Where ChatGPT Search works and how searching starts
OpenAI provides ChatGPT Search across its supported ChatGPT plans. Logged-out Free users can also access the search feature. Access is available through chatgpt.com, desktop apps, and mobile apps.
Readers can open View all tools and select Search. Typing a slash also opens the Search option. ChatGPT may search automatically when web information could improve its response.
Search answers may contain linked text, images, or maps. Desktop users can open citation previews by hovering over citations. Mobile results may include maps when location information is useful.
Search activity remains subject to current ChatGPT usage limits. OpenAI lists current access methods within its ChatGPT Search documentation.
How ChatGPT Search turns a prompt into web queries
ChatGPT Search may rewrite one prompt into several targeted queries. Those queries can reach third-party search providers, including Bing and Shopify. ChatGPT reviews returned results before preparing its response.
Additional queries may follow after ChatGPT reviews the first results. Earlier messages can also influence queries used for follow-up questions. OpenAI does not publish a fixed source-selection formula.
For example, someone asks about Delhi Metro service changes during June. Search may create narrower queries covering dates, lines, and official notices. ChatGPT can then combine relevant findings within one cited response.
Actual queries and sources may differ for every search. The ChatGPT Search help page documents query rewriting and named search providers.
How citations and the Sources panel identify webpages
Inline citations can appear beside claims within a ChatGPT Search answer. Selecting a citation opens the webpage connected with that claim. Desktop users can hover over citations for additional source information.
When inline citations are absent, readers can open the Sources panel. The panel lists cited webpages and other relevant web links. Images may also link to their original webpage sources.
One citation may cover only one part of an answer. Readers should compare each important claim with its cited passage. Publication dates and named authors can reveal outdated or missing context.
OpenAI adds utm_source=chatgpt.com to ChatGPT referral URLs. Analytics tools can use the UTM parameter to identify referral visits. Referral tracking cannot measure brand mentions without a linked visit.
OpenAI documents the tracking parameter for publishers. The Answer Engine Optimization entry covers wider citation and direct-answer practices.
What third-party search providers receive
ChatGPT Search may send rewritten queries to third-party search providers. It may also share a general location for nearby results. OpenAI does not share account information or the IP address itself.
ChatGPT estimates general location from an internet connection. That estimate may identify a country, state, or city. General location can improve weather, news, and nearby business results.
Memory can shape rewritten queries when users enable that feature. A saved dietary preference may narrow a nearby restaurant search. Users can review, remove, or disable saved memories.
Device location sharing remains optional and disabled by default. Precise location can improve nearby recommendations when users enable access. OpenAI deletes precise location after producing the requested response.
Place names and maps may remain within chat history. Users can disable location access through ChatGPT Data controls. Each third-party provider processes shared queries under its privacy policy.
Search prompts should omit personal details that the request does not require. Removing private information limits unnecessary sharing with third-party search providers. OpenAI documents these controls within its ChatGPT Search privacy information.
Which OpenAI crawlers affect search inclusion
OpenAI uses separate user agents for search, training, and user-requested visits. Each user agent controls a different web access purpose.
| User agent | Main purpose | Effect on ChatGPT Search |
|---|---|---|
| OAI-SearchBot | Automatic crawl access for search | Controls eligibility for ChatGPT Search answers |
| GPTBot | Possible model-training use | Does not control normal search inclusion |
| ChatGPT-User | Visits requested by ChatGPT users | Does not determine automatic search inclusion |
OAI-SearchBot helps surface public webpages within ChatGPT Search answers. Blocked pages cannot appear as content within normal search answers. OpenAI notes that blocked pages may remain as navigational links.
Website owners should permit OAI-SearchBot through their robots.txt rules. Their hosting service or content delivery network must also permit OpenAI IP ranges. Crawl permission creates eligibility but never promises citation or placement.
GPTBot controls whether crawled content may enter model training processes. Website owners can allow OAI-SearchBot while blocking GPTBot. Search inclusion and model training use separate crawler controls.
ChatGPT-User visits webpages after certain requests from ChatGPT users. Automatic web crawling does not use the ChatGPT-User agent. OpenAI states that robots.txt rules may not govern user-requested visits.
The following file allows search crawling while blocking possible training use. Each rule applies to one named OpenAI user agent.
OpenAI systems may need about 24 hours after a rule update. The OpenAI crawler documentation describes every crawler role and published IP range.
Crawler permission never protects private files from unauthorized access. RFC 9309 defines robots.txt as crawler instructions, not access authorization. Private content needs login protection or server-level access controls.
RFC 9309 documents that important security boundary. Crawler access and website security require different technical controls.
How ChatGPT Search differs from SearchGPT and answers without web search
SearchGPT was a temporary search prototype announced during July 2024. OpenAI launched ChatGPT Search on October 31, 2024. The current feature brought selected SearchGPT ideas into ChatGPT conversations.
ChatGPT Search can retrieve current web information and provide linked sources. An answer without web search uses available model and conversation context. ChatGPT can choose web search without a manual reader selection.
Traditional search pages generally list results for readers to open. ChatGPT Search writes a response and connects claims with cited webpages. Both formats require source checks before important personal or business decisions.
The Generative Engine Optimization entry covers visibility across generated answer systems. ChatGPT Search is one product where GEO practices may apply.
OpenAI records the prototype and launch history within its ChatGPT Search announcement.
How readers can verify a ChatGPT Search answer
ChatGPT Search provides a sourced starting point for online research. Important claims still need direct checks against their cited webpages.
- Open the citation connected with each important claim.
- Find the exact passage supporting the stated information.
- Check the publication date, author, and source owner.
- Prefer official documents for product, policy, or legal information.
- Compare additional sources when evidence conflicts or remains incomplete.
Search results can vary across prompts, dates, locations, and accounts. Different wording may produce different targeted queries and cited sources. A citation may also omit context found elsewhere on its webpage.
OpenAI does not guarantee top placement for any public webpage. Crawler permission creates search eligibility without guaranteeing answer inclusion. Readers should verify changing facts again before using them.
Manish Singh is the Team Lead at IMMWIT, where he brings over 14 years of experience in SEO, UX, and digital marketing. Known for helping businesses rank, scale, and grow smarter online, he blends strategic thinking with AI and NLP-backed insights. His hands-on approach to semantic SEO and UX design turns ideas into real results clients can see and trust.