Google AI Mode Explained: How the New Search Names People

For twenty-five years, Google answered a question by handing you a reading list. Now it increasingly answers by just answering, and when the question is "who should I hire for this?", the reply contains names. Whose names, and chosen how? That is the question worth sitting with.

What is Google AI Mode?

AI Mode is a conversational search experience built directly into Google Search and powered by Gemini. Instead of the familiar page of blue links, it generates a written answer, with links to supporting sources, and invites you to keep going: refine, compare, ask the follow-up you would ask a knowledgeable friend. Google introduced it as a Labs experiment in March 2025, describing it as its most powerful AI search, then rolled it out to everyone in the US at I/O in May 2025.

The uptake was not subtle. By I/O 2026, Google reported that AI Mode had surpassed one billion monthly users, with queries more than doubling every quarter since launch, and made Gemini 3.5 Flash the default model for everyone globally. Whatever one thinks of AI answers, the behavioural verdict is in: given the choice between a reading list and a conversation, a billion people chose the conversation.

Access is deliberately frictionless. AI Mode sits as a tab inside ordinary Google Search and the Google app, and the redesigned AI search box is rolling out in every country and language where AI Mode operates. There is no separate destination to remember and nothing to install, which is precisely the point: Google is not asking users to change their habit, it is changing what the existing habit returns.

It matters for our purposes because AI Mode is not a chatbot bolted onto the side of search. It is search, from the company that still owns the world's question pipeline. When it names a person in an answer, that is the closest thing the modern web has to being pointed at.

How is AI Mode different from AI Overviews?

The two get conflated constantly, and the distinction matters if you are trying to appear in either. We covered the older feature in our AI Overviews explainer; here is the side-by-side.

Google has also stitched them together: ask a follow-up from an AI Overview and you slide straight into AI Mode with your context intact. The summary box is the doorway; the conversation is the destination.

What is query fan-out and why does it matter for your name?

Here is the mechanism that changes the game. When you ask AI Mode something, it does not run your question as one search. Google describes a query fan-out technique: the system breaks your question into subtopics and issues a multitude of searches simultaneously, across the web and Google's own data, then synthesises what comes back into a single answer.

Consider what that does to a people-question. "Who is a good employment lawyer for a startup in Leeds?" quietly becomes a small research project: employment law specialists, startup experience, Leeds, reviews and reputation, recent commentary on the relevant regulations, perhaps fee structures. A person who exists strongly in three of those hidden searches and not at all in the others is a weaker candidate than someone who appears, modestly but consistently, across all of them.

Under the old search, you optimised a page for a query. Under fan-out, your name is being assembled from fragments gathered by searches you never see. The strategic consequence is almost philosophical: you are no longer trying to rank for a question. You are trying to be the most coherent answer across a dozen questions at once.

Fan-out also quietly rehabilitates the narrow question. A specialist who has answered twenty small, unglamorous questions in their field, each on its own clear page, holds twenty tickets in every draw the fan-out runs; the generalist with one impressive homepage holds one. So the practical advice that follows from AI Mode is less about writing the definitive guide and more about leaving no reasonable sub-question of your specialty unanswered in your own words. The machine is asking around behind the scenes, and the winners are already wherever it asks.

The research-assistant analogy, and what it means for you

Query fan-out is easier to hold in your head with an analogy than with the engineering description. Picture a very thorough research assistant, the kind who, when you ask a simple question, does not just glance at one source and report back. Instead, they quietly go and ask several different, specific follow-up questions to several different places, cross-check what comes back, notice where the answers agree, and only then walk into your office with one clean, combined answer. You never see the legwork. You only see the tidy paragraph at the end. That is precisely the difference between an old-style single lookup and what AI Mode is doing every time someone types a question into it: it is not consulting one shelf, it is dispatching a small team of researchers who each check a different angle before anyone reports back.

Imagine someone types "best divorce mediator for a high-conflict case near Bristol" into AI Mode. The visible question is one sentence. The invisible research assistant, though, is quietly running separate checks: who practices mediation near Bristol, who among them has specific experience with high-conflict cases, who holds recognized accreditation, what recent reviews and reputation look like, and roughly what the process and cost involve. Each of those is its own small search, happening in the background, and the person or people who keep showing up, consistently and specifically, across the most of those separate checks are the ones who make it into the final paragraph the user actually reads.

