Why ChatGPT Doesn't Recommend Your Business: The 6 Real Causes

You run a restaurant and you have already looked yourself up? You are a plumber and you never come up for your own neighbourhood? ChatGPT answered confidently, naming three or four businesses. Yours was not among them.

This is not a penalty. It is almost always a mechanical problem, with specific causes you can diagnose one at a time.

The question has stopped being theoretical: the share of consumers using AI to find a local business has gone from 6% to 45% in a single year (BrightLocal, March 2026).

The short answer

Why doesn’t my business show up in ChatGPT? If you are asking, the answer is one of six causes: your information contradicts itself across platforms, your reviews are too few or in the wrong places, your site blocks the crawlers that feed AI search, no outside source talks about you, your positioning is too vague to be recommended. Or, deeper still: there is no structured source the assistant can query about you.

The first five are fixed by producing more signal. The sixth requires something else.

Summary

• ChatGPT does not read your site live. It relies on its own web index, a shared cache, and, more rarely, opening a page on demand. None of those three paths is a real-time look at your website.
• AI visibility is far harder to earn than Google visibility: across nearly 350,000 locations analysed, ChatGPT recommends only 1.2% of them, against 35.9% appearing in Google’s local 3-pack (SOCi, Local Visibility Index 2026).
• Being mentioned isn’t enough: when ChatGPT does describe a business, its information is accurate roughly 68% of the time. It describes you from memory, out of fragments written by other people.

How ChatGPT finds information about a business

This is the founding misunderstanding. Until it is cleared up, every remedy risks missing the point. ChatGPT does not go and read your website when a customer asks a question. It takes three routes, and none of them is your live site.

A proprietary web index. ChatGPT no longer queries Bing’s index as it did early on: OpenAI now runs its own, and the most detailed technical analysis published to date shows very little overlap with Bing: roughly 1.5% of the URLs it retains also appear in Bing’s top 20 for the same query (Search Engine Land, August 2026). Ranking well on a classic search engine does not guarantee existing in that one.

A shared cache. Retrieved pages are converted to text and then stored for all users, staying fresh for around thirty minutes. During that window your server sees nothing, and the cache ignores Cache-Control: no-store headers as well as noindex tags (same source).

An on-demand page open, rarer. Sometimes the assistant genuinely fetches a page, through the ChatGPT-User agent. It is a minority of cases but a decisive one: pages actually opened are cited in the answer around 74% of the time, against 7% for pages merely surfaced by the index.

Training data, finally, freezes a snapshot of the web at a past date. If your business opened last year, moved, or changed its name, the model does not know and will not know until its next version.

The key point. You are not missing from a database somebody forgot to fill in. You are missing from a reconstruction the assistant assembles out of what other people have written about you.

The six causes, from the most common to the deepest

1. Your information contradicts itself across platforms

Your Google listing says 7pm to 11pm, your Facebook page says 6.30pm to 10pm, your website still shows summer hours, and a local directory has you at an address two moves ago. To you, that is administrative clutter. To an assistant reassembling your business from those sources, it is a conflict with no arbiter.

The consequence isn’t only absence. It’s error. When ChatGPT describes a business location, the data is accurate about 68% of the time, against 100% for Gemini, which draws directly on Google Maps (SOCi, 2026). That gap says everything: an assistant backed by a single authoritative source almost never gets it wrong. An assistant stitching fragments together does.

The fix: make name, address, phone and hours identical everywhere, starting with the platforms you control.

2. Your reviews are too few, too old, or in the wrong place

An assistant that has to recommend looks for trust markers it can read. Reviews are among the densest of them. Three reviews from two years ago carry little weight; forty recent ones do.

But volume isn’t everything: what also matters is where they sit. The review platforms that feed AI answers vary sharply by sector and by country, and reviews piling up on a platform the assistant never consults produce no effect at all.

The fix: identify the two or three platforms actually cited in your sector and concentrate your effort there rather than spreading it.

3. Your site blocks AI crawlers (and not the ones you think)

This is the most frequently mishandled cause, because “blocking AI” is not a single action. Each provider runs several crawlers with distinct roles, and confusing them can be expensive.

