Query Fan-Out: How Google AI Mode Splits One Question Into Many And How To Win Each Sub-Query

A customer types one question into Google AI Mode. Behind the scenes, Gemini fires off six. Each one hits Google's index. Each one pulls a different shortlist of pages. The synthesized answer the user sees borrows from all six. If your page only ranks for the top-level query, you are visible for one slice of the retrieval pool and invisible for the rest.

Google announced this pattern at I/O 2025 and gave it a name: query fan-out. The implication is bigger than most SEO teams have absorbed. The unit of optimization is no longer the user-facing keyword. The unit is the set of sub-queries Gemini generates on the way to an answer, and most of those sub-queries are never typed by a human.

Teams that win in AI Mode are not the teams with the cleanest H1 or the most matched-keyword title tag. They are the teams whose pages happen to be the best answer to several of the fanned-out sub-queries at once. This guide breaks down what query fan-out actually does, how to discover the sub-queries Gemini runs on your category, and the structural changes that let one page win four or five sub-queries instead of just the one its title was written for.

What Query Fan-Out Actually Does

Query fan-out is the technique Google AI Mode uses to decompose a user prompt into several parallel related queries, run each one against Google's index, and synthesize a single answer from the union of retrieved passages. Google documented the technique in its AI Mode launch materials at I/O 2025, framing it as a way to handle complex, multi-step questions that no single keyword search would resolve.

The mechanics matter. A traditional Google search runs one ranking pass per user query. AI Mode runs many. When a user types "is the Feno smart toothbrush worth it for sensitive teeth," Gemini does not simply look up that string. It expands the query into a constellation of related searches. Some of those searches map to traditional keyword intent (best smart toothbrush for sensitive teeth). Others map to product attributes (Feno toothbrush brush pressure sensor). Others map to comparison intent (Feno vs Oclean for gum sensitivity). Each sub-query produces its own short list of candidate pages. The model assembles the answer from a curated selection across the lists.

This is not the same as the query expansion classical search engines have done for years. Classical expansion adds synonyms and morphological variants to one query. Fan-out generates conceptually distinct sub-queries that probe different facets of the user's question, retrieves independently for each, and reconciles the results. The retrieval surface is wider, deeper, and more specific.

Where Fan-Out Sits In The AI Mode Pipeline

The pipeline runs in roughly four stages. The user prompt is parsed and a planning step decides which sub-queries are worth running. Each sub-query goes through a retrieval cycle against Google's index, including Knowledge Graph entries, structured data, and standard web results. Retrieved passages are scored for relevance to the original prompt, not just to the sub-query. The model writes the answer and attaches citations to the passages it relied on.

The optimization implication is straightforward. Your page can be retrieved in stage two for a sub-query you never targeted, get scored as the best evidence in stage three, and earn a citation in stage four, all without your team realizing that sub-query existed in the first place.

Why Traditional Keyword Ranking Misses Sub-Query Demand

Most SEO programs are built around the visible keyword, the one a customer types or speaks. Search Console reports it. Keyword research tools rank it by volume. Editorial briefs anchor on it. The whole apparatus rewards optimizing for the surface query.

Query fan-out breaks this model in two places. First, the sub-queries Gemini generates often have low or zero search volume because they were never typed by a human user. Standard keyword tools will not surface them. Second, the user-facing query may not be the one that triggers your citation. A page can be cited because it happened to rank well for a sub-query the model generated, even though the page targets a different head term entirely.

The result is a measurement gap. Teams see their rank tracking tool show position three for the primary keyword, then watch AI Mode answer the same prompt while never mentioning their page. The page was not present for the right sub-queries. Rank tracking lies because the unit it tracks is no longer the unit that matters.

This shift mirrors what happened with AI Overviews, but with a wider impact. AI Overviews typically pull from the top organic results for the user-facing query plus a few related queries. AI Mode is more aggressive, more multi-step, and more willing to pull from pages that never ranked for the top-level term at all.

Reverse Engineering The Sub-Queries Google Runs

You cannot optimize for sub-queries you cannot see. The first practical task is making the invisible visible.

Several approaches surface the sub-queries Gemini is likely running. None is a perfect mirror of Google's internal pipeline, but in combination they give a usable working set.

• Ask AI Mode directly - The model will often disclose the questions it considered when prompted. Type your target query into AI Mode, then follow up with "what searches did you run to answer that." The response is not always literal, but it gives you a list of related intents the model treated as relevant.
• Use the People Also Ask block - PAA expansions on the standard SERP for your target query are a strong proxy for sub-queries Gemini might generate. They share the same underlying entity-relationship model Google's ranking systems use across surfaces.
• Inspect autocomplete and related searches - Google's autocomplete and "Related searches" carousel surface adjacent intents that the index has clustered around your target. These are not guaranteed to be Gemini's fan-out, but they overlap heavily.
• Use specialist tracking tools - Profound, Otterly.ai, AthenaHQ, and Ahrefs Brand Radar each maintain query libraries that simulate fan-out behavior across AI Mode, ChatGPT, Perplexity, and Gemini. They sample prompts at scale and surface which sub-queries trigger citations of your domain and competitor domains. These tools are paid but cut the discovery time from weeks to hours.

