How ChatGPT Search actually works — and what it means for AEO

OpenAI has published detailed documentation on how ChatGPT Search works, and the technical specifics matter considerably more than the headline feature. The documentation confirms several behaviours that have significant implications for brands and publishers trying to optimise for AI-generated answers — from how queries are rewritten before they reach the web, to which third-party data providers are involved, to how the system handles location, memory, and commerce.

<h2>Query rewriting is the core mechanism</h2>

The most important detail in the documentation is one that most coverage has underplayed: ChatGPT does not search the web with your words. It rewrites your prompt into one or more targeted queries before anything reaches a search provider. The documentation gives a concrete example: a researcher asking about drugs targeting CCR8 for cancer might trigger an initial query of "CCR8 immunotherapy drug development 2025" followed by a more specific follow-up query of "CHS-114 conference 2025" sent to a different provider. This multi-step, multi-provider query expansion means that the relationship between what a user asks and what content gets retrieved is not direct. A brand optimising purely for its own brand terms may miss the intermediate queries that actually drive retrieval. The implication for AEO practitioners is that content needs to be structured around the kinds of intermediate, specific queries that a rewriting system would generate — not just the broad conversational prompts that users type.

<h2>Bing and Shopify are the named data partners</h2>

The documentation names two third-party search providers: Microsoft Bing and Shopify. Bing's inclusion is not surprising — it has been the underlying web index for ChatGPT Search since launch. Shopify's inclusion is more significant. It signals that ChatGPT Search is drawing on structured commerce data for product and merchant queries, not just general web content. For brands selling through Shopify, this means that product catalogue quality, merchant data completeness, and Shopify's own structured data may influence how ChatGPT represents those brands in commerce-related answers. For brands not on Shopify, it raises a question about which commerce data sources ChatGPT will expand to over time.

<h2>The restaurant reservation layer is a preview of agentic commerce</h2>

The documentation describes a fully functional restaurant reservation flow within ChatGPT: users can search for restaurants, view availability from third-party providers, and complete a booking without leaving the conversation. ChatGPT prefills details from the prompt — party size, date, time — and passes them to the reservation provider. This is not a minor feature. It is a working example of ChatGPT acting as a transactional layer between a consumer and a business, with the AI controlling which businesses appear, which are shown as bookable, and which reservation providers are surfaced. For hospitality brands, the practical implication is immediate: visibility in ChatGPT Search is no longer just about being mentioned — it is about being bookable. Brands that are not connected to a supported reservation provider will not receive the Reserve button, regardless of how well their content is optimised.

<h2>Memory integration changes the query context</h2>

When a user has Memory enabled, ChatGPT can incorporate stored preferences into its query rewriting. The documentation gives the example of a user asking for restaurant recommendations who has a stored memory that they are vegan and live in San Francisco — ChatGPT rewrites the query to "good vegan restaurants San Francisco" without the user specifying either detail. For brands, this means that the query context ChatGPT uses to retrieve content is not always visible in the user's original prompt. A brand that appears for "vegan restaurants San Francisco" may receive traffic from users who asked something much more generic. Conversely, a brand that does not appear for the specific rewritten query may be invisible to users who would otherwise be highly relevant prospects.

<h2>OAI-Searchbot crawling is the minimum requirement</h2>

The documentation is explicit that there is no way to guarantee top placement in ChatGPT Search, but it identifies one necessary condition: allowing OAI-Searchbot to crawl the site. Publishers and brands that have blocked OpenAI's crawler — either deliberately as part of an AI opt-out strategy, or inadvertently through CDN or firewall rules — will not be included in ChatGPT Search results. This connects directly to the broader debate about AI opt-out costs. As ChatGPT Search becomes a more significant traffic and commerce channel, the cost of blocking OAI-Searchbot rises. The documentation also notes that the site's host and content delivery network must allow traffic from OpenAI's published IP addresses — a technical requirement that many brands may not have verified.

<h2>Location data is layered and partially automatic</h2>

The documentation distinguishes between two types of location data: IP-based location, which is used by default and shared with search providers, and device location, which is optional and more precise. IP-based location is already in use without any user action — ChatGPT may rewrite a query like "restaurants near me" into "top restaurants San Francisco" based purely on IP geolocation. For local businesses, this means that local SEO signals and structured location data are directly relevant to ChatGPT Search visibility, even for users who have not explicitly enabled location sharing.

<h2>AEO Updates Takeaway</h2>

The ChatGPT Search documentation is one of the clearest primary sources available on how a major AI search system actually retrieves and presents content. The key implications for AEO practitioners are: first, optimise for intermediate queries, not just brand terms; second, ensure OAI-Searchbot is not blocked at the CDN or firewall level; third, for commerce brands, evaluate whether Shopify integration and third-party reservation connectivity are relevant to your category; and fourth, treat location data quality as an AEO signal, not just a local SEO concern. The documentation also confirms that ChatGPT Search is already a transactional layer, not just an informational one — a distinction that will become increasingly important as the reservation and commerce features expand beyond restaurants.

<h3>References</h3>

[1] OpenAI, <a href='https://help.openai.com/en/articles/8077698-how-chatgpt-search-works' target='_blank' rel='noopener noreferrer'>ChatGPT Search Help Documentation</a>, accessed August 2026.