What is AI search?

AI search means asking an AI assistant a question and receiving one synthesised answer rather than a list of links. The assistant reads or recalls multiple sources, combines them, and replies in prose, often with a handful of citations. The main surfaces today are ChatGPT, Claude, Perplexity, Google's Gemini, Google AI Overviews and AI Mode inside classic search results, and Microsoft Copilot in Bing and Windows.

The mechanics differ from classic search in two ways. First, the answer can come from the model's training data alone, with no live lookup, which means it can be months out of date or simply wrong. Second, when the assistant does search the web, a process called retrieval-augmented generation, it fetches a small set of pages, extracts passages, and writes its own summary. A page ranked tenth in Google can still pick up occasional clicks, but if the assistant retrieves only a few pages and yours is not among them, nothing from your site reaches the reader.

For brands, the practical consequence is that the assistant writes the first description a prospective buyer sees. Assistants lean on sources they already trust, such as Wikipedia, Reddit threads, G2 reviews, and established trade publications, so a brand with a thin footprint on those surfaces gets described from stale or secondhand information. In our July 2026 study, an AI assistant answering from memory materially misdescribed 27 of 30 brands, and 13 of 30 were still misdescribed when live web search was switched on.

The discipline of improving how assistants find and describe you is called Generative Engine Optimization. The starting point is measurement: ask each assistant what it says about your brand, your category, and your competitors, then record where the answers are wrong or missing. Tools automate this; Discoverable, for example, offers a free AI visibility checker that runs a snapshot with no login, with Claude checks live today and other engines rolling out.