ChatGPT Search Visibility: Crawler Controls and Measurement

By ClickRadius · Published April 14, 2026

OpenAI publishes access roles for three named user agents. That documentation is useful, but it does not disclose how ChatGPT ranks sources or chooses a displayed link. This guide keeps the documented controls, publisher-side checks, sampled answers, and unknown provider decisions separate.

OpenAI evidence boundary. OpenAI's public crawler documentation defines access roles for OAI-SearchBot, ChatGPT-User, and GPTBot. It does not publish source-ranking weights, a passage-selection formula, a Bing dependency, recrawl-to-citation timing, or a 60–90-day result window. No eligible closed ClickRadius customer-month cohort establishes a typical ChatGPT source-link or citation outcome.

What OpenAI documents

The current OpenAI crawler documentation separates the three roles:

1. OAI-SearchBot is used to surface websites in ChatGPT search results. OpenAI says a site that opts out will not be shown in ChatGPT search answers, although it may still appear as a navigational link. Publishers control its automatic crawl through robots.txt and can use OpenAI's published IP ranges for network rules.
2. ChatGPT-User supports visits initiated by a user. OpenAI says it is not used for automatic web crawling or to determine whether content appears in Search, and that robots.txt rules may not apply to these user-initiated visits.
3. GPTBot is the training crawler. A publisher's GPTBot decision is independent of the OAI-SearchBot control used for ChatGPT search visibility.

These are access roles, not a source-ranking specification. Allowing OAI-SearchBot can preserve the documented Search-appearance route, but it does not guarantee a crawl, index state, answer appearance, recommendation, or displayed source link.

Audit access as a chain of observations

1. Policy. Record the exact robots.txt response and effective rule for OAI-SearchBot. Review GPTBot separately as a training-policy decision. Do not treat ChatGPT-User as the automatic Search crawler.
2. Network handling. Compare CDN and firewall rules with OpenAI's published IP ranges. A user-agent string by itself is not authentication.
3. Public response. Through an authorized check, preserve the requested URL, status, redirects, content type, and public body. OpenAI's crawler page does not publish a JavaScript-rendering contract, so report what was served rather than declaring client-rendered content retrievable or invisible.
4. Server evidence. A verified log entry can show that a request reached the origin and what response the origin recorded. It does not prove later indexing, retrieval, ranking, or source selection.

Make the publisher's evidence checkable

Content review is still worthwhile when its objective is accurate, inspectable publishing rather than a promised ChatGPT outcome:

• Answer the page's stated question clearly and keep material qualifications beside the claim they limit.
• For a statistic, state the source, measured population, method, geography, collection date, and limitations when applicable.
• Identify the accountable author or publisher and provide a public path to inspect cited sources.
• Use Article, FAQPage, Organization, or other structured data only when applicable and consistent with visible content. OpenAI does not publish a rule on this page assigning those schema types a ChatGPT citation weight.
• Correct unsupported superlatives and factual conflicts for readers. Measure any later provider outcome separately.

What the GEO study does and does not establish

The Princeton-led GEO: Generative Engine Optimization study, presented at KDD 2024, tested nine content interventions in a bounded benchmark. It began with the top five Google sources, randomly selected one source for optimization, and measured that source's relative visibility or impression in generated responses. Interventions including quotations, statistics, and source citations produced reported gains of up to about 40% in some settings.

The study did not measure current ChatGPT retrieval probability or citation-inclusion probability, and effects varied by domain. It is evidence about that benchmark, not a private OpenAI selection rule. Use it to form a disclosed editorial hypothesis, then measure the named provider result without relabeling the hypothesis as a guarantee.

Keep off-site evidence within its authority

Directories, business profiles, official records, reviews, and third-party coverage are controlled outside the website. Inventory each source's controller, subject, claim scope, public URL, and review date. Preserve legitimate location, formatting, and historical differences; correct actual conflicts only through an authorized workflow and verify the public result.

That inventory can establish what the public sources say. It cannot expose ChatGPT's authority assessment or prove that a profile, review, or article caused a displayed source link.

Measure ChatGPT Search as a named surface

One answer is one observation. A repeatable protocol should record:

1. The fixed prompt and the classification rules for a brand mention, displayed source link, and recommendation.
2. The ChatGPT product surface, exposed model label, date, locale, and relevant account or device state.
3. The returned wording and exact displayed source URLs.
4. A completed no-link result separately from provider failure, unsupported work, or unavailable evidence.
5. Site-change authority, an exact change receipt, and public readback before comparing later samples.

ClickRadius records a monthly AI Presence sample across ChatGPT, Gemini, Perplexity, Claude, and Grok for stored queries. Copilot is not supported. Unavailable provider evidence remains unavailable. The separate five-family AI Readiness Score estimates readiness inputs observed at the submitted URL; it does not predict the AI Presence result.

Interpret logs and zeros narrowly

• A robots.txt allowance records publisher policy; it is not evidence that OAI-SearchBot requested the page.
• A verified request log records transport to the origin and the recorded response; it is not evidence of indexing or use in an answer.
• A completed zero records no qualifying mention or displayed source link under the declared sample rules. It does not diagnose access, content, authority, or timing by itself.
• A difference between providers or months is descriptive until the protocol addresses prompt, surface, provider, site, and other changes. It does not localize a private failure stage on its own.

An eight-week evidence-collection plan, not a result window

Weeks 1–2: Establish the access record

• Read and date OpenAI's crawler documentation. Record separate decisions for OAI-SearchBot and GPTBot.
• Verify the effective robots.txt, CDN, and firewall state through authorized checks using the documented identities and published IP guidance.
• Capture the public response for priority URLs, including status, redirects, content type, and visible claim readback.

Weeks 3–6: Repair publisher-controlled evidence

• Choose pages from observed business questions and factual gaps, not an invented citation quota.
• Make claims specific, sourced, dated, scoped, and visibly qualified.
• Reconcile visible content and applicable structured data. If a tool is used, require current authorization, proven connection capability, an exact effect receipt, and public readback.

Weeks 7–8: Freeze the measurement protocol

• Predefine the prompt set, surface context, classifications, and unavailable states.
• Record the monthly ChatGPT observation alongside the other supported providers without treating cross-provider differences as a diagnosis.
• Retain a dated baseline for later comparison. The eight-week sequence is an operating framework, not evidence that a source link should appear by its end.

Frequently asked questions

What does OAI-SearchBot control?

OpenAI documents OAI-SearchBot as the crawler used to surface websites in ChatGPT search results. A publisher can control its automatic crawl through robots.txt and should use OpenAI's published IP ranges when configuring network rules. Access does not guarantee that a page will appear or receive a displayed source link.

Do I need to allow ChatGPT-User in robots.txt for Search visibility?

OpenAI says ChatGPT-User supports user-initiated visits, is not used for automatic web crawling, and is not used to determine whether content appears in Search. Because the visit is user initiated, robots.txt rules may not apply. OAI-SearchBot is the documented control for automatic Search crawling.

How long after a change should I expect a ChatGPT source link?

OpenAI's crawler documentation publishes no recrawl-to-citation result window, and no eligible closed ClickRadius customer-month cohort supplies one. Record the change, public readback, provider, surface, prompt, answer, source URLs, and observation date without turning an editorial schedule into a promise.

Want a bounded site baseline? Get your free AI Readiness Score — a submitted-URL estimate across five readiness evidence families. It is separate from measured AI Presence and does not reveal ChatGPT's private selection process or predict a source link. You can also review the current plans and enrollment status.