AI Mode (Google)

Definition : Google's conversational search tab, where the user holds a dialogue with the AI instead of scanning a list of links. Live in France since 22 July 2026.

AI Mode is a full search tab in its own right, sitting alongside Images, News and Maps. You ask a question in plain language, you get a written answer with a few sources, and above all you can follow up: refine, correct, dig deeper across one continuous exchange. In practice it is a ChatGPT installed inside Google. It is also reachable directly at google.com/aimode, while signed in to a Google account.

It is routinely confused with AI Overviews, though the difference is clean. An AI Overview is imposed: a summary that appears by itself at the top of the ordinary results page, unrequested. AI Mode is chosen: you decide to go in. The first changes everyone's behaviour immediately; the second builds a habit more slowly, among those who adopt it.

Technically, AI Mode runs on a mechanism called query fan-out. Rather than looking for the page that answers, it breaks the question into several sub-queries which it runs simultaneously against Google's index, then synthesises one sourced answer. Analyses published in 2026 commonly observe 8 to 12 sub-queries for an ordinary question; the Deep Search feature can trigger several hundred. Google says this relies on a custom version of Gemini.

The consequence for a website is substantial, and it is the point most coverage omits: the unit selected is no longer the page, it is the passage. AI Mode picks fragments that each answer one sub-question. An excellent page whose useful fact is buried in a long paragraph loses to an average page that answers in two clean sentences under an explicit heading. I have detailed the mechanism and the practical levers in this article on query fan-out.

Common mistakes

  • Confusing AI Mode and AI Overviews: one is a tab you deliberately enter, the other a panel imposed above the results. The optimisations overlap, but the traffic stakes differ.
  • Still targeting one keyword per page: fan-out judges your page across a cluster of sub-questions you cannot see. Completeness matters, not repetition of a phrase.
  • Splitting a site into thin pages: one page per sub-question produces exactly the generic content models discard. One solid page, cut into clean sections, works better.
  • Neglecting SEO: AI Mode draws on Google's index. A page absent from that index is a candidate for no sub-query at all.

And for your business, concretely.