Google AI Mode and SEO: Query Fan-Out, and What It Actually Changes

Most explanations of Google AI Mode are written in one of two registers. Either it is the end of search as a traffic channel, or it is a gimmick nobody uses. The evidence supports neither, and the interesting part sits in the gap between them: AI Mode is available to nearly everyone on earth, is used by a very small fraction of searches, and has already changed something structural about which pages get shown as sources.

That last point is the one worth your attention, because it is a mechanism rather than a prediction. This guide covers AI Mode specifically — what it is, how its retrieval actually works, what the strongest available numbers say about its scale, and what genuinely follows for how you plan content.

It deliberately stays off two neighbouring topics. For AI Overviews as a surface, including the click-behaviour research, read our explainer on what AI Overviews are and what changed. For the practical playbook on earning citations in Google’s AI answers, read how to show up in AI Overviews. This post is about the mechanism.

AI Mode is not an AI Overview

The two get conflated constantly, and the distinction matters because they behave differently.

An AI Overview is a generated summary that appears at the top of an otherwise normal results page. The ten blue links are still underneath it. The user did not ask for it; Google decided the query warranted one.

AI Mode is a separate conversational surface the user chooses to enter. There is no ranked list of results underneath — the answer is the page, with inline citations, and the user can ask follow-up questions that carry context forward. Google has described people in AI Mode “asking questions nearly three times longer than traditional searches”, which is the behavioural signature of a surface people use for genuinely complex questions rather than for navigation.

The strategic difference: an AI Overview competes with your blue link on the same page. AI Mode replaces the page entirely. In AI Mode there is no position three to fall back on.

Query fan-out: the mechanism that matters

Here is the part that actually changes SEO, and Google now documents it in its own guidance. Query fan-out is defined by Google as “a set of concurrent, related queries generated by the model to request more information” and fetch additional relevant results to address the user’s question.

In plain terms: when someone asks AI Mode a question, the system does not run that one query against the index and summarise the top results. It decomposes the question into a set of related sub-questions, runs them in parallel, gathers results from each, filters them, and synthesises a single answer from the combined pool.

Take a question like “is it worth hiring an SEO agency for a small Singapore business”. A single-query system retrieves pages targeting that phrase. A fan-out system might simultaneously retrieve for what SEO agencies charge in Singapore, how long SEO takes to work, in-house versus agency trade-offs, what a small business should expect for its budget, and how to tell a good agency from a bad one — then assemble one answer from all of it.

Two consequences follow directly, and they are the whole strategic content of this post.

First, the pages that get cited are frequently not the pages that rank for the question the user typed. They are pages that rank for sub-questions the user never typed and never saw. This is the accepted explanation for the most striking measured trend in AI search: Ahrefs, analysing 863,000 keyword result pages and 4 million AI Overview URLs in March 2026, found only 38% of cited pages ranked in the top 10 for the query — down from 76% in July 2025. Roughly 31% ranked at positions 11 to 100, and another 31% ranked beyond position 100 entirely. The visible results page has stopped being the source pool.

Second, topical coverage beats keyword targeting in a way that is now mechanical rather than philosophical. “Cover the topic properly” has been advice for a decade, and it was always a bit hand-wavy. Under fan-out it has a concrete payoff: each sub-question your content genuinely answers is another retrieval path into the same answer. A page that answers one question well has one chance to be pulled in. A well-structured cluster that answers twelve related questions has twelve.

How big is AI Mode, really?

Here the reporting and the measurement point in opposite directions, and both are correct.

On reach, AI Mode is enormous. Google announced in October 2025 that AI Mode “will be available in over 200 countries and territories total”, having launched “in more than 35 new languages and over 40 new countries and territories” in that wave alone. Subsequent expansions took it to close to 100 languages. Singapore is well inside that footprint.

On usage, it is currently tiny. The largest clickstream study published in 2026 — SparkToro’s analysis of Similarweb data covering US Google searches from January to April 2026 — found that just 0.34% of searches transitioned into AI Mode. That is roughly one search in three hundred. The same study found 68.01% of Google searches ended without a click, a figure that has been climbing for years for reasons that mostly predate generative AI.

