Why you rank on Google but AI assistants never name you
Ranking on Google and being named by ChatGPT are different outcomes with different inputs. Five causes account for most cases: the assistant cites directories and roundups you are absent from; your ranking pages are built to convert rather than to be quoted; AI crawlers are blocked from your site; third-party descriptions of you are thin or contradictory; or you describe your service in words nobody actually asks about.
Two different competitions
A search results page is a shortlist of links. An assistant's answer is a sentence naming two or three businesses. Those are not the same contest, and doing well in one does not enter you into the other.
The reason sits in the mechanism. When an assistant answers a buying question it usually searches, pulls a small number of pages, and writes from what those pages say. So the names in the answer come from the retrieved documents, not from a ranking. How assistants choose who to name covers that pipeline in full; this article is about diagnosing why you specifically are missing from it.
Five causes account for nearly every case we see. They are not mutually exclusive, and most businesses have two or three at once. Read each one and decide honestly whether it describes you.
Cause 1: the cited sources are pages you are not on
This is the big one, and for most businesses it is sufficient on its own.
Ask an assistant which accountants serve small businesses in your city. Then look at what it cited. Very often it is not individual firms' websites at all: it is a directory, a review platform, a professional body's member search, and one or two "best accountants in X" roundups published by local media or an industry blog. Those documents contain the names. Your own ranking page was never in the room.
The test is simple and takes ten minutes. Run the question, list the sources, open each one, and search it for your business name. If you are absent from all of them, you have found your problem and nothing else on this list matters until it is fixed.
Fixing it is unglamorous: get listed, get reviewed, and get included in the roundups that are actually cited rather than the hundred that are not. Reviews, directories and third-party signals covers which ones are worth the effort and which are noise.
Cause 2: your ranking pages are conversion-shaped
Suppose you clear the first hurdle and your page is retrieved. It still has to give a model something to quote.
A homepage built to convert typically opens with a large image, a short emotive headline such as "Craftsmanship you can trust", three icons, a testimonial slider and a contact form. Strip out the markup and you may be left with forty words, none of which state what the business does or where it does it. There is nothing to extract.
Compare that with a page that says, in ordinary prose: "We are a family-run joinery business in Mansfield, Nottinghamshire, making bespoke staircases, fitted wardrobes and sash windows for homes across Nottinghamshire and north Derbyshire. We have traded since 2004." That sentence can be lifted whole into an answer. The first page cannot.
The same problem affects service pages written as sales letters. Benefit language is not descriptive language, and a model summarising a market needs description. Making your pages quotable is mostly a matter of adding plain statements, not of removing the persuasive ones.
Cause 3: the crawlers cannot get in
Search engine crawlers and AI crawlers are not the same set of user agents, and a site can be perfectly open to one while closed to the other.
Three things cause this in practice. A robots.txt file copied from a template or a plugin that disallows agents the owner has never heard of. A bot-protection or firewall layer that challenges unfamiliar user agents and returns a block page instead of content. And server-side rules added during a scraping scare and never revisited.
The symptom is invisible, which is what makes this cause so durable. The site loads fine in your browser, Google still ranks you, and nothing in your analytics suggests a problem. The only reliable way to know is to check your robots.txt deliberately and look at server logs for the named AI agents. AI crawlers and your robots.txt lists the agents worth checking and the difference between the ones that fetch pages for an answer and the ones that collect training data.
Worth being precise here: unblocking a crawler removes an obstacle, it does not create demand. Plenty of sites are fully accessible and still never named, because causes one, four and five are untouched. Treat access as a prerequisite rather than a strategy, and be sceptical of anyone selling a robots.txt edit as an AI visibility service.
Cause 4: thin or contradictory third-party corroboration
A model producing a recommendation is effectively weighing evidence from several publishers at once. One source saying something is a claim. Several independent sources saying the same thing is a description it can repeat with confidence.
Thin corroboration looks like this: a good website, a Google Business Profile, and almost nothing else. No reviews outside one platform, no trade body listing, no local press, no supplier or partner pages mentioning you. There is only your own account of yourself.
