AI Mode

AI Mode is built for questions that used to take several searches. Google describes it as most helpful where exploration, reasoning or complex comparisons are needed^[4], and launched it on a custom version of Gemini 2.0 that makes a plan, runs searches and adjusts the plan on what it finds^[1]. It draws on web pages and on Google’s own systems, the Knowledge Graph and shopping data among them.

It differs from AI Overviews in where it sits and how it is used. An overview appears on a classic results page only when Google judges it adds something; AI Mode is a mode the searcher chooses, and the answer continues as a conversation. Google says the two may use different models and techniques, so their answers and links can differ for the same query^[4]. Google says both may use query fan-out^[4], which in AI Mode breaks a question into subtopics and searches each one^[2].

Google also says when AI Mode steps back: where it lacks confidence in the quality or helpfulness of an AI response, it may show a set of web links instead^[3]. The links are the constant, and they follow the same eligibility rules as the links of an overview.

Access to AI Mode is set by the robots.txt rules for Googlebot, not by a token of its own^[4]. In QueryRev’s study, 1,339 of the 6,294 sites read, 21.3%, write a robots.txt rule for at least one of the AI crawlers covered^[7]; none of those tokens reaches AI Mode, which reads pages through Googlebot.

Search Console counts AI Mode with the standard rules: a click on a link to an external page is a click, position follows the method of a results page, and a follow-up question is a new query with data of its own^[5]. Google reports it together with the rest of Web search, so AI Mode performance is read through the queries it produces, which tend to be longer and phrased as questions.

In Google Analytics 4 a visit from AI Mode is Organic Search, not part of the AI Assistant channel^[6]; the difference is set out under AI referral traffic.

The answer itself has to be observed: ask the tracked questions in AI Mode on a schedule, with the follow-ups a buyer would ask next, and record the supporting links and the brands named in each turn.

Sample data: an invented site, to show the method.

A bed retailer asks AI Mode to compare oak and walnut frames for a small bedroom. The response weighs hardness, colour and price, and links three pages: two review sites and the retailer’s own guide to wood types, cited for the durability point.

The follow-up, which is easier to look after, is a new query in Search Console, with impressions and clicks of its own. The retailer’s guide is not linked in that second answer, so its section on care is the passage to strengthen.