Answers / Getting cited

How does ChatGPT decide which businesses to recommend?

By Hoss · Published July 13, 2026 · Built on the engines' own documentation; sources at the end

Step 1: it decides whether to search at all

Not every question triggers retrieval. When ChatGPT judges it can answer from what it already knows, it does — no citations, no live sources, just its training-data impression of the world. When it searches, the answer is grounded in retrieved pages it then cites. This split matters practically: a grounded answer can change as soon as the index picks up new evidence, while an ungrounded one only changes when the model itself is retrained. (It's also why answers vary between runs of the same question — a reason single spot-checks mislead, covered in the self-check guide.)

Step 2: retrieval — can it even see you?

ChatGPT's search features are fed by OpenAI's own crawler. From OpenAI's crawler documentation:

"OAI-SearchBot is used to surface websites in search results in ChatGPT's search features. Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers."OPENAI CRAWLER DOCUMENTATION [1]

Three consequences follow directly. First, retrieval is a gate: if your site blocks OAI-SearchBot — or a firewall challenges it — you're not in contention no matter how good your reputation is (this is a separate control from the GPTBot training crawler; the distinction trips up a lot of sites). Second, what the crawler can extract matters: pages whose substance renders as clean, readable text get used; pages that are effectively blank to a machine don't. Third, the underlying index leans on Bing's, so being indexed there is part of eligibility — one legal-market analysis found 87% of what ChatGPT cites comes from top-10 search results [2].

Step 3: evidence — why third parties decide it

A recommendation is a confidence claim, and the engine builds confidence by cross-referencing. Independent citation research keeps finding the same pattern: for business-recommendation queries, the cited sources are dominated by review platforms, directories, community discussion, and roundup articles rather than the businesses' own sites. A Semrush analysis of 150,000+ AI citations found 40.1% pointed to Reddit and 26.3% to Wikipedia [3]; an Ahrefs study of 9.6 million queries put Reddit at 29.4% of ChatGPT citations [4]. For local businesses specifically, published analyses point to business profiles, Yelp and industry directories, reviews, local press, and "best of" roundups as the recurring source set [5].

Your own website still matters — it supplies the facts the answer states about you (services, location, hours) and the extractable statements that let the engine describe you accurately. But it functions as confirmation, not nomination. If the rest of the web is silent about you, there's nothing to corroborate [6].

Step 4: commitment — the short list

Unlike a results page, an answer doesn't rank twenty options — it names two or three and moves on. That's the structural difference that makes this worth caring about: there's no page two, and absence is invisible. Which names make the cut varies run to run (the evidence is probabilistic, and so is the model), which is why measurement in this field means repeated runs of fixed questions, not a screenshot — that's our whole methodology.

And Google's AI Overviews?

Same principle, different plumbing — worth spelling out because the controls are opposite. Google's documentation states that to appear in AI Overviews or AI Mode, "a page must be indexed and eligible to be shown in Google Search with a snippet... There are no additional technical requirements" [7] — Googlebot is the control, and there is no separate AI-Overviews bot. Meanwhile Google-Extended, the token many sites block expecting to escape AI features, "does not impact a site's inclusion in Google Search nor is it used as a ranking signal" — it only governs Gemini training and grounding [8]. And the engines barely share sources: a five-engine analysis of 127,198 citations found only 2.7% of cited domains were cited by all five [9]. Being named in one engine says little about the others; each is its own checklist.

What this means if you want to be the name

• Be retrievable: crawler access (robots.txt and firewall), indexed in Bing and Google, content that survives extraction.
• Be legible: one unambiguous identity, structured data matching your visible text — Google's guidance is explicit on that match [7] — and pages that answer the exact questions buyers ask.
• Be corroborated: reviews, directories, community presence, third-party coverage — the evidence layer that actually nominates you.
• Then measure: repeated runs of your buyers' real questions, per engine, over time — because every one of the steps above is invisible until an answer changes.