How Does a Business Get Found and Recommended by ChatGPT?

The study states its own limits, and they travel with the figure. ChatGPT there is a model with web search enabled behind a vendor API, which is not the product people type into, and the consumer app may retrieve differently. It counted cited sources rather than businesses named or recommended, because, in its words, this corpus measured links. What it does show is the pattern a business is up against: the uncited answers were written from the model rather than from retrieved pages, and a brand cannot be cited in an answer that cites nobody.

Whether a particular brand is missing because the answer cited nobody, or because it cited others, is a separate check, set out on What Is AI-Mediated Search.

OpenAI also says that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links, and that after a robots.txt change it can take about 24 hours for OpenAI’s systems to adjust. Its other crawlers do different jobs. GPTBot collects content that may be used to train OpenAI’s models, and OpenAI treats the two settings as independent, so a site can allow search while disallowing training. ChatGPT-User visits pages in response to a person’s request in ChatGPT, and OpenAI says it is not used to determine whether content may appear in search.

Whether allowing the training crawler changes which businesses ChatGPT names when it does not search is not something OpenAI documents, and Hendricks has not measured it. Crawler access makes a site eligible to be found by ChatGPT search. It does not make it chosen.

For products, OpenAI says more. A product appears when ChatGPT perceives it to be relevant to the person’s intent, considering the query and context such as memory or custom instructions, and the merchants offering a product are ranked on factors like availability, price, quality, and whether they are the maker or primary seller of the item.

Two things follow, and both are inference rather than documentation. The query ChatGPT runs is not always the question a person typed, so a business can be passed over for a search no site owner sees. And because OpenAI names relevance and reliability as the aim without saying how either is judged, anything sold as the ChatGPT ranking formula is a claim OpenAI has not made. How AI systems in general move from retrieval to recommendation is modeled, as testable handoffs rather than a documented mechanism, on What Is Search Intelligence Engineering.

A list of businesses one person saw is therefore not a ranking anyone else can be assumed to see, and that sentence is an inference from OpenAI’s documentation rather than a measured rate. A single answer screen is one observation under one set of conditions.

The Answer Index offers no stability figure for ChatGPT: ChatGPT and Gemini were not re-run, so no self-agreement is claimed for them. How a Hendricks reading handles personal context, without claiming to reproduce any one person’s account, is set out with the methodology’s context panels.

An ad is a labeled placement shown separately from the answer, not a place in it, and OpenAI says seeing one does not mean OpenAI endorses or recommends the advertiser. A firm that sells a ChatGPT ranking is selling a position OpenAI says it does not guarantee, which is the first red flag in the answer engine optimization guide.

A position on Google is a different outcome again, and why a brand can hold one and be absent from AI answers is set out on What Is AI-Mediated Search.

For questions about nearby businesses, OpenAI says ChatGPT may share a person’s location with trusted third-party providers of local information, such as business listings. That passage does not say which providers supply those listings or how ChatGPT uses them, so it is an inference, not a documented mechanism, that a business’s listings elsewhere are part of how it is found for nearby questions.

Hendricks has run no controlled test of any of those conditions on ChatGPT, and this guide recommends none as a lever with a measured effect. No firm, Hendricks included, controls the recommendation itself, as the Search Presence Engineering page sets out beside the conditions a brand can change.

The order the work should run in, and how to judge anyone selling it, is in the answer engine optimization guide.

What that reading measures, and why a count of appearances is not the same as being considered, is defined on What Is Selection Intelligence. A free one-run reading of a single brand is the AI Visibility Check.

The conditions this guide says a business controls are the work Hendricks calls Search Presence Engineering, which starts from a reading of where a brand is missing or misdescribed rather than from a list of tactics.

Last reviewed November 9, 2026.

This guide states the Hendricks point of view on how a business is found and recommended by ChatGPT. Every statement about how ChatGPT works is taken from OpenAI’s own documentation, listed below, the figures are quoted from one Hendricks study, and no firm or tool is recommended.

The position in this guide is applied in the Search Presence Engineering solution and the Search Intelligence Diagnostic.