Google AI Mode.

Google AI Mode is a separate mode inside Google Search in which a Gemini model writes the answer instead of returning a list of results. It is conversational: once the first answer appears, you can ask follow-up questions without rephrasing the search. The difference to AI Overviews is one of depth. AI Overviews sit in a box above the classic results, which stay in place underneath. AI Mode replaces the result list altogether and draws on far more sources to do it. For a website that shifts the goal. Position one matters less than whether your page is used as evidence inside the generated answer.

Technically, AI Mode runs on query fan-out. The model breaks the input into sub-questions, side conditions and likely follow-ups, then sends all of them to the search index at the same time. A single input typically triggers anywhere from a handful to well over a dozen individual queries. The model then weighs all returned passages together and writes flowing text with source links. What counts is therefore not whether a page ranks for the literal input, but whether it supplies solid material for as many of the derived sub-questions as possible. Pages that answer one question completely, in clearly separated sections, have a structural advantage here.

AI Mode queries show up in Google Search Console as ordinary search queries, usually as fully written-out questions. On dlm-digital.ch that includes lines such as «welches ist die beste geo-agentur in zürich?» with 147 impressions at position 10.9, or «welche geo-spezialisten kannst für unternehmen in zürich empfehlen?» with 133 impressions at position 8.7. Those rows are the practical proof that conversational search creates real demand. They are also the best working material available: once you know how people phrase the question in Search Console, you have a fair idea of which sub-questions the fan-out will ask.

A concrete case. A fiduciary office on Lake Zurich wants to appear in AI Mode when someone asks who supports small firms with their VAT returns. A general services page will not do it. What is needed is a page that answers scope, process, price range and regional coverage each in its own headed section, because those sections are exactly the units the fan-out collects. Clean Organization data as JSON-LD belongs with it, so the model can identify the provider unambiguously. Visibility in AI Mode is therefore not a new trick. It is editorial clarity taken to its logical end.