Why does your business show up in ChatGPT but not Google AI Overviews?

Your business shows up in ChatGPT and not in Google AI Overviews because the two systems are not looking at the same web. Each assistant answers from its own retrieval layer, and those layers are built by different crawlers, refreshed on different schedules and ranked by different models. Presence in one says almost nothing about presence in another, which is why a single screenshot of a good ChatGPT answer is such a poor report on AI visibility.

Google AI Overviews sit on top of the ordinary Google index. The pages considered for an overview are pages Googlebot has crawled and Google Search already ranks, which is why AI Overview citations correlate with organic position more tightly than any other engine's citations do. A site that has never ranked for a query is unlikely to be quoted in the overview for it.

ChatGPT search works differently. OpenAI runs its own search crawler and index and supplements them with a commercial search partner, so a page can be entirely absent from Google's first pages and still be the cleanest available answer inside ChatGPT's candidate set. The reverse also happens, and it is the more painful version: strong organic rankings, invisible in ChatGPT, because the crawler that builds that index was never allowed in.

The practical consequence is that "AI visibility" is not one number. It is one number per engine, and a programme that reports a single blended score will hide the specific failure that is costing you the specific answer you care about. Our guide to how AI engines choose which sources to cite covers the shared mechanism underneath these differences.

How much do the engines overlap in what they cite?

The engines overlap far less than most teams assume, and the largest published dataset puts a number on it. Profound's analysis of 680 million citations collected between August 2024 and June 2025 found that only 11% of domains cited by ChatGPT were also cited by Perplexity, and that commercial .com domains made up more than 80% of citations across the platforms it studied.

Citation volume differs as much as citation identity. Profound reported Perplexity averaging roughly 21.87 citations per response, the most of any platform it measured, against 7.92 for ChatGPT. An engine that cites twenty sources has room for the fourth-best page on a topic. An engine that cites eight does not, and on that engine being second best is much the same as being absent.

Those two facts together explain a common and confusing pattern. A company with genuinely good content often finds itself routinely cited by the engine with wide retrieval, occasionally cited by the engine with narrow retrieval, and absent from a third whose index never picked the site up at all. Nothing about the pages changed between those three outcomes.

Engine	What feeds its answers	Crawler that matters	What that implies
Google AI Overviews and AI Mode	The standard Google Search index	Googlebot	Organic ranking and indexation decide eligibility
ChatGPT search	OpenAI's own search index plus a commercial search partner	OAI-SearchBot	Blocking GPTBot alone does not remove you; blocking OAI-SearchBot does
Perplexity	Its own crawl and index, with wide reference lists	PerplexityBot	More citation slots per answer, so breadth of coverage can land
Gemini	Google Search grounding	Googlebot, with Google-Extended governing training use	Content that Google cannot rank is unlikely to be grounded

Does blocking GPTBot stop you appearing in ChatGPT?

Blocking GPTBot does not stop you appearing in ChatGPT's answers, and this single confusion accounts for a large share of the cases we see. OpenAI's own crawler documentation lists three separate agents with three separate jobs: GPTBot collects pages for model training, OAI-SearchBot builds the index behind ChatGPT's search results, and ChatGPT-User fetches a page when a person in a conversation asks for it.

A robots.txt written in 2023 to keep a site out of model training therefore does exactly that, and nothing else. The visibility question is governed by OAI-SearchBot, and a rule that names only GPTBot leaves search access untouched. Blocking OAI-SearchBot, by contrast, removes the site from ChatGPT search answers while leaving the training block unchanged.

Google has the same trap with different names. Google-Extended controls whether content is used for Gemini model training and has no bearing on eligibility for AI Overviews, because those draw from the standard Googlebot index. Blocking Google-Extended as an AI opt-out does not opt you out of AI Overviews, which is the surface most businesses were actually worried about.

One newer directive deserves a check on every template. Google has updated its robots meta tag documentation so that nosnippet applies to its AI surfaces and max-snippet limits what they may use, which means a snippet preference set years ago for ordinary search results can now be quietly suppressing AI Overview eligibility. It fails silently, it is one line, and nobody remembers setting it. Our guide on whether to block AI crawlers works through the trade-offs in full.

How do you tell which engine is failing you, and why?

Telling which engine is failing you takes a test that separates three outcomes rather than one. For each buyer question, run it several times on each engine and record whether you were named in the prose, whether you were cited with a link, and which pages were cited in your place. A brand mentioned without a citation and a brand cited without a mention are different problems with different fixes.

Read the pattern across engines before touching a page. Absent everywhere usually means the question itself never reaches your content: wrong vocabulary, no page on the topic, or a retrieval block. Present on one engine and absent on another points at that engine's index or at the corroboration it relies on, not at the quality of the writing, since the writing is identical in both cases.

Check the mechanical explanations in order, because they are cheap and they are common. Is the engine's crawler allowed in robots.txt by name. Does the page deliver its answer in server-rendered HTML rather than only after JavaScript executes. Is the page indexed in Google at all, which is a precondition for AI Overviews. Is there a stray nosnippet or a low max-snippet in the template.

Only then look at the content, and look at it against the competitor being cited instead. Usually the cited page states the answer in the first line under a heading worded the way the buyer asked, and yours reaches the same answer in paragraph four under a heading written in company vocabulary. Our walkthrough of how to check whether AI can read your website covers the access half of this diagnosis.

What should you fix first when one engine ignores you?

Fix access first, because no amount of rewriting helps a page an engine never retrieved. Allow the specific crawler by name, confirm it is fetching successfully in your server logs, and remove any directive that limits snippet use. Verify the page returns its substance in the initial HTML response. This is unglamorous work that changes outcomes more often than a content rewrite does.

Fix vocabulary second. An engine matches the buyer's question to text, so a page that uses the market's words wins retrieval against a page that uses the company's words. Where the market says one thing and your positioning document says another, the market wins on the page, and the internal vocabulary can live in the sales deck. Alignment with how buyers phrase the question carries 35% of the score in the SIGNALS framework, more than any other dimension.

Fix corroboration third, and expect it to be slow. An engine that cannot resolve who you are from independent sources will hedge by naming a competitor it can resolve. Consistent descriptions across your own site, accurate entries wherever your category is compared, and genuine presence where your customers discuss the problem all feed that resolution. None of it is a trick, and all of it takes months rather than days.

Then re-measure per engine, not in aggregate. The point of the diagnosis is that the engines disagree, so a blended score will mask the recovery of one and the decline of another. Keep the prompt set fixed, sample repeatedly because AI answers vary between runs, and read the source list as carefully as the mention count.

What else do people ask about ChatGPT and Google AI Overviews?