What this changes: run your check twice, once with web search on and once without, and record them as two different measurements. The two numbers answer two different questions, and mixing them is how people conclude their work did nothing.

Three: the number of pages you have is not the lever

This is our own result and it is the most useful thing in the run. Our site had 1,235 live pages on the day of the audit and was cited 3 times in 597 readings. In the same run, rank4ai.co.uk was cited in 39 readings and hawkorbit.com in 33, and neither is a large company. They win because their pages are specific to one market and one question, with the detail an engine can lift as an answer.

What this changes: stop counting pages and start counting questions covered. One page that answers a real buyer question with a dated number beats twenty that restate a category. If your current programme is being sold to you as volume, that is the conversation to reopen, and where this work sits beside ordinary search is set out in what AI SEO covers.

Four: the two engines do not behave the same way

Our 3 citations were all on ChatGPT: none in the United States, one in the United Kingdom, two in India. Gemini cited us zero times in every market. A brand can be present on one engine and absent on the other in the same market on the same day, and a single blended visibility figure hides exactly that. Markets diverge too, which is why 30 of the 102 questions in each market were written for that market rather than translated.

What this changes: measure per engine and per market, and fix per engine. All 3 of our citations came from one of the two engines, so a change that gets ChatGPT to read your page is evidence about ChatGPT and about nothing else until the same questions have been run on Gemini.

Working out which reason is yours

Take twenty questions your buyers actually ask. Put your brand name in none of them, because a question that names you returns a flattering answer and measures the question rather than the engine. Run each in a fresh conversation on both engines, and record four things per answer: the date, the engine, whether you were named, and every domain cited. Then read the domain column, which is where the diagnosis lives. If the same three competitor domains keep appearing, reason one is yours. If many answers cite nothing, reason two. If the cited pages are thin pages on small sites, reason three. If one engine names you and the other never does, reason four.

The full protocol, including the definitions of a mention, a recommendation and a citation, is written up at how to measure whether ChatGPT mentions your company, and the reason the blind rule is not optional is covered at why you must never name your own brand in a tracking question.

Questions about not being named in ChatGPT answers

Can I just ask ChatGPT why it does not mention my company?

Ask it, keep the answer as a lead, and verify it against the sources the answer cites. Engines describe their own behaviour unreliably: in an earlier capture of ours, one assistant said it had run a live search and then withdrew the claim when it was asked for the sources.

How many questions before the number means anything?

Forty or more gives you a rate you can compare month to month. Twenty gives you a baseline and a diagnosis, which is what this page is for. At twenty questions a single extra answer moves the figure by five points, so report the counts rather than the percentage.

We are cited in one country and not another. Why?

Because the source set differs by market. Our own 3 citations were one in the United Kingdom and two in India with none in the United States, on identical core questions, which is a sourcing difference rather than a quality one. Measure each market you sell in separately.

Does blocking AI crawlers explain it?

It explains it completely when it is true, so check that first: it takes ten minutes to read your robots rules and confirm the assistant user agents are allowed. Most sites we look at are not blocking anything, and the gap is in what the pages say.

How long after publishing does a page get cited?

Record the publication date and the first citation date for every page you write, because that interval is the only honest way to judge how quickly your own category refreshes, and it differs by category. Our own run is dated 15 October 2026 and is re-run every 60 days, which is the interval we would suggest you copy.

What we did with our own answer to this question

We published it. The run that produced these numbers also produced the worst result in it, 3 citations in 597 readings for the company selling the fix, and it is on our own benchmark page with the method and the question split. Our product exists to answer this exact question for a business: AI Knows Us puts the buyer questions to five named engines, ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and names the sources that decided each answer. The data on this page comes from two of them. The reason to say all of that on a page about why you are not named is that the diagnosis above is only worth anything if it was applied to us first.