Definition
Generative AI search is search in which a language model writes the answer. The system retrieves relevant documents, synthesizes them into a response in natural language, and cites some of the sources, rather than returning a ranked list of links for the user to read. Google AI Overviews and AI Mode, ChatGPT search, Perplexity, Microsoft Copilot, Gemini and Grok all work this way.
The shift happened in a few years. Google tested generative answers as the Search Generative Experience from May 2023, renamed them AI Overviews in May 2024, and launched the conversational AI Mode in the US in May 2025. By Google I/O in May 2026, Google said AI Overviews had more than 2.5 billion monthly active users and AI Mode more than 1 billion. OpenAI said in October 2026 that more than 1.2 billion people used ChatGPT each week. Generative answers are now the default experience for a large share of informational queries, not an add-on.
Most systems follow the same pipeline, usually some form of retrieval-augmented generation:
- The model interprets the question and often splits it into several sub-queries through query fan-out.
- A search layer retrieves candidate pages or passages, from the vendor's own index (Google, Bing) or from a partner or proprietary crawler.
- Results are reranked, and the most useful passages are passed into the model's context.
- The model writes an answer, attaching source citations to some claims.
- The user can ask follow-ups, which start the loop again with the conversation as context.
The model's training data also shapes the answer, especially which brands it already associates with a topic and what it chooses to search for.
This changes what visibility means. In a list of links, position decides attention. In a generated answer, the unit of visibility is inclusion: is your brand named, where in the answer, in what tone, and is your page one of the cited sources? Those are separate outcomes. A brand can be recommended without being cited, or cited without being named, which is why GEO measurement tracks mentions and citations independently. Clicks change too. Google says its AI features send billions of clicks to websites every week, but many questions are now answered on the results page itself, so zero-click search metrics and answer-level tracking matter more than before.
The discipline of earning visibility in these systems is generative engine optimization. It builds on SEO, since retrieval still depends on crawlable, indexable pages, but adds work specific to generation: writing passages that answer one question clearly, keeping facts consistent across your site and third-party sources so the model does not see conflicting claims, and covering the sub-questions that fan-out generates. Each platform retrieves differently, so the same prompt can cite completely different sources on ChatGPT, AI Mode and Perplexity. Our study of average sources per response shows how widely citation counts vary between platforms.
Examples of Generative AI Search
- A user asks Google AI Mode 'which heat pump works best in a cold climate with an old radiator system' and gets a three-paragraph answer citing a manufacturer's spec page, an energy agency guide and a Reddit thread.
- A B2B brand finds ChatGPT names it in 30% of answers about its category but cites its website in only 6%, because most answers rely on review sites for the supporting links.
- A publisher sees impressions in Search Console's generative AI report rise while clicks from the same pages fall, a typical pattern when AI Overviews answer the question directly.
- A healthcare provider compares the same symptom question across ChatGPT, Perplexity and AI Overviews and finds no overlap in cited sources, so it tracks each platform separately.
