VectorCiteVectorCite
OpenAI

ChatGPT Search

OpenAI's web-retrieval-enabled chat surface — the largest AEO target by user volume.

How it works

ChatGPT Search is the web-search-enabled variant of ChatGPT, launched 2024. When a query requires fresh information, the model invokes its web_search tool, retrieves ~5–8 candidate pages via Bing's index + OpenAI's own crawler, then synthesizes an answer with inline citations. The retrieval is hybrid (dense embeddings + BM25 + authority). Citations are presented as numbered footnotes [1][2] that expand to source cards. Not every ChatGPT query triggers retrieval — the model decides per turn based on query freshness needs.

Official docs
User-agents
GPTBotOAI-SearchBotChatGPT-User

Allow these in your robots.txt explicitly. Audit your current setup with the robots.txt analyzer.

Signals it weights heavily
  • JSON-LD schema
  • Author E-E-A-T
  • Citation density
  • Freshness for time-sensitive queries
  • Open Graph metadata for source cards
How to optimize for ChatGPT Search
  • Allow GPTBot + OAI-SearchBot in robots.txt explicitly (not just defaults).
  • Add og:image + og:title — source cards in ChatGPT render visually.
  • BLUF-format your top pages — ChatGPT prefers direct-answer leads.
  • Strong Author + Date markup signals freshness for time-sensitive queries.

See your score on ChatGPT Search's signals.

Run free audit

Other engines