ChatGPT search: discovery, citations, and measurement
ChatGPT search can cite public web pages. Learn the role of OAI-SearchBot, the limits of crawler access, and a defensible measurement workflow.
ChatGPT Search
ChatGPT search is OpenAI’s web-search experience inside ChatGPT. It can produce a synthesized response with linked sources, but a page’s presence depends on more than one crawler request.
OpenAI tells publishers not to block OAI-SearchBot if they want public content to be discoverable, surfaced, cited, and linked. GPTBot has a different purpose: content that may be used to improve foundation models. Allowing GPTBot is therefore not a requirement for ChatGPT search visibility.
The technical access path
A useful audit separates four stages:
- Discovery: OAI-SearchBot or another discovery source learns the URL.
- Fetchability: robots.txt, the CDN, WAF, status code, and rendered response allow access.
- Eligibility: canonical and index controls do not exclude the intended URL.
- Selection: the system decides whether the page helps answer a particular query and deserves a citation.
Passing one stage does not prove the next. Structured data can clarify entities and page meaning, but OpenAI does not document a special ChatGPT-search schema requirement.
Publisher checklist
- Allow
OAI-SearchBoton the pages intended for discovery. - Keep canonical URLs indexable and return a stable
200response. - Make the core answer available in rendered HTML without requiring interaction.
- Identify authors, dates, evidence, and primary sources where they matter.
- Use specific passages, tables, definitions, and constraints that can support an answer.
- Avoid publishing near-duplicate pages for small query variations.
- Verify edge access with current OpenAI IP information where appropriate.
If a page must not appear, use noindex and permit the crawler to read it. OpenAI notes that a disallowed URL may still be surfaced as a link and title in some circumstances; blocking crawl is not equivalent to removing an URL from every surface.
Measurement without overclaiming
Track three different signals:
- validated bot requests in server or CDN logs;
- repeatable citation checks for a fixed prompt set;
- referral sessions and conversions from ChatGPT properties.
Record prompt, locale, account state, date, cited URL, and answer position. Results can vary, so a single manual prompt is not a stable ranking report. Use share of model and AI referral traffic as trend metrics, not proof that a specific edit caused selection.
Primary source
See also: GPTBot, AI citations, and source-of-truth content.