Why Your Business Doesn't Show Up in ChatGPT (And How to Check)

A customer opens ChatGPT and types something like "best accountant in Manchester for a small ecommerce business". They get three names, a short explanation of each, and a decision. You were not one of the three.

Nothing appeared in your analytics. No impression, no click, no trace. As one agency put it, there is an entire layer of research happening that will not show up in any analytics platform you own, because the person never visited a website at all.

Here is why that happens, and how to check your own situation in about ten minutes.

Why doesn't ChatGPT mention my business?

Usually for one of five reasons: the model never learned you exist, your site blocks the crawlers that would look you up, your pages cannot be quoted in usable pieces, no credible third party sources mention you by name, or the system cannot confirm you are a real, distinct organization. These are separate failures with separate fixes, and most businesses have more than one.

Key takeaways

• An AI answering from memory alone and an AI searching the live web are two different systems with two different failure modes.
• Blocking AI crawlers in robots.txt removes you from consideration entirely, and many sites do it without realizing.
• Outside mentions matter more than most owners expect. Ahrefs found unlinked brand mentions correlated with AI visibility roughly three times as strongly as backlinks.
• Checking your own visibility while logged in gives you a distorted answer, because memory and location personalize the result.
• The same prompt returns different answers on different runs, so one check proves very little.

Does the AI even know your business exists?

There are two entirely different ways an AI tool can answer a question about you, and confusing them wastes a lot of money.

The first is memory. A language model learns from a fixed snapshot of text taken before a cutoff date, commonly twelve to eighteen months before the model is released (Wikipedia). If you opened last year, or nobody wrote about you before that snapshot, the model simply has no stored impression of you, and no amount of website work changes that until a future training run.

The second is live retrieval. When ChatGPT searches, or when you use Perplexity or Google AI Overviews, the system runs a real web search at the moment you ask, pulls back documents, and writes its answer from those (Firecrawl).

This second path is the tractable one. It is fought fresh on every query and does not require waiting for anyone to retrain a model. It is also growing fast: Cloudflare reported that crawling tied to answering live user requests grew more than fifteenfold across 2025, the fastest growing crawler category they tracked (Cloudflare).

Are you blocking the crawlers that look you up?

This is the most common fixable cause, and the most embarrassing one.

Many sites, security plugins and hosting providers block unfamiliar bots by default. If your robots.txt or your firewall turns away the agents that fetch pages for AI answers, you are not competing badly. You are not in the contest at all.

HubSpot co founder Dharmesh Shah puts this at the top of his checklist: explicitly allow OAI-SearchBot, which fetches pages to answer live questions, and GPTBot, which crawls for training. Without that, in his words, you have no chance of winning.

Check this before you change anything else. It costs nothing and takes five minutes.

Can your pages be quoted in pieces?

An AI answer engine does not read your page the way a visitor does. It retrieves a passage, strips it out of context, and passes that fragment to the model alongside fragments from four or five other sites.

The funnel is brutally narrow. A technical breakdown of Google AI Overviews describes an initial pool of roughly 200 to 500 candidate documents being cut down to the 5 to 15 sources actually cited, with an average of about 8 sources appearing per answer (ZipTie.dev).

So a page written as a flowing narrative, where the useful sentence only makes sense after the setup above it, is a weak candidate. A page where each section opens with a complete, self contained answer under a heading phrased as a real question is a strong one.

This is also why the content that earns citations is more boring than most marketing teams want. Practitioners report that specific product content, named features and use cases, FAQs and help center pages earn citations, while general thought leadership posts largely do not.

Does anyone else on the web mention you?

This is the reason most owners have not considered, and the evidence behind it is unusually strong.

Ahrefs analyzed 75,000 brands to see which signals correlated with appearing in Google AI Overviews. Unlinked mentions of the brand across the web correlated at 0.664. Traditional backlinks correlated at 0.218 (Ahrefs). Being talked about was roughly three times as predictive as being linked to.

Where those mentions live matters too, and it differs by platform. In a study of citation patterns across a very large sample of AI answers, Wikipedia alone accounted for about 7.8 percent of ChatGPT citations and nearly half of its ten most cited sources, while Perplexity leaned heavily on Reddit, which made up around 6.6 percent of its total citations (Profound).

