Query Fan-Out — Why AI Search Doesn't Actually Run Your Keyword

You type "cold email deliverability" into Google AI Mode. Google doesn't run that query. It runs about thirty.

This is "query fan-out," and it's the single biggest reason most SEO measurement from 2024 stopped predicting AI-search traffic in 2026. The keyword your analytics dashboard says you rank for is not the query the AI engine actually ran. It ran dozens, synthesized the answers, and cited whoever showed up in enough of the retrieval buckets to survive the re-ranking pass.

The new Query Fan-Out Generator on jwatte.com maps the fan-out for any seed keyword across eight intent buckets. Before the tool, the explanation.

What fan-out actually looks like

A user asks an AI-answer engine: "best tools for cold email deliverability."

Behind the scenes the engine decomposes this into parallel sub-queries that look roughly like:

• Informational: "what is cold email deliverability", "how does cold email deliverability work", "why does cold email deliverability matter"
• Comparison: "best cold email deliverability tools 2026", "top cold email deliverability software", "alternatives to Instantly"
• Commercial: "cold email deliverability pricing", "is Instantly worth it", "MailReach reviews"
• Navigational: "Instantly official site", "MailReach founder", "Smartlead linkedin"
• Branded (if a brand is the query source): "Instantly vs MailReach", "what does Smartlead do"
• Voice: "can you tell me how to improve cold email deliverability", "what went wrong with my cold emails"
• Follow-up: "cold email deliverability case study", "cold email deliverability failure modes", "SPF DKIM DMARC for cold email"

Each of those is sent to a different retrieval index. The engine pulls the top few results per sub-query, de-duplicates the citation set, ranks by source quality + coverage + freshness, and emits one synthesized answer with three to eight citations at the bottom.

If your site ranks for the main keyword but not for any of the sub-queries, you don't get cited. Ranking on a single phrase isn't the job any more. Covering the fan-out is.

Why fan-out is different from "related keywords"

Classic SEO tools have sold "related keywords" for a decade. Fan-out is not that, and it's important to know why.

Related keyword tools pull co-occurring phrases from a bulk search-log index (what do users who searched X also search for within a session?). They're statistical cousins of the seed keyword. They may or may not be what an AI engine actually decomposes the query into.

Fan-out is the engine's own internal decomposition — an engineering decision baked into the retrieval pipeline, not a statistical artifact of user behavior. The patterns are stable: every AI-answer engine fans out across roughly the same eight intent buckets. The specific wording varies a little, the structure doesn't.

Which means the practical content-planning question changes. You don't need to "rank for 500 related keywords." You need to have one answer-shaped paragraph for each of the ~30 fan-out queries that cluster around your seed. Do that on one page and you show up in the citation list a retrieval-per-retrieval pass.

The eight intent buckets

The tool emits queries across eight buckets because that's the shape every mainstream AI-answer engine fans out into. They won't all fire on every query — an engine reads the seed and decides which buckets are relevant — but the full set is:

1. Informational. Definition, mechanism, importance. Fires first on any unfamiliar seed.
2. Comparison. Best-of lists, alternatives, pros and cons. Fires whenever the seed has competitive substitutes.
3. Commercial / transactional. Pricing, reviews, free vs paid, trials. Fires when purchase intent is sensed.
4. Navigational / entity. Who, where, official, profiles. Resolves the seed to a specific named entity.
5. Local. City-scoped variants. Only fires when geography is signaled or implied.
6. Branded. Reputation queries specifically about one brand. Fires whenever the seed mentions a brand or a brand is inferred.
7. Voice / question-shaped. Full-sentence versions of the information bucket. Often the exact prompt the LLM paraphrases internally before retrieval.
8. Follow-up / deep-dive. Second-turn questions the user is likely to ask after the first answer — retrieved preemptively to anticipate.

A good content plan hits at least six of these on the same page.

What the tool outputs

Type a seed keyword. Add optional modifiers (brand, competitor, city, audience). The generator expands into 30-60 fan-out queries grouped by intent. You get:

1. A visual intent-grid showing which bucket each query belongs to, color-coded
2. Per-bucket counts so you see where the fan-out is thin (e.g. "no local queries because you didn't supply a city")
3. Copy-all-as-text for pasting into a content brief
4. CSV download with bucket + query columns for bulk content planning
5. An AI content-plan prompt — paste into Claude or ChatGPT and get H2-shaped section titles, article-cluster outlines, and citation-ready one-sentence statements for every fan-out query

The prompt is the important part. One click, one paste, and an LLM turns the fan-out list into a content plan that covers the retrieval space. The tool doesn't do the writing; it does the architecture.

How to read a fan-out pack

Three things to look for when the tool emits a pack.

First — which buckets fired? If branded fired (because you supplied a brand) and reputation queries are lopsided negative, you have an AEO-reputation problem before you have a content problem. Fix the knowledge-graph entries first.

Second — which queries would need original research? Most fan-out queries are satisfied by summarization. But a few — usually the follow-up / deep-dive bucket — are better answered by first-hand data, case studies, or numbers that don't exist elsewhere. Those are the queries where you can lock in a citation by publishing original findings. Every AI engine prefers primary sources in its re-rank pass.

Third — which are orphan queries on your site? If your site already has content, run a site-crawl and check which of the fan-out queries already have a matching landing page. The gap list is the content pipeline.

The Index Coverage Delta tool already does orphan detection for indexing. A future tool (on the roadmap) pairs the fan-out generator with a sitemap import to automate orphan scoring against the fan-out list.

Why this tool is free and client-side

Paid AI-SEO platforms charge $99-$499/month for fan-out analysis, and their methodology is usually proprietary. Our tool runs rule-based templates derived from observing real AI-engine retrieval patterns across the big four (Google AI Mode, ChatGPT Search, Perplexity, Gemini). Rule-based templates aren't as elaborate as a paid tool's LLM-driven fan-out, but they cover the structural backbone — the eight intent buckets, the template patterns within each — which is 80% of what a content planner actually needs.

If you want to extend it, fork the page: the pattern library lives in PATTERNS in the inline script, and adding new templates is a one-line change.

Related reading

• AI Citation Readiness — score a single article for LLM citation probability across 14 on-page signals
• Entity Citation Radar — check whether high-authority sources (Wikipedia, Wikidata, Internet Archive) reference your brand
• AI Visibility Prompt Pack — run fan-out queries manually through ChatGPT, Claude, Perplexity, and Gemini; log share-of-voice in a CSV
• FAQ Harvester — pull every FAQ from the top 10 SERP results; complementary to fan-out for question-shaped content
• Why passage retrieval changed SEO — related piece on how AI engines rank at the paragraph level, not the page level

Fact-check notes and sources

• Google "AI Mode" query fan-out (announced 2025 I/O): blog.google/products/search/google-search-ai-mode-update
• Perplexity architecture on decomposition: docs.perplexity.ai/docs/search-api
• OpenAI ChatGPT Search retrieval notes: openai.com/index/introducing-chatgpt-search
• Microsoft "Copilot" retrieval decomposition approach: microsoft.com/en-us/research/publication/query-rewriting-retrieval-augmented-generation

---

The $100 Network covers how to build content libraries that cover the fan-out surface for dozens of related keywords across a site network. Pairs directly with this tool.