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.

Published 2026-06-19
·
Updated 2026-07-22
·
1 min read

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:

  1. Discovery: OAI-SearchBot or another discovery source learns the URL.
  2. Fetchability: robots.txt, the CDN, WAF, status code, and rendered response allow access.
  3. Eligibility: canonical and index controls do not exclude the intended URL.
  4. 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-SearchBot on the pages intended for discovery.
  • Keep canonical URLs indexable and return a stable 200 response.
  • 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.

Privacy & Cookies

We use cookies to enhance your experience. By continuing to visit this site you agree to our use of cookies.