Build A Sub-Query Map

Once you have a working set of sub-queries for a target prompt, document them in a sub-query map. The format is simple: target user query in column one, list of likely sub-queries in column two, the page on your site best positioned to answer each sub-query in column three. A gap in column three is a content opportunity. A duplicate page in column three is a cannibalization risk.

Five to twenty target user queries with their sub-query maps is enough to start. Each map roughly takes 30 to 45 minutes to build when you start, dropping to 10 minutes once the workflow is familiar.

Structuring Pages To Win Multiple Sub-Queries

Once you know the sub-queries, the structural work begins. One page winning multiple sub-queries is the goal because it concentrates topical authority and limits the canonicalization headaches that come from spreading sub-queries across many thin pages.

The single biggest structural lever is what we have called content chunking in earlier posts. Each H2 section becomes a self-contained passage that answers one specific sub-query in full, with a heading that mirrors the question intent and a first sentence that delivers the extractable answer. Done well, a single 3,000-word guide can host eight independently retrievable passages.

Schema markup compounds this advantage. FAQPage, HowTo, and Article schemas annotated with sub-section anchors give Google's retrieval system explicit signals about which parts of your page answer which questions. Pages with proper schema tend to earn richer treatment in AI Mode citations because the model can identify the relevant passage faster.

Internal linking matters too, but in a specific way. Anchor text that uses the natural language phrasing of likely sub-queries gives Google a signal that the linked page is the canonical answer for that sub-intent. A page titled "Smart Toothbrush Buying Guide" should be linked from related pages using anchor text like "how to pick a smart toothbrush for sensitive teeth" rather than just the title or a generic phrase.

Question-Shaped Headings Win More Often Than Topic Headings

Compare two H2 conventions. "Battery Life" labels a topic. "How long does a smart toothbrush battery last between charges" labels an answer. The second pattern matches the natural language phrasing Gemini's planner is more likely to generate when it fans out for a battery-related sub-query. Question-shaped headings cost nothing to adopt and consistently outperform topic-style headings in AI retrieval contexts.

A Worked Example: The Smart Toothbrush Query

A worked example clarifies the workflow. Imagine a smart toothbrush brand called Feno. Their top-of-funnel target query is "is the Feno smart toothbrush worth it."

When we typed that prompt into Google AI Mode and asked the model to explain its search behavior, it returned a list of sub-queries it had considered. The list was approximately:

• Feno smart toothbrush features and specs
• Feno smart toothbrush price and where to buy
• Feno smart toothbrush vs traditional electric toothbrush
• Feno smart toothbrush vs Oclean or Quip alternatives
• Feno smart toothbrush user reviews on Reddit and Trustpilot
• Feno smart toothbrush warranty and customer support
• Smart toothbrush effectiveness for gum health and sensitive teeth

The brand's existing site had a strong product page that targeted the head term but no dedicated content for five of those seven sub-queries. The single feature blog post on the site mentioned brush pressure but did not directly answer questions about gum sensitivity or warranty support.

The rewrite created one comprehensive guide structured as eight H2 sections, each anchored to a likely sub-query. The first three sections covered product features, pricing, and comparison alternatives. The next three covered third-party reviews, warranty terms, and customer support response times. The final two sections covered effectiveness for specific dental conditions and the long-term cost comparison versus traditional electric brushes.

The page also kept its commercial intent through a tight CTA in the closing synthesis and structured product data in JSON-LD. After 60 days, AI Mode citation rate for any of the seven sub-queries climbed from one in seven to five in seven, with the new guide cited in 71 percent of test prompts.

The lesson is not the specific numbers. The lesson is that one well-structured page outperformed seven thin pages because Google's retrieval system rewards depth on related sub-questions when they live together with clear semantic boundaries.

Measuring Visibility Across Fanned-Out Queries

Traditional rank tracking is necessary but insufficient. To measure AI Mode visibility properly, you need a second layer of metrics tied to the sub-queries themselves.

Three measurement habits make the difference. The first is sub-query citation tracking. For each of your target user queries, sample 25 to 50 fanned-out prompts via AI Mode (manually or via Profound, Otterly.ai, or AthenaHQ) and record whether your domain is cited, whether competitor domains are cited, and which page on your site earns the citation when one occurs. Run this monthly. The metric to watch is the share of fanned-out prompts that cite your domain.

The second habit is passage-level extraction. When your domain is cited, note which heading or passage the model surfaced. Patterns emerge fast. Question-shaped H2s typically beat topic-style ones. Self-contained passages with a citable first sentence beat passages that require surrounding context.

The third habit is correlated with citation analytics more broadly. Cross-reference your AI Mode citation share with traditional organic ranking for the same user-facing query. When ranking is high but citation share is low, the page is reaching the right SERP but losing the fan-out. When citation share is high but ranking is low, the page is winning sub-queries despite not topping the surface results. Both patterns require different fixes.

