It is a particular kind of unsettling. You ask ChatGPT to recommend a company in your sector and your city, and it names four businesses. None of them are you. One of them, you are fairly sure, is smaller than you.
This is now a common reason for a first conversation with us, and the explanation is rarely the one people expect. It is almost never that the AI "doesn't know you exist". It is that nothing it can read makes the case for naming you.
How AI Tools Actually Choose Who to Name
Language models do not hold a ranked directory of businesses. When someone asks for a recommendation, roughly two things happen.
First, there is what the model absorbed during training — a compressed, slightly out-of-date impression of who gets talked about in your sector. Your brand is in there only to the extent that other people wrote about you.
Second, and increasingly dominant, there is retrieval: the tool searches the live web, reads a handful of pages, and builds an answer from them. ChatGPT with browsing, Perplexity, Google's AI Overviews and Gemini all work broadly this way.
That second mechanism is the one you can influence, and it changes the question entirely. The question is not "does the model know me?" It is: when a tool searches for the answer to this question, does it land on a page that names me as one of the options?
Usually it does not — and here is why.
Reason One: Nothing Lists You
AI recommendations are heavily shaped by pages that already contain lists. When a model needs five agencies in Manchester, the cheapest route to a confident answer is a page that already names five agencies in Manchester.
Those pages are directories, industry roundups, "best of" articles, trade publications, awards shortlists and comparison posts. If your competitors appear on them and you do not, the model is not making a judgement about quality. It is reading the only list available.
This is the single most common cause, and it is also the least technical. The fix is unglamorous: get accurately listed where your sector gets listed. Not link farms — the real directories and publications your industry actually uses, with consistent details across all of them.
Reason Two: Your Own Site Never States It Plainly
Models extract claims. They are poor at inferring them.
A great many business websites never say, in plain declarative text, what the business does and who for. They say things like "we turn ambition into outcomes" above a hero image. A human reads the rest of the page and works it out. A retrieval system extracting a sentence to support a recommendation finds nothing it can quote.
The pages that get cited tend to contain unambiguous statements: what the service is, where it is delivered, who it suits, what it costs or how pricing works, and what makes it different in concrete terms. Not keyword stuffing — just saying the thing.
Reason Three: The Content Is There, But Not Extractable
This one is technical and surprisingly common.
If your key content renders only after JavaScript executes, some crawlers will see an empty page. If your FAQs are in an accordion that loads content on click rather than hiding content that is already in the HTML, the answers may not exist in the document at all. If your service details live inside an image or an embedded PDF, they are invisible to text retrieval.
Three checks worth running:
- View the page source — not the inspector, the raw source — and search for a sentence you can see on screen. If it isn't there, it rendered client-side.
- Confirm your FAQ answers are in the HTML, collapsed visually rather than absent.
- Check whether your
robots.txtblocks the AI crawlers. Some businesses block them deliberately, which is a legitimate choice — but you cannot block a crawler and then wonder why its model never names you.
That last one catches people out. If you have asked an agency to block AI crawlers to protect your content, you have also opted out of AI recommendations. Both positions are defensible; holding both at once is not. Our LLM audit checks exactly this sort of contradiction.
Reason Four: No Third-Party Corroboration
Models hedge towards consensus. A claim that appears on one website is a marketing statement; the same claim appearing on your site, a trade publication, a review platform and a directory becomes a fact as far as a retrieval system is concerned.
This is why businesses with modest websites and strong external presence often outperform better-looking competitors in AI answers. Reviews, mentions, case studies published elsewhere, speaking appearances, and being quoted in industry press all serve as corroboration.
It is also why reviews matter more than their star rating suggests. A model summarising "is this company any good" is reading the text of reviews, not just averaging the scores.
Reason Five: You're Asking a Question Nobody Writes About
Sometimes the brand is fine and the query is the problem. If you ask for "the best X in Y" and nobody has ever published a page comparing X in Y, the model will improvise from whatever is adjacent — often national brands or whoever happens to rank for something similar.
Before concluding you have a visibility problem, try the question several ways: with and without location, in the words a customer would use, and as a problem statement rather than a vendor request ("who can help me with…"). You will often find you appear for some phrasings and not others, which is a far more actionable finding than a blanket absence. We cover how to test this systematically in tracking brand visibility in ChatGPT.
What Actually Changes the Answer
In rough order of how quickly it tends to show results:
Fix extraction first. If your content is not in the HTML, nothing else you do will register. This is usually days of work, not months.
Say the plain thing. Rewrite your main service pages so the core claims are stated in quotable sentences. Add a genuine FAQ section answering the questions people actually ask before buying, with the answers visible on the page.
Get on the lists. Audit where your competitors are named and you are not. Fix your presence on the directories and publications that matter in your sector, with consistent details.
Build corroboration. Reviews, press mentions, published case studies, anywhere your claims can be verified independently.
Then monitor. Ask the questions that matter to your business, repeatedly, and record what comes back. AI answers vary between sessions, so one check tells you very little and twenty tell you a pattern.
The uncomfortable truth is that most of this is recognisable as good SEO and good PR. The mechanism is new; the work mostly is not. What has changed is that vagueness now has a direct cost — a human visitor will forgive a woolly homepage and read on, and a retrieval system simply will not quote it.
If you want to know where you currently stand, our free LLM audit tests how AI tools describe your business and where the gaps are, and our AI SEO and generative engine optimisation services cover the work itself. Our guide to what an LLM audit involves explains what that assessment actually looks at.
Frequently Asked Questions
Why doesn't my brand appear when people ask ChatGPT for recommendations?
Usually because nothing the tool can read names you as an option. AI recommendations lean heavily on pages that already list businesses in your sector, so if directories and roundups name your competitors and not you, the model is reading the only list available rather than judging quality.
Can I pay to appear in ChatGPT recommendations?
Not directly — there is no advertising slot in an AI recommendation in the way there is in search results. What you can influence is the evidence the tool reads: your own site's clarity and extractability, your presence on the lists and publications it draws from, and independent corroboration like reviews and press coverage.
Does blocking AI crawlers stop my business being recommended?
Yes, for the tools you block. Blocking AI crawlers is a legitimate decision for protecting content, but it means opting out of being cited by those systems. If you want AI visibility, check your robots.txt before anything else — some sites block the crawlers without realising it.
How long does it take to start appearing in AI answers?
Extraction fixes can register within days or weeks of a recrawl, because the content was already there and simply unreadable. Getting listed on the sources models draw from, and building third-party corroboration, works on a slower timescale — closer to the pace of PR and link building than technical SEO.
Is appearing in ChatGPT different from ranking on Google?
They overlap more than most people expect, because retrieval-based AI tools search the web and read the results. The main differences are that AI answers favour content stating claims plainly enough to quote, and that being named on third-party lists matters more than it does for a conventional ranking.