Crawler	Role	Blocking it means
GPTBot	Training OpenAI models	Your content stays out of future models
OAI-SearchBot	Feeds search inside ChatGPT	You disappear from ChatGPT’s answers
ChatGPT-User	Opens a page at a user’s request	User-initiated; robots.txt rules may not apply
ClaudeBot	Training Anthropic models	Your content stays out of future models
Claude-SearchBot	Feeds search inside Claude	You disappear from Claude’s answers
PerplexityBot	Feeds Perplexity search (no training)	You disappear from Perplexity’s answers

Sources: OpenAI, Anthropic, Perplexity.

The classic mistake is writing a broad rule to “protect my content from AI” and cutting off, in the same move, the crawler your presence in answers depends on. Refusing training is a legitimate decision. Refusing search is a different one, with an immediate commercial cost.

Since September 2025 a syntax exists to separate the two: Cloudflare’s Content Signals Policy, which declares three distinct uses in robots.txt: search (indexing and search results), ai-input (using the content to compose a generative answer) and ai-train (training). That lets you allow the first two and refuse the third, in a single line added to your robots.txt.

One honest caveat: these signals express a preference, not a technical block. Cloudflare says so explicitly. Nothing compels any provider to honour them.

The fix: open yourdomain.com/robots.txt and check, line by line, which agents are being refused.

4. Nobody outside is talking about you

An assistant builds its answer from what the web says about you, not only from what you say about yourself. Your own site is one source among many, and not necessarily the most heavily weighted. A local press article, a listing in a sector guide, a mention in a comparison, a forum thread: those third-party traces are what let a name surface.

This is the slowest cause to fix, and the most unfair to a young or discreet business.

The fix: aim for few mentions, but mentions that count: the ones already cited in your sector’s answers.

5. Your positioning is too vague to be recommended

An assistant doesn’t answer “which is the best business?”. It answers a specific question: an Italian restaurant with a terrace, open on Sundays, in this neighbourhood. To be the answer, you have to be assignable to a case.

“We support our clients through their projects” assigns to nothing. “We restore slate roofs within 40 km of Rennes” assigns immediately. Vagueness isn’t a stylistic flaw: it is a technical impossibility to be selected.

The fix: write down plainly what you do, for whom, where, and at roughly what price.

6. Nothing about you is structured to be queried

You have fixed everything. Your information is consistent, your reviews solid, your crawlers open, people talk about you and your positioning is sharp. And still the assistant describes you approximately, or skips you in favour of a booking platform.

Because it still has nothing to query.

Everything you did amounts to improving the raw material of a reconstruction. The assistant goes on rebuilding you from memory, out of fragments. Better written, more consistent fragments, but fragments all the same.

It cannot ask whether you have a table free on Thursday at 8pm, because nowhere does a source exist to put the question to.

That is the mechanism we unpack in why your website is a black box to AI, and the reason for what we call an AI presence: a structured, up-to-date source that assistants query directly instead of guessing at you. See what that looks like in practice.

How to show up in ChatGPT as a local business

If you want an order of priority rather than a list of causes, here it is. The fastest wins are at the top.

1. Check your robots.txt. It is the only cause that can make you invisible overnight, and the only one you can fix in ten minutes.
2. Align your information across every platform you control, starting with your Google listing.
3. Concentrate your reviews on the two or three platforms cited in your sector, rather than spreading them.
4. Make your positioning assignable: what you do, for whom, where, at roughly what price.
5. Look for outside mentions, few in number but in sources assistants already cite.

Those five steps are useful and worth doing. But they all share one limit, which is the subject of the section below.

The diagnostic table

Find your symptom: you probably have your cause.

What you observe	Likely cause	The test
It never names you, but names competitors	4 then 2	Ask it to cite its sources on a competitor
It names you, but the details are wrong or stale	1	Compare your hours on Google, Facebook and your site
It names you but never recommends you	5	Ask a customer’s question, not your brand name
It finds nothing although you rank first on Google	3	Open your robots.txt, look for OAI-SearchBot
It describes you well, then sends you to a platform to book	6	Ask it to book. Watch where it sends you
It quotes an address or name you left behind	Frozen training data	Check whether that detail is still live elsewhere on the web

The three-question test

Five minutes are enough to place your case. Open ChatGPT in a private window, with no account history, and ask the question a customer who doesn’t know you would ask: their need, never your name.

