What AI search means
AI search is any search experience where the primary result is a generated answer rather than a ranked list of links. The user asks a question in natural language; a large language model — usually combined with live web retrieval — reads relevant sources and writes a synthesized response, often with citations. The ten blue links haven't disappeared, but in AI search they've been demoted to raw material: inputs the engine reads so the user doesn't have to.
The category spans two kinds of product that arrived from opposite directions. Chat assistants (ChatGPT, Claude, Gemini) added web search to conversation; search engines added generation to results (Google AI Overviews and AI Mode, Perplexity as a purpose-built hybrid). They've converged on the same experience: one question, one composed answer.
The major AI search engines
- ChatGPT — the largest consumer AI assistant; browses the web for current topics and cites sources when it does.
- Google Gemini and AI Overviews — Gemini powers Google's assistant and the AI-generated summaries that now sit above traditional results (AI Overviews).
- Perplexity — retrieval-first by design; nearly every sentence carries a numbered citation.
- Claude — Anthropic's assistant, with web search grounding answers about current topics.
These four span the architectural spectrum — from memory-heavy to retrieval-heavy — which is why tracking all of them matters: they frequently disagree about the same brand.
How AI search actually works
Under the hood, most AI search runs on retrieval-augmented generation: the engine expands the user's question into background searches (see query fan-out), retrieves top results from a conventional search index, and generates its answer from what it read — blended with what the model already knows from training data. This layered architecture is why classic SEO still feeds AI search, and why it no longer suffices on its own: ranking gets you retrieved, but the model decides what gets written into the answer.
Why AI search changes brand discovery
In generative search, the answer is the whole result. Three consequences follow for brands:
- The consideration set is written for the buyer. Instead of the user compiling a shortlist from ten links, the engine hands them one — typically three to five names. Being outside it means being invisible, with no "page two" to be found on.
- Most interactions are zero-click, so the recommendation is consumed without any measurable site visit.
- Visibility becomes probabilistic. The same question can yield different brand lists run to run, which is why measurement uses repeated sampling — scheduled prompt tracking across engines rather than one-off spot checks.
Optimizing for this surface is the discipline of generative engine optimization: understanding which questions your buyers ask AI search, measuring whether the answers include you, and working the sources engines read until they do.