How AI search engines actually break down your question
When you type a question into ChatGPT or Perplexity, you probably imagine the system doing one search, finding some pages, and stitching together an answer. The reality is quite different. Modern AI search engines use a technique called query fan-out to decompose your single question into a cluster of related sub-queries, run those in parallel (or in sequence), and then synthesise the results into one coherent response.
Think of it like this. You ask: "What's the best skincare routine for oily skin in your 30s?" A naive system would search exactly that phrase. A fan-out system might generate five or six sub-queries: "sebum production in your 30s", "niacinamide for oily skin", "lightweight moisturiser recommendations", "SPF for oily skin types", "common skincare mistakes oily skin". Each sub-query retrieves a different set of sources. The final answer draws from all of them.
This is not a theoretical future feature. Perplexity has been open about its multi-step reasoning approach. Google's AI Mode uses what Google calls "query fan-out" explicitly in its documentation. ChatGPT with browsing enabled runs multiple searches per session. If your content only answers the surface-level question and ignores the surrounding sub-topics, you are invisible to large portions of every AI-generated answer.
Why this changes everything about content strategy
Traditional SEO optimised one page for one keyword. You wrote a page about "oily skin moisturiser", targeted that phrase, and hoped to rank. Query fan-out breaks that model. The AI is not looking for the single best page about a topic. It is pulling fragments from multiple pages, each of which answers a different sub-query well.
This has two big implications.
Depth within a topic beats breadth across topics
If your skincare brand has one generic "skin types" page that mentions oily skin briefly, you will almost certainly lose to a competitor who has a dedicated page on oily skin in your 30s, another on the role of ceramides, and a third specifically comparing SPF formulations for oily skin. The AI's fan-out retrieval rewards specialists, not generalists.
That said, depth does not mean padding. A 3,000-word page that spends 1,500 words on introductions and disclaimers will be less useful than a tight 900-word page that directly answers the sub-query. AI systems extract the useful paragraph, not the whole page. This is covered in detail in the post about why AI quotes one paragraph and ignores the rest of your page.
Your content needs to cover the full query cluster, not just the head term
If you sell project management software, your target keyword might be "best project management tool for remote teams". But the fan-out sub-queries could include things like "how to track tasks across time zones", "async communication tools compared", "project management software pricing 2025", and "Gantt chart alternatives". If your site only covers the head term, you will be cited for almost nothing.
The practical fix is to map out the likely sub-queries before you write. Ask yourself: if someone asked an AI to research this topic thoroughly, what ten things would it need to know? Then make sure your site content, taken as a whole, covers at least seven of them directly.
How query fan-out interacts with structured data
Structured data (schema markup) does not directly influence which sub-queries get generated. But it does dramatically affect whether your page gets selected as the answer to a given sub-query. Here is why.
When a fan-out system retrieves candidate pages for a sub-query like "lightweight SPF for oily skin", it needs to quickly understand what each page is actually about. Schema markup gives AI crawlers a machine-readable summary of your content's type, topic, and key attributes. A Product schema with a clear name, description, and relevant properties tells the crawler "this page is specifically about this product and these attributes" in a fraction of a second. A page with no schema requires the crawler to infer meaning from HTML structure and prose, which introduces uncertainty.
Beyond Product schema, FAQPage, Article, and HowTo schemas are particularly valuable for fan-out optimisation because they map naturally onto the question-and-answer structure that sub-queries produce. If a sub-query is "how do you apply SPF to oily skin", a page with HowTo schema and clear step markup is a much cleaner match than a blog post with the same information buried in flowing prose.
This is one of the core reasons FlinnSchema focuses on structured data implementation alongside content strategy. Schema is not just about Google rich results anymore. It is a communication layer between your content and AI retrieval systems, and fan-out makes that communication more important, not less.
Practical steps to optimise for query fan-out
Here is a concrete process you can follow, regardless of your platform.
Step 1: Run a prompt research session
Open ChatGPT or Perplexity and ask questions your customers would ask. Watch what follow-up questions the AI asks itself. Perplexity, in particular, shows you its "related searches" at the bottom of every response. Those are a direct window into its fan-out sub-queries. Collect 20 to 30 of these and group them by theme. That grouping becomes your content map.