Google has also said on the record that AI Mode is not the default way it serves results. It is an opt-in surface where Google ships new capability first.

Put those together and you get the correct posture: AI Mode is not currently where your traffic is, and it is where the retrieval architecture is being tested. Reorganising your entire content operation around a surface used in 0.34% of searches would be a mistake. Ignoring a retrieval mechanism that has already halved the correlation between ranking and citation across Google’s AI features would be a bigger one.

What Google says you should do about it

Short version: nothing special, and Google is unusually direct about this.

From Google’s guide to optimising for generative AI features: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” On eligibility: “to be eligible to be shown in generative AI features on Google Search, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements.” On markup: “structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” Google’s AI features documentation adds that “you don’t need to create new machine readable files, AI text files, or markup”.

The one piece of substantive content advice Google does give is worth quoting because it is unusually pointed: “don’t just recycle what others on the internet have already said, or could easily be produced by a generative AI model.”

Read that alongside the fan-out mechanism and it stops being a platitude. A synthesis engine that pools results from a dozen sub-queries has no reason to cite the fourth page that says the same thing as the first three. What it needs is something the pool does not already contain: an original figure, a first-hand observation, a local specific, a genuinely different framing. Commodity content is not penalised so much as it is simply redundant.

The corollary for anyone being sold an “AI Mode optimisation” package: there is no technical entry fee. If someone is charging you for llms.txt files, AI-specific markup or a separate AI submission process, our guide to llms.txt and AI crawlers covers what the evidence actually supports, which is considerably less than the sales pitch.

What actually changes in how you plan content

Four changes we would make, in priority order.

1. Plan in question sets, not keywords. Before writing, list every sub-question a reasonable person would need answered to act on the main question. Cost, timeline, alternatives, risks, how to choose, what it looks like when it goes wrong. Then make sure each is answered somewhere in your cluster with enough specificity to be quotable on its own. That is a direct response to fan-out, not a general content-quality nicety.

2. Make each answer independently extractable. Fan-out retrieves passages to synthesise, not pages to rank. A clear heading that states the question, followed by a direct answer in the first two sentences, then the supporting detail, is easier to pull from than the same information distributed across four paragraphs of build-up. Our on-page SEO checklist covers the structural fundamentals this sits on.

3. Put something in the page that the pool does not have. Your own pricing ranges, your own observed timelines, a local regulatory specific, a worked example with real numbers. This is the only durable defence against being the redundant fourth source.

4. Stop treating position one as the goal for informational queries. With citations drawn substantially from pages ranking outside the top ten, the return on grinding a commodity informational page from position four to position two is lower than it was. Breadth of genuine coverage now competes with depth of optimisation on a single URL. That is a real reallocation of effort, and it is the argument developed in our AI SEO strategy guide.

Measuring AI Mode, honestly

You largely cannot, and it is important to know why before you buy a tool that claims otherwise.

A click from AI Mode arrives at your site as a click from Google. GA4’s own channel definitions classify Organic Search as including Google’s AI Overviews and AI Mode, and the new AI Assistants channel explicitly excludes them. There is no configuration that separates AI Mode traffic in analytics, because the distinguishing signal is not sent.

The only official view is Search Console’s generative AI performance report, which began a limited rollout in June 2026 and reports impressions only — no clicks, no click-through rate, no position, no query data — broken down by page, country and device. It is a breakout of data already inside the Web search type, so it cannot be added to your organic totals without double-counting.

That means the practical answer to “how is AI Mode affecting us” is triangulation rather than measurement: watch total organic sessions and conversions for the pages that matter, watch AI feature impressions in Search Console if the report has reached your property, and treat any third-party number claiming to isolate your AI Mode traffic with suspicion. The full method is in our guide to tracking AI search traffic.

The Singapore read

Three things are specific to this market.