One finding is worth flagging because it contradicts what most people assume: in a separate study of US ChatGPT citations, major outlets including the Wall Street Journal, the New York Times and Bloomberg did not appear in the top 20 sources at all (5W Public Relations, via PR Newswire). A national press hit is not the lever people think it is.

Are you a recognizable entity, or just a website?

Answer systems need to confirm that you are a specific, real organization and not an ambiguous name. That confirmation runs through structured data: an organization schema block on your site, consistent name, address and phone details everywhere they appear, and the sameAs property, which explicitly tells machines that your website, your LinkedIn page and your other profiles are all the same real world entity (1Digital).

If three businesses share a similar name in your city and none of you have done this, the system has no reliable way to tell you apart, and it will often name whichever one it can verify.

How can I check whether ChatGPT mentions my business?

Ask the question your customer would ask, not your brand name, and ask it several times in a clean session. Searching for your own company name proves nothing, because you have just told the system the answer.

Three rules make a manual check useful rather than misleading:

• Log out, or use a private window with memory disabled. A practitioner recently described running a visibility test with account memory left on and getting a confidently stated, completely false answer back. Personalization contaminates the test.
• Ask the buying question, phrased naturally. Real AI prompts run much longer than search queries, often ten words or more rather than three.
• Run it at least five times, and across more than one tool. One run is an anecdote.

SignalCheck automates the tedious parts of this. It checks your crawler access and page structure, then runs a live grounded query asking an AI model who it would recommend in your category and city from a neutral session with no memory or location bias, and shows you the raw answer text rather than only a score. Seeing the actual sentence that named three competitors and not you tends to be more useful than any number.

Why does the answer change every time I ask?

Because these systems are probabilistic, and this is a genuine limitation rather than a bug in your testing.

Visibility varies across paraphrases, interfaces, models, logged in states, locations and repeated runs (Search Engine Land). An academic study of the problem found that measurement error only drops to an acceptable level after roughly 7 repeated runs of the same prompt (arXiv:2604.07585).

Treat any tool that reports a precise AI visibility score from a single query with suspicion, including ours. What a single check reliably tells you is directional: whether you appear at all, and who does appear instead.

What should I fix first?

In this order, because the early items are cheap and the later ones are slow:

1. Unblock the AI crawlers, then confirm the fix.
2. Rewrite your two or three most commercially important pages so each section opens with a direct answer under a question shaped heading.
3. Add organization structured data, with sameAs links to your verified profiles.
4. Identify the specific sites that appear in the AI answers for your category, then work honestly to be present on them.

Step four is where the real work lives, and it is the slow part. Getting genuinely discussed in the places your buyers already trust cannot be sprinted.

One last honest note. Being cited is not the same as being paid. Agencies working in this space have publicly pointed out that brands earn citations in ChatGPT and AI Overviews without a matching lift in conversions, usually because the page the AI sends people to was never built to convert a visitor who arrives already half decided. Fix the visibility problem, then check that what happens next is worth the traffic.

Frequently asked questions

Does ChatGPT actually browse the web when someone asks about my business? Sometimes. In search mode it retrieves live pages before answering. In a plain conversation with no search tool, it answers from training data with a cutoff date and cannot know about anything newer. The two paths need different work.

If I rank number one on Google, will ChatGPT recommend me? Not reliably. Kevin Indig's research found only about 62 percent overlap between what ranks well in classic Google and what appears in AI answers (Search Engine Land). Strong rankings help, but they are not a guarantee.

How long until this changes after I fix things? Technical and structural fixes tend to show movement in about 2 to 4 weeks, while stable citation behavior across engines takes months (WebFX). The authority work is the slow half.

Can I pay to appear in ChatGPT answers? Not in the way you buy search ads today. The current levers are being crawlable, being extractable, and being credibly discussed elsewhere.

Should I worry that AI reads my content and sends nobody back? It is a legitimate concern. Cloudflare measured Anthropic's crawlers fetching roughly 38,065 pages for every referral visit sent back, against 5 to 1 for Google (Cloudflare). Since the content is being read regardless, being named in the answer is the only part of that exchange that returns anything to you.