This process is sometimes called prompt research, and it is quickly becoming the most useful form of keyword research for AI-first content strategies. You can read more about that approach in the post on prompt research as the new keyword research for AI search.
Step 2: Audit your content against the sub-query map
Take your grouped sub-queries and check whether your existing content covers each one. Be honest. "Mentions it briefly" is not the same as "answers it directly". For any sub-query that is not clearly answered somewhere on your site, that is a content gap with a direct impact on AI visibility.
A free AI visibility audit can help identify these gaps systematically, rather than just eyeballing your site and guessing.
Step 3: Create dedicated, focused pages for high-value sub-queries
Resist the temptation to cram all sub-query content onto one mega-page. Fan-out systems retrieve pages, not sections. A dedicated 800-word page that answers one sub-query brilliantly will outperform a 4,000-word page that answers it adequately alongside nine other things.
Think of each page as a candidate answer to a specific sub-query. Write the title and opening paragraph as if you are directly answering that sub-query. Use the relevant schema type. Keep it focused.
Step 4: Build entity relationships across your content
Fan-out systems are entity-aware. They understand that "retinol", "vitamin A", and "tretinoin" are related concepts. Your content should explicitly name and connect these entities, not assume the AI will infer the relationship. This means using the proper names for products, ingredients, processes, and people. It means linking between related pages on your own site. And it means using schema properties like sameAs to connect your entities to known external references (Wikipedia, Wikidata, industry databases).
Step 5: Structure your prose for extractability
Each page should contain at least one paragraph that could stand alone as a complete answer to the sub-query it targets. This "golden paragraph" should appear early, state the answer directly, include the key entity names, and be between 40 and 120 words. This is the paragraph the AI will quote. Everything else on the page provides supporting context that builds your authority on the topic.
The difference between fan-out visibility and traditional ranking
With traditional search, you either rank on page one or you effectively do not exist. Fan-out AI search is more forgiving in one way and less forgiving in another.
More forgiving: you do not need to rank number one for any single query. If your page is the best answer to one specific sub-query, you get cited in the final answer, even if you would never rank on page one of Google for the head term. This is genuinely good news for smaller brands and specialists. As covered in the post on whether small businesses can beat big brands in AI search, specificity and depth can outperform domain authority in fan-out retrieval.
Less forgiving: if your content is generic and covers topics at surface level only, you will be invisible to every sub-query, not just the hard ones. Generic content gets crowded out by specialists at every point of the fan-out.
The brands winning in AI search right now are not necessarily the ones with the biggest budgets or the most backlinks. They are the ones whose content maps most closely onto the sub-query structures that AI systems generate. Understanding query fan-out is the first step to building that kind of content.
Frequently Asked Questions
What is query fan-out in simple terms?
Query fan-out is the process by which an AI search engine takes a single user question and splits it into multiple smaller sub-queries. It runs those sub-queries separately, retrieves different sources for each, and then combines the results into one answer. The "fan-out" refers to the way one question fans out into many parallel searches.
Does query fan-out affect Google search as well as AI search?
Yes. Google's AI Mode explicitly uses query fan-out as part of its retrieval process, and Google has referenced this in its own documentation. Traditional Google web search does not fan out in the same explicit way, but AI Mode and AI Overviews both use multi-query retrieval. If you are optimising for Google AI Mode, fan-out optimisation is directly relevant.
How many sub-queries does a typical fan-out produce?
This varies by system and by the complexity of the original question. Simple factual questions may produce two or three sub-queries. Research-style questions ("what is the best CRM for a 10-person sales team") can produce eight to twelve or more. Perplexity's "related searches" feature gives you a rough indication. Google has mentioned fan-out counts of five to ten sub-queries for moderately complex questions in its developer documentation.
Can schema markup alone fix my fan-out visibility?
No. Schema markup improves how clearly your existing content is understood by AI crawlers, but it cannot compensate for missing content. If your site simply does not cover the sub-queries that a fan-out produces, no amount of structured data will make you appear in those answers. The right approach is to close content gaps first, then apply schema to ensure each page is retrieved and interpreted correctly. Both parts matter.