What this means practically, for anyone hoping to be one of the names that surfaces, is a shift in what "covering your topic" actually requires. It is not enough to have one strong, general page that impresses a human reader who lands on it directly, the way the next generation increasingly searches by asking a full question rather than typing keywords. You need your specific expertise answered in several separate, specific places, because the research assistant is checking several separate, specific things. A family lawyer who has written distinctly about high-conflict mediation, about their credentials, and about their process and typical costs, each covered clearly somewhere in their own published record, gives the invisible research assistant several places to find them saying the same consistent thing. A lawyer with only a polished "About Us" paragraph gives it exactly one chance, and that chance may not even match the sub-question being asked. This is also why understanding the actual prompts buyers type into AI matters more than guessing at keywords: the sub-questions fan-out generates tend to mirror the real, specific concerns a buyer has, not the generic categories a directory might use.

A single AI Mode question is actually several hidden searches, quietly combined into one answer, which is also the logic behind answer engine optimization as a discipline.

What has AI Mode become in 2026?

The 2026 announcements pushed it well past question-and-answer. As of I/O 2026, AI Mode runs Gemini 3.5 Flash by default worldwide, and around it Google introduced a redesigned, AI-powered search box, Search agents, and expanded Personal Intelligence.

Each of these deepens the same trend. Search agents are background researchers: you describe what you want to stay informed about, and an agent monitors the web continuously, sending synthesised updates. Personal Intelligence, an opt-in, lets AI Mode draw on your Gmail and Google Photos for context, in nearly 200 countries. Generative interfaces let Search build custom visual layouts and small tools on the fly. Agentic features let it check availability and set up bookings with local providers.

Read those together and a pattern emerges: the human is delegating more of the research, and sometimes all of it. When an agent works in the background for weeks and returns a recommendation, there is no results page at all. There is only the conclusion, and whoever is named in it. If you have wondered whether AI search is killing Google, the more precise observation is that Google is killing the results page itself, on its own schedule, and replacing it with answers.

How does AI Mode decide which people to name?

Google publishes mechanisms, not recipes, so some of this is inference from how the system demonstrably works. But three requirements follow directly from the architecture.

First, you must survive entity resolution. Fan-out gathers fragments from many searches; the synthesis step has to decide whether the consultant on one site, the speaker on another and the author on a third are the same person. If your name, title and bio drift across the web, you arrive at the final answer as three weak candidates instead of one strong one, the failure we dissected in Entity, Not Ego.

Second, you need presence on the sources the sub-queries actually reach: professional directories, press, review platforms, publications with authority in your field. A beautiful homepage covers one fragment; the other eleven searches went elsewhere.

Third, you need statements a language model can lift and defend. Concrete claims, verifiable credentials, specifics with dates and outcomes. Fan-out is a fact-gathering exercise, and vague positioning gives it nothing to gather. We trace the whole selection pipeline step by step in Anatomy of an AI Answer.

There is a fourth advantage worth naming: Google has been building machinery for understanding people longer than anyone, through its Knowledge Graph and its long investment in structured data. Marking your site up so it states, in machine-readable terms, who you are, what you do and which profiles belong to you is not busywork; it feeds the one company whose people-database predates the chatbot era. AI Mode did not start learning who you are in 2025. It inherited everything Google already believed, which makes correcting and completing that inherited record part of the job.

What should you actually do about AI Mode?

A billion-user channel that names people deserves a deliberate response rather than dread. A practical sequence:

1. Ask it about yourself. Open AI Mode, ask what it knows about your name, then ask for recommendations in your specialty. You are reading your machine reputation, unfiltered.
2. Map the fan-out. Write down the six to ten sub-questions a buyer's query about your field would decompose into. That list is your visibility syllabus.
3. Fill your gaps deliberately. For each sub-question where you are absent, decide the fix: a published piece, a directory profile, a third-party mention.
4. Unify your entity. One spelling of your name, one title, one bio, everywhere, so the synthesis step can assemble you confidently.
5. Recheck quarterly. AI Mode's sources and models move fast; treat the check-up as recurring hygiene, not a one-off audit.

None of this is exotic. It is the ordinary discipline of being legible, applied to a reader that happens to run twelve searches per question. If you would rather have the audit and the gap-map done for you, that is precisely what our services are built around, and the rest of the journal covers each signal in depth.