Query Fan-Out Explained: How Google AI Mode Turns One Search Into a Dozen Questions

Short answer: Query fan-out is the single most important concept for understanding why AI-era SEO looks nothing like blue-link SEO. When a user asks AI Mode "how do I improve my restaurant's online visibility," AI Mode doesn't run one search. It silently generates 8–15 sub-questions — "how does Google Business Profile work for restaurants," "what schema helps restaurants," "what's the review threshold for local pack," and so on — runs a search for each, and synthesizes the sources into one answer. That's why AI citations match Google's top 10 only 7–12% of the time: the AI is citing sub-question sources, not head-query sources.

How Fan-Out Actually Works

1. User query enters AI Mode. "Best way to launch a Shopify store in 2026?"
2. Query decomposition. Gemini (via Gemini 3.5 Flash as of I/O 2026) rewrites the query into a set of self-contained sub-questions:
   • What's the current Shopify plan structure and pricing?
   • What theme performs best for conversion?
   • What apps are essential at launch?
   • What's the pre-launch SEO checklist?
   • What payment gateways are available in [user region]?
   • What's the return-policy standard for new stores?
   • How do you set up shipping zones?
   • What's the first-week marketing playbook?
3. Parallel retrieval. Each sub-question triggers its own search (Google Search API + AI Mode's index).
4. Source selection. The best 1–3 sources per sub-question are pulled in.
5. Synthesis. Gemini stitches the sub-answers into one coherent response with inline citations.
6. Follow-up generation. Related questions the user might want to ask next are surfaced.

Why This Broke Traditional SEO

For 20 years, SEO optimized for the head query. Rank for "best way to launch a Shopify store" and you win the traffic. In an AI Mode world, that page might not appear in the final answer at all — because the winning sources are the ones that rank for the sub-questions. Your comprehensive 4,000-word guide is out-cited by:

• Shopify's own pricing page (sub-question 1)
• A theme-comparison blog (sub-question 2)
• An app-review site (sub-question 3)
• Your competitor's pre-launch checklist (sub-question 4)

The head-query SERP winner earns clicks. The sub-question winners earn citations. In 2026, those are two different games.

The Structural Implication: One H2 Per Sub-Question

The tactical response is simple in principle, hard in execution: every important sub-topic should be independently discoverable and self-contained.

What "independently discoverable" means

• Each sub-topic gets a question-shaped H2 with an ID (so it can be linked to directly).
• The answer under it stands alone — a reader who lands via anchor link should get a complete answer without needing to read the paragraphs above.
• Schema is applied at the appropriate level: HowTo per procedure, FAQPage per Q&A cluster, Article at the page level.
• Internal links connect related sub-topics — AI Mode follows these when synthesizing.

What "self-contained" looks like

Compare:

• Bad: "As mentioned above, the process starts with X. Then, as we discussed in the previous section, you do Y."
• Good: "To do Z, start with X (see [linked page] for the setup), then Y (a 3-step process: A, B, C)."

The bad version depends on document context AI can't carry into a synthesized answer. The good version is a citable island.

How to Reverse-Engineer the Fan-Out for Any Query

1. Pick a target head query.
2. Ask ChatGPT or Perplexity the same query. Read the inline citations — those are approximately the sub-question winners.
3. For each cited source, identify the sub-question it answers.
4. Ask yourself: does your site have a dedicated, well-structured page for each of those sub-questions?
5. The gaps are your content plan.

The Content Strategy Shift

Old model: one pillar page ranks for the head query and captures traffic. New model:

• One pillar page anchors the topic and ranks for the head query.
• 8–15 sub-topic pages each optimized for one sub-question, each independently citable.
• Tight internal linking between pillar and sub-pages so AI Mode can traverse the cluster.
• Table of contents and jump links on the pillar so users (and AI extractors) can navigate cleanly.

This is the "topic cluster" model that content strategists have advocated for years. Fan-out finally makes it the highest-ROI content shape, not just a nice-to-have.

What This Means for Small Sites

Good news for SMBs: fan-out favors specificity. A local dental clinic with a well-structured page for each sub-question ("what does a dental cleaning cost in Vancouver," "how often should I get X-rays," "is invisalign covered by BC MSP") can out-cite a national aggregator for local queries — because the aggregator's head-query authority doesn't translate into sub-question dominance. Depth on the specific beats breadth on the generic in an AI Mode world.