Query fan-out: why one blog post cannot win AI search
AI engines decompose every complex prompt into 8–12 sub-queries and search them in parallel. Hub-and-spoke topic clusters are the only structure that survives fan-out. Here is how to build one.
Founder 5 min read
Summarize with AI Open this article in your preferred assistant
In this article
What is query fan-out?
Query fan-out is the AI-era retrieval pattern in which a single user prompt is decomposed into 8 to 12 narrower sub-queries, each retrieved in parallel, before the engine synthesizes a single answer. It is the default behavior of ChatGPT search, Perplexity, Claude with web search, Gemini, and Google AI Overviews. The user types one question. The engine runs a dozen.
The strategic implication is direct: an isolated blog post can satisfy one sub-query. It cannot satisfy a fan-out. If a competitor has published the pillar plus the spokes that cover the remaining 7 to 11 sub-queries, the competitor will be cited 7 to 11 times more often inside the same answer — for the same parent prompt your isolated post was written to win.
The 2025 Profound Share of Model Report documented that in mature B2B categories, the top three brands capture 70–80% of all AI citations combined. The mechanism behind that concentration is not authority alone. It is coverage. The brands that own the top citation share own complete clusters. The brands underneath them own scattered posts.
How does fan-out actually decompose a query?
Most fan-out decompositions split along four predictable intent classes. The cluster strategy follows the decomposition. For the parent prompt “What is GEO and how do I do it?”, a typical fan-out produces sub-queries across these classes:
- Definitional — What is GEO? What is the difference between GEO and SEO? What does Share of Answer mean?
- Comparative — GEO vs traditional SEO? Citable vs Profound vs in-house? Which AI engine cites which brands most?
- Procedural — How do I run a GEO audit? How do I optimize for Perplexity specifically? How do I structure FAQ schema for ChatGPT?
- Evidence — What are GEO case studies? What does GEO ROI look like? What are the data points behind AI citation patterns?
A pillar page can navigate all four classes in one document, but a single document cannot answer all four classes in depth — and depth is what determines citation. The spoke layer is where depth lives. The pillar is the index. The spokes are the answers.
What does a complete hub-and-spoke cluster look like?
A complete cluster is one pillar page plus 5 to 10 spokes, bidirectionally linked, with each spoke owning one sub-query intent. Bidirectional means the pillar links to every spoke and every spoke links back to the pillar — and where it is editorially honest, spokes also link to each other. This is the link topology AI engines and traditional search crawlers both reward.
A working Citable cluster on the topic AI Visibility Measurement would ship as:
- Pillar — The Complete Guide to AI Visibility Measurement (3,000–4,000 words, indexes every sub-question, links to every spoke)
- Spoke (definitional) — What is Share of Answer?
- Spoke (definitional) — Share of Answer vs Share of Model
- Spoke (comparative) — Profound vs Peec vs Otterly: AI Visibility Tools Compared
- Spoke (procedural) — How to Run an AI Visibility Audit in 7 Days
- Spoke (procedural) — How to Construct a 50-Prompt Test Set
- Spoke (evidence) — Q1 2026 Share of Answer Benchmarks by Sector
- Spoke (evidence) — Citable Case Study: SaaS Brand Lifts Share of Answer from 4% to 18%
Eight URLs. One topic. One semantic web that an AI engine fanning out the parent query can land on regardless of which sub-query fires.
Why does bidirectional linking matter?
Bidirectional linking does two things that one-directional linking does not. First, it distributes internal link equity across the cluster so that traditional Google retrieval — still the base layer that AI engines query — sees the spokes as authoritative, not orphaned. Second, it creates the navigable structure that LLM crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) use to discover related content from a single entry point.
Pages without inbound internal links from their topic cluster are roughly invisible to AI crawlers that enter via the pillar — even if they rank for the spoke’s primary keyword on Google. The crawler does not know they belong to the cluster, so it does not treat them as part of the brand’s answer set for the parent topic. Bidirectional linking is the cheapest, most underused lift in GEO content strategy.
How does this connect to the 60-day freshness loop?
A pillar-and-spoke cluster is a maintenance unit, not a launch artifact. AI engines actively down-weight content older than 30 days, with citation potential dropping up to 40% for pages older than 30 days that have not been refreshed (Citable internal data, Q1 2026 cohort, n=312 tracked pages). Citable’s Citation Freshness Loop SKU treats every cluster as a living asset: statistics refreshed, Last Updated timestamps cycled, schema markup re-validated, internal links re-checked every 14 to 60 days.
The strategic asymmetry: brands that ship one cluster and walk away will see their citation share decay inside a single quarter. Brands that ship one cluster and maintain it on a 14–60 day loop compound their Share of Answer month over month while competitors’ isolated posts age out of the citation set.
What about the social distribution layer?
The same pillar-and-spoke cluster doubles as the foundation for owned-channel distribution. A 3,500-word pillar can be broken into 5 to 7 LinkedIn posts, each carrying one sub-query insight back to the cluster. The architecture works one way: the long-form anchor is what AI engines retrieve and cite, the social posts are what humans engage with, and the link in every social post points back to the pillar where the citation lives. Inverting this — leading with social, treating the blog as derivative — leaves the GEO surface unbuilt.
Hub-and-spoke is not a content tactic. It is the minimum unit of GEO surface area in 2026. Anything smaller is leaking citation share to brands that built the cluster while you were publishing isolated posts.
Citable builds pillar-and-spoke clusters with the AED+BLUF editorial standard and maintains them under the Citation Freshness Loop. Start with the AI Visibility Audit to see which sub-queries your category fans out into.
Frequently asked
Questions buyers ask before booking
What is query fan-out in AI search?
Query fan-out is the process by which an AI engine takes a single user prompt and decomposes it into 8 to 12 narrower sub-queries, retrieves documents for each in parallel, and synthesizes the results into one answer. The user sees one response. The engine ran a dozen retrievals underneath it. This means the surface area of a 'single AI question' is roughly an order of magnitude larger than the surface area of a single Google query.
How many spokes does a pillar page need?
Five at minimum, ten at the ceiling for most B2B categories. Below five, the cluster cannot cover the sub-query intent classes (definitional, comparative, procedural, evidence). Above ten, internal-link equity starts to dilute and editorial maintenance under a 14–60 day freshness loop becomes the bottleneck. The sweet spot for most Citable engagements is one pillar plus seven spokes.
Does the pillar page need to be longer than the spokes?
Yes, but length is not the point — coverage is. The pillar must contain a navigable answer to every sub-query the topic generates, with each section linking out to the spoke that goes deep. Pillars in well-executed clusters run 2,500–4,500 words; spokes typically 1,000–1,800. If your pillar is shorter than your average spoke, the architecture is inverted.
How is hub-and-spoke different from traditional SEO topic clusters?
The structure is the same. The selection criteria changed. Traditional SEO clusters were built around keyword volume and search intent inferred from SERP analysis. GEO clusters are built around the sub-query set an AI engine actually fans out — measurable by prompting ChatGPT, Perplexity, and Gemini with the parent question and logging the follow-up retrievals. The spokes are dictated by what the engine asks, not by what a keyword tool suggests.
How long does a complete hub-and-spoke cluster take to ship?
On a disciplined editorial cadence with the AED+BLUF standard applied to every piece, a complete pillar plus seven spokes ships in 6 to 10 weeks. Two weeks for prompt-set construction and outline. Four to six weeks for drafting and editorial review. One to two weeks for schema markup, internal linking, and citation-loop entry. Faster than that and the architecture drifts. Slower than that and the freshness clock starts costing citation share before the cluster is complete.