Google no longer searches your query — it searches 15 others

What query fan-out is

A provocation backed by a patent: Google no longer searches exactly your query. It unpacks it into 5–15 parallel sub-queries and answers each from its own pool of pages. The AI answer is assembled from the pages that land most often across that whole fan — not from the top-1 on the original query.

The official example from Google’s guide (2026-05-15): the query “how to fix a lawn full of weeds” is automatically expanded into “best herbicides”, “remove weeds without chemicals”, “prevent weeds”. The mechanic is confirmed by the Google I/O 2026 schema and patent US20240289407A1.

The ranking paradox: top-3 misses, top-15 makes it

Since citation goes by fan frequency, not position, the counterintuitive happens:

• A top-3 site on the main query may not appear at all in the AI answer.
• A top-15 competitor gets in — because it answers one of the sub-questions more precisely.
• A page at position #150 that answers one sub-query exactly has a citation shot on par with #3.

The numbers agree:

• The “top-10 ↔ AI-citation” correlation fell from 76% (2025) to 38% (2026) — Ahrefs, March 2026, 863k queries, 4M URLs.
• ~60% of AI-Overview citations come from URLs beyond the top 20 organic (AirOps, State of AI Search 2026).
• 31% of citations come from pages beyond the top 100.

How much the fan covers

The fan weighs differently across Google’s two engines (see the AIO vs AIM breakdown):

Answer source	AI Overview (AIO)	AI Mode (AIM)
From the original query	~60%	~15%
From the extra sub-queries	+10–15%	+30%
Total covered by fan-out	~75%	~45%

In AI Mode the fan decides almost half the answer — sub-intent coverage is critical there.

How to write for the fan

The strategy shifts from “rank #1” to “cover the fan”:

1. One page = one sub-question, deeply. Not “all about X” shallowly, but a precise, exhaustive answer to a single sub-intent.
2. A network of topical pages/sections — 8–12 reformulations of one query across the understand / compare / buy / verify axes.
3. Atomic 40–60-word answers with a number in the first words for each sub-question — that’s what the engine lifts.
4. FAQPage markup doubles the retrieval surface: one FAQ block can close several sub-queries at once.
5. An explicit “fan-out map” in the brief for the writer: a list of 8–12 phrasings across 8 axes — so no sub-intent is missed.

You can’t read the exact sub-queries (Google removed their visibility), so you model them — from the sections of the AI answer (often 1 section = 1 sub-query), from PAA, and from derived queries.

How to check it on your own site

We built this into the audit:

• facet-coverage — whether the page covers the fan’s facets (comparison / alternatives / price / pros-cons / specs / FAQ) as distinct sections, and what’s missing across the cluster.
• coverage-gap — takes the real PAA sub-questions for your key query and shows which of them the article body doesn’t answer.

What to do

1. Change the goal. Not “rank #1 on the key,” but “cover the sub-query fan.”
2. Unpack the key into 8–12 sub-intents and answer each — deeply and separately.
3. Atomize the answers (40–60 words, number up front) + FAQPage.
4. Measure presence in the answer, not just position — they’re different KPIs now.

You still need Google rankings. But in the AI era the winner isn’t whoever took one summit — it’s whoever covered the whole fan.

Sources

• Query Fan-Out — Google patent US20240289407A1 + Google’s official AI-search guide (2026-05-15): splitting a query into parallel sub-queries; the “fix a lawn” example.
• Google I/O 2026 — the Query Fan-Out schema (via Bormintsev’s talk, GEO-HowTo 2026).
• Ahrefs, March 2026 (863k queries, 4M URLs) — top-10 ↔ AI-citation correlation 76% → 38%, 31% of citations beyond top-100 (via Runov’s talk, GEO-HowTo 2026).
• AirOps — State of AI Search 2026 — ~60% of AIO citations beyond top-20 organic.
• GEO-HowTo 2026 — Kurdyukova–Volovich talk — measuring the AIO vs AIM fan shares.