Google AI Mode

What did Google actually announce?

On 20 May 2025 Google described AI Mode as using the query fan-out technique, “breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf”. Its documentation positions AI Mode as “particularly helpful for queries where further exploration, reasoning, or complex comparisons are needed”, and its longer-running Deep Search mode “can issue hundreds of searches” for one question.

Two things follow from that description and neither is obvious from the interface. The answer is assembled from several searches rather than one, so there is no single query a page competes on. And the surface is conversational, so a second question carries the first one’s context, which no keyword tool models and no rank tracker can represent.

Why do AI Mode citations ignore the organic top ten?

Because the searches being answered are not the search that was typed. Moz’s study of 40,000 queries found that 88% of AI Mode citations do not match the organic top 10 for the same query, and that only about one citation in five comes from a domain that appears in that top ten at all.

That is a genuinely different competitive shape rather than a rounding error, and it has one clear consequence: a page that answers a narrow sub-question completely can be cited on a query it could never rank for. Ranking first for the head term is neither required nor sufficient. It cuts the other way too, and this is the part nobody selling AI Mode optimisation mentions: a page that ranks first can be absent from the answer to the query it ranks for, with nothing wrong with it and nothing to fix.

How do you measure AI Mode, honestly?

With difficulty, and with less precision than any vendor will admit. Google reports AI Mode traffic inside Search Console’s Web search type with no separate breakdown, so the only defensible figures are the ones you gather yourself: whether a query returns an answer naming you, run repeatedly over a panel of prompts rather than once.

One check is worth nothing here. Repeat the same prompt and the answer changes, which is measured rather than folklore, and it is the reason a citation rate across many prompts is the smallest unit worth reporting. The second reason to distrust a single check is that AI Mode is personalised and location-aware, so two people running the same query on the same day are not running the same test. A panel run from one place, repeatedly, at least holds those variables still.

Related terms

• AI OverviewAn AI Overview is the written summary Google places above its ordinary search results, assembled from web pages and shown with links to a few of them.
• Query Fan-OutQuery fan-out is Google’s own term for how AI Overviews and AI Mode answer a question: the engine breaks the question into subtopics and issues many searches at once, then writes one answer from the results.
• Answer EngineAn answer engine is a search product that returns a written answer rather than a ranked list of links, citing a handful of sources inside it.
• Citation RateCitation rate is the share of tracked AI answers that link your site as a source, counted over the answers where a source list was actually returned.