1. Do you appear? If not, look at causes 3 and 4 first.
2. Is the information correct? If not, that’s cause 1.
3. Can the assistant do anything, or does it merely name you and send you elsewhere? If it sends you elsewhere, that’s cause 6.

Repeat on Claude, Gemini and Perplexity: the gaps between assistants are wide, and they tell you as much as the results themselves.

What the first five causes don’t solve

The first five causes share one thing: they ask you to produce more signal so that someone else describes you better. The sixth asks a different question: what if the AI could ask you directly?

That is the whole difference between being read and being queryable.

An assistant that reads you produces a description: dated, approximate, dependent on what it happened to find. An assistant that queries you gets an answer: accurate, current, and something it can act on.

The first five fixes remain useful, and nothing here says to neglect them. But they optimise a reconstruction. They don’t replace it.

Which is precisely why getting cited by AI isn’t enough.

See what this looks like in your line of work →

FAQ

Does ChatGPT read my site live?

No, not by default. It relies on its own web index, on a shared cache of already-retrieved pages, and more rarely on opening a page on demand through the ChatGPT-User agent. Your live site is consulted only in that last, minority case.

Why do I rank first on Google but never appear in ChatGPT?

Because they are two different indexes. OpenAI stopped relying on Bing and runs its own, which overlaps very little with classic search results: roughly 1.5% of the URLs it retains also appear in Bing’s top 20 for the same query (Search Engine Land, August 2026). A good Google position does not carry over.

Do Google reviews influence ChatGPT’s answers?

Indirectly, and less than people assume. Gemini draws directly on Google Maps, which is why its business information is 100% accurate, against roughly 68% for ChatGPT and Perplexity (SOCi, 2026). A well-kept Google listing therefore helps Gemini a great deal, and ChatGPT only in part, since it looks elsewhere.

What is the difference between local SEO and GEO?

Local SEO targets a position in a results page, with a human clicking afterwards. GEO targets a mention inside a generated answer, with no results page and no guaranteed click. The two overlap little in practice, as the gap between the indexes shows. The full vocabulary is laid out in our AI search glossary.

Do I have to pay to appear in ChatGPT?

No. There is no way to buy a position in the answers. OpenAI runs a crawler dedicated to checking pages submitted as advertisements (OAI-AdsBot), but that is an advertising system distinct from organic recommendations.

How long before ChatGPT knows my business?

It depends on the route. An information fix can show up within days to weeks through the search index. Training data only updates with the model’s next version: expect several months, with no guarantee.

Does blocking AI crawlers protect my content?

Only partly, and at a cost that is often overlooked. Blocking GPTBot refuses training; blocking OAI-SearchBot removes you from ChatGPT’s answers. Those are two different decisions. And in both cases, robots.txt expresses a preference, not a technical lock.

Is a Google Business Profile enough?

It helps a great deal, but unevenly depending on the assistant. Gemini draws on Google Maps and shows 100% accuracy on business information for that reason, against roughly 68% for ChatGPT and Perplexity (SOCi, 2026). An up-to-date profile is a necessary base, not full coverage.

My competitor appears and I don’t. Why?

Usually because they are cited elsewhere (local press, a sector directory, a comparison, a booking platform), and not because they are better. The assistant isn’t comparing qualities; it’s comparing what it manages to read.

Sources

• SOCi, Local Visibility Index 2026, ~350,000 locations across 2,751 multi-location brands, via Search Engine Land, 28 January 2026 (link)
• BrightLocal, Local Consumer Review Survey 2026, 1,002 US consumers, March 2026 (link)
• Search Engine Land, “Inside ChatGPT’s retrieval stack”, 17 August 2026 (link)
• OpenAI, crawler documentation (link)
• Anthropic, ClaudeBot documentation (link)
• Perplexity, crawler documentation (link)
• Cloudflare, Content Signals Policy, 24 September 2025 (link)