Concentration raises the stakes. StatCounter’s July 2026 figures put Google at 92.46% of Singapore search against Bing’s 3.37%. There is no meaningful alternative channel. Any structural change to how Google assembles answers reaches essentially all of your search demand at once, which argues for watching the total organic line closely rather than for panic about any one surface.

Multilingual fan-out is more relevant here than in most markets. AI Mode now runs in close to 100 languages, and Singapore’s search behaviour spans English, Mandarin, Malay and Tamil, often with code-switching inside a single query. A fan-out system decomposing a mixed-language question can retrieve across languages in a way keyword-matched search historically did not. We have not found published data measuring this specifically for Singapore, so we state it as a reasoned expectation rather than a finding — but it argues for making sure your key pages are genuinely comprehensible rather than optimised for one exact phrase.

Local and transactional intent remains comparatively insulated. AI features skew heavily toward informational and question-shaped queries. A Singapore business whose demand is dominated by “near me”, brand and ready-to-buy searches is less exposed than a publisher living on explainer traffic — which is also why local fundamentals continue to carry weight, as covered in our guide to AI and local SEO.

Frequently asked questions

What is the difference between AI Mode and AI Overviews?

An AI Overview is a generated summary at the top of a normal results page, with the usual ranked links still below it, shown because Google decided the query warranted one. AI Mode is a separate conversational surface the user chooses to enter, where the generated answer replaces the results list entirely and follow-up questions carry context forward. AI Overviews compete with your listing; AI Mode replaces the page it would have appeared on.

What is query fan-out?

Google defines it as “a set of concurrent, related queries generated by the model to request more information”. Instead of running the user’s question as a single search, the system breaks it into multiple related sub-questions, runs them in parallel, and synthesises one answer from the pooled results. The practical effect is that pages can be cited for sub-questions the user never typed, which is why citation and ranking have come apart.

How do I optimise for Google AI Mode?

There is no separate technical process, and Google states that optimising for generative AI search “is optimizing for the search experience, and thus still SEO”, with no special markup or files required. What changes is emphasis: cover the full question space around a topic rather than one keyword, make each answer independently extractable, and include something original that a synthesis engine could not get from the other sources it retrieved.

Is AI Mode replacing normal Google search?

Not on current evidence. Google has said AI Mode is not the default way it serves results, and the largest 2026 clickstream study found just 0.34% of US Google searches transitioned into it between January and April 2026. Alphabet also reported Google Search and other revenue up 17% year on year in Q2 2026. Reach is near-universal; usage is currently very small.

Can I see how much traffic I get from AI Mode?

No. GA4 classifies AI Mode clicks as Organic Search by design, and the AI Assistants channel excludes them. Search Console’s generative AI performance report shows impressions for AI Overviews and AI Mode but no clicks, click-through rate or queries, and is still rolling out. Any tool claiming to isolate your AI Mode traffic is estimating.

Should I block AI Mode from using my content?

You can, using the standard preview controls Google documents — nosnippet, data-nosnippet, max-snippet and noindex — but understand the trade-off: these are the same controls that govern ordinary search snippets, so restricting AI features means restricting your normal listings too. For most businesses that is a bad trade. The crawler-level controls and what they actually do are covered in our guide to llms.txt and AI crawlers.

The short version

AI Mode is a conversational surface with near-universal availability and, so far, very small usage. The thing to take seriously is not its traffic share but its retrieval mechanism: query fan-out decomposes one question into many concurrent searches, which is why only 38% of pages cited in Google’s AI answers now rank in the top ten, down from 76% eight months earlier.

The response is a reallocation, not a rebuild. Plan in question sets rather than keywords, make each answer extractable on its own, put something original in every page, and stop over-investing in single-URL optimisation for commodity informational queries. Google’s own documentation says there is no separate technical discipline to buy, and the measurement tools confirm there is no separate number to report. Anyone selling you either is ahead of the evidence.

If you want a content plan built around how retrieval actually works now, that is what our AI SEO service does, and our case studies show the reporting that comes with it. The wider cluster starts at our AI SEO guide for Singapore, and the sourced numbers behind this post are collected in our AI SEO statistics reference. Get in touch if you want a second opinion on your current plan.