The Lag Between Organic Ranking And AI Mode Citation

Pages that newly enter top organic positions do not immediately earn AI Mode citations. There is a lag, typically 2 to 6 weeks in our observation, during which Gemini's retrieval and synthesis layers test the new page across multiple sub-queries before stably citing it. Patience matters. Most teams give up on a page that is one week old and not yet cited, then are surprised when it dominates fan-out six weeks later.

Six Patterns That Hide Pages From Query Fan-Out

Several recurring failures keep otherwise strong pages out of fan-out retrieval. Avoiding them is a high-leverage starting point before deeper structural work.

1. Topic-style H2s instead of question-style H2s. "Battery Life" loses to "How long does the battery last between charges." The first targets a topic; the second targets the actual phrasing of a likely sub-query.
2. Buried answers. When the answer to a question lives in paragraph four of a section, Gemini's passage scorer often picks a competing page where the answer sits in the first sentence. Move the answer up.
3. Sub-queries scattered across thin pages. A separate 800-word post for each sub-query splits authority and underwhelms on depth. Consolidate related sub-queries into one substantial guide with clear section boundaries.
4. No FAQPage or HowTo schema. The structured data costs almost nothing to add and gives Gemini explicit cues about which passages answer which questions. Skipping it leaves signal on the table.
5. Generic internal anchor text. Linking with "click here" or just the page title misses the chance to tell Google which sub-query the linked page is the canonical answer for. Use natural language anchor text that mirrors likely sub-queries.
6. Stale comparison content. Comparison sub-queries (X vs Y) are over-represented in fan-out because users genuinely need decision support. Comparison pages that have not been refreshed for the current generation of competitors drop out of citation quickly. Audit and refresh them every six months minimum.

Frequently Asked Questions

Is query fan-out the same as AI Overviews retrieval?

No. AI Overviews and AI Mode are both Google AI surfaces, but they retrieve differently. AI Overviews mostly draws from the top organic results for the user-facing query plus a small number of related queries. AI Mode runs a more aggressive query fan-out, generating multiple semantically distinct sub-queries and pulling from a wider retrieval pool. A page can win AI Overviews citations through traditional SEO alone. Winning AI Mode reliably requires explicit optimization for the sub-queries themselves.

How can I see the actual sub-queries Gemini ran for a prompt?

Google does not formally expose the sub-queries Gemini generates, but several proxies get close. Asking AI Mode "what searches did you run for that" prompts the model to disclose the intents it considered. Tools like Profound, Otterly.ai, AthenaHQ, and Ahrefs Brand Radar maintain query libraries that simulate fan-out behavior at scale. Google's own People Also Ask and Related Searches surfaces are useful adjacent signals. None of these is a literal mirror of the pipeline, but in combination they give a working set within 10 percent of accuracy in our experience.

Should I split one comprehensive guide into multiple pages targeting individual sub-queries?

Almost always no. Consolidation wins for query fan-out because one page hosting eight self-contained passages concentrates topical authority and signals to Google that the page is comprehensive on the topic. Splitting into thin pages spreads authority and triggers cannibalization audits. The exception is when sub-queries map to genuinely different user intents (informational vs transactional) or different funnel stages, in which case separate pages serve different reader needs.

How long does it take for a new page to start earning AI Mode citations?

In our observation, two to six weeks after the page enters reasonable organic positions for its target query. During the lag, Gemini's retrieval and synthesis layers test the page across multiple sub-queries before stably citing it. Pages that earn early citations tend to do so first on long-tail sub-queries (where competition is thin) and then on head sub-queries as the model gains confidence. Teams that give up on a page after one week miss the typical citation curve.

Does query fan-out work the same for non-English queries?

Largely yes, with a caveat. Google AI Mode rolled out to English first and has expanded to additional languages since launch. The fan-out mechanic is language-agnostic in principle, but the breadth of fan-out is correlated with the density of Google's index for that language. Sub-query coverage in Spanish, French, German, and Japanese has been comparable to English in our testing. Smaller language indexes (Vietnamese, Filipino, Polish) often see narrower fan-out and fewer sub-queries per prompt.

The practical takeaway is that classical keyword optimization is no longer sufficient for AI Mode visibility. The unit that matters is the set of sub-queries Gemini fans out behind the surface query, and the best way to win is to make one substantial page the canonical answer to several related sub-queries at once.

The workflow that delivers this is not exotic. Map the sub-queries for each target user prompt using AI Mode disclosures, PAA expansions, and specialist tracking tools. Consolidate related sub-queries into one well-structured guide with question-shaped H2 headings, citable first sentences, and FAQPage or HowTo schema where it fits. Measure citation share across fanned-out prompts, not just rank on the surface query. Refresh comparison content every six months because users keep asking new versions of the same question.

If your team wants a structural audit of how your current pages perform across query fan-out for your top buyer-intent prompts, that work sits inside our generative engine optimization program. Optimize for the sub-queries Gemini actually runs and the citations follow.