AI Search

AI search is a way of answering search queries in which a large language model retrieves web pages or other sources, reads them, and writes a direct answer, usually with links to the sources it used. It replaces or sits above the ranked list of links that defined web search for 25 years. ChatGPT search, Google AI Overviews and AI Mode, Perplexity, Microsoft Copilot, Gemini and Claude with web search all work this way.

Most AI search systems follow the same pattern, a form of retrieval-augmented generation:

• The model interprets the question and decides whether it needs fresh information or can answer from training data.
• If it searches, it often splits the question into several sub-queries, a step called query fan-out, and runs them against a search index.
• It reads the retrieved pages, often at the passage level, and selects the material that supports an answer.
• It writes one response, names products or brands where relevant, and attaches citations.

The index behind each step differs by provider. Google uses its own search index; Perplexity says its index tracks more than 200 billion URLs; ChatGPT relies on OpenAI's OAI-SearchBot crawler plus third-party search providers; and Meta has been crawling heavily to build its own. That is one reason the same question can produce different sources on different platforms. See AI search index.

Usage reached a large scale in 2026. OpenAI said in August 2026 that ChatGPT had more than 1 billion weekly active users. Google said in May 2026 that AI Overviews had more than 2.5 billion monthly users and AI Mode more than 1 billion, and Alphabet reported the Gemini app at 950 million monthly active users in July 2026. Google also said its AI features were sending billions of clicks to websites every week.

The outcome for a website is different from classic search in three ways. First, visibility is split into two things that can move independently: being mentioned in the answer text and being cited as a source. Second, fewer users click: Pew Research Center found Google users clicked a regular result on 8% of visits with an AI summary versus 15% without one in March 2025, which feeds zero-click search. Third, answers vary by platform, country, phrasing and even run to run, so a single check says little. Promptwatch's data on average sources per response shows how differently the engines cite.

For teams working on GEO, AI search changes the unit of work from keywords to prompts and from rankings to answers. The practical questions are whether AI crawlers can reach your pages, whether those pages answer the sub-questions buyers ask in a form a model can quote, whether third-party sources describe your brand consistently, and how often each platform names and cites you compared with competitors. Google's own guidance for its AI features is that standard SEO fundamentals still apply and no special AI markup is required; the other platforms publish less, so measurement fills the gap.

For the products themselves and how they differ, see AI-powered search tools.