AI Engines

Deep dives into AI search engines: how each one retrieves, ranks, and cites — and how to optimize for it.

• ChatGPT Search

  ChatGPT Search is retrieval-augmented chat: ChatGPT fetches the live web only when the model decides to, so citations are sparser than Perplexity's. Its edge is distribution scale. The load-bearing GEO fact: OAI-SearchBot — not GPTBot — controls Search visibility.

• Claude

  Claude adds live web retrieval to its chat, Research, and API surfaces. Search is conditional, citations appear when web results support an answer, and Anthropic uses three separate agents for potential training collection, search indexing, and user-requested fetching.

• DeepSeek

  DeepSeek combines an official chat product with optional web search, open-weight models, and a separate API that now supports server-side search. Its crawler identity and ranking logic remain undisclosed, so consumer, API, and self-hosted results must be measured separately.

• Google AI Overviews

  Google AI Overviews is the SERP-embedded engine: an AI summary block atop classic Google Search, grounded on the normal web index via query fan-out. The load-bearing GEO fact — there is no AI Overviews crawler. Googlebot controls eligibility; Google-Extended does not.

• Google Gemini

  Gemini is the Google-stack retrieval-augmented-chat engine: a chat model given a Grounding-with-Google-Search tool, conditional not SERP-embedded. It shares AI Overviews' Google index, but the GEO fact is inverted: Google-Extended is the control here.

• Microsoft Bing Copilot

  Copilot Search in Bing grounds its answers on Bing search results. Bing indexing determines whether a page enters the web candidate pool. Retrieval and citation selection determine whether it appears in a particular answer.

• Perplexity AI

  Perplexity is an answer-engine-native generative engine: it retrieves the live web by default and ships every answer with numbered inline citations — the most citation-dense, transparent mainstream engine, and the live-engine baseline GEO research and field tests use most.