Contradictory corroboration is worse and much more common than people realise. The business trades as one name but is listed under its registered company name elsewhere. An old address survives on three directories. One listing says "plumbing and heating", the next says "bathroom installation", a third says "property maintenance". Each is true. In combination they describe three different businesses, and none of them strongly.
The fix is tedious and effective: pick one name, one address, one phone number and one sentence of description, then make every place you appear agree with it.
Cause 5: you describe yourself in words nobody asks
Retrieval still starts with a query. If your pages and listings use category language that never appears in the questions people actually type, you are hard to retrieve even when everything else is in order.
Examples of the gap: a firm calling itself a "integrated built environment consultancy" when customers ask for an architect; a clinic offering "aesthetic dermatological interventions" when people search for acne treatment; an agency selling "digital transformation" when the question is who builds websites for restaurants. The formal term may be more accurate professionally. It is not the term in the question.
Location wording has the same failure mode, with an extra trap. An unqualified town name is ambiguous across countries — Mansfield exists in England, Texas and Ohio, and Boston, Birmingham, Perth, Richmond and Hamilton all exist in several. Writing the county, state or country alongside the town removes a genuine source of wrong-continent answers.
You do not have to abandon your professional vocabulary. Add the plain words alongside it, once, in a sentence that states the service and the area in the terms a customer would use.
Working out which one is yours
Run these in order. Each step either finds your cause or rules it out, and stopping early is fine.
- Ask the real question. Put the question a customer would ask to two or three assistants, with your town qualified by county, state or country. Record which businesses are named.
- Read the citations. List every source the assistant linked. Open each and search for your business name. Absent from all of them points straight at cause one.
- Strip your own page. Take the page that ranks, view it as plain text, and ask whether a stranger could state from it what you do and where. If not, cause two.
- Check access. Read your robots.txt line by line, then check server logs or your firewall rules for blocked AI agents. Cause three hides here.
- Audit your mentions. Search your business name and look at the first two pages of results. Count how many independent sources describe you, and whether they agree. Sparse or contradictory means cause four.
- Compare vocabulary. Put your own category wording next to the wording in the question you asked in step one. A mismatch is cause five.
- Repeat the whole thing a week later. Answers vary between runs, so one pass tells you about one run, not about your position.
What to expect once you fix it
Be realistic about mechanism and timescale. Where an assistant browses, improvements to your own pages and new third-party mentions can register within days or weeks, because the next search simply returns different documents. Where an answer comes from training data, the timescale is much longer and largely outside anyone's control.
Nobody can promise you a place in an answer. The selection happens inside systems that are not published, the outputs are probabilistic, and your competitors are working on the same problem. What you can do is remove the reasons you are currently excluded, which in most cases are mundane and entirely fixable.
Start by establishing where you actually stand rather than guessing, whether you do that by hand with the checklist above or with a quick baseline check. A diagnosis you trust is worth more than a fix you picked at random.
Common questions
Does ranking on Google help at all with AI visibility?
It helps, but indirectly and less than people expect. Ranking makes you more likely to be in a search result set, which is a necessary step. It does nothing about whether the specific pages an assistant chooses to cite happen to mention you, which is the step that decides the names.
How do I find out which pages the assistant is citing?
Ask it the buying question your customers ask, then read the citations it shows. Most assistants display source links when they have browsed. Do this five or six times with slightly different wording and the same two or three domains will usually keep appearing. Those are the pages that matter.
Could my site be blocking the AI crawlers without me knowing?
Yes, and it is more common than you would think. Blocks arrive through robots.txt rules copied from a template, a bot-protection or firewall product set to challenge unknown agents, or a hosting default. Nothing visibly breaks, so nobody notices until someone checks the server logs directly.
How long after fixing this would I expect to be named?
Where assistants browse live, changes to your pages and to pages describing you can register within days or a few weeks. Where an answer comes from training data instead, the timescale is far longer and largely outside your influence. Nobody can promise a specific result or date.
See whether the assistants name you
Three real customer questions for your trade and town, no signup, about a minute.
Run my free check