Query Fan-Out

Source

Kasra Dash (@Kasra_Dash) recapping James Dooley (@james_dooley) at the SEO.Domains Mastery Summit, after @MadsSingers asked how Dooley got cited for 5,000+ phrases in AI — x.com/Kasra_Dash/status/2098379471615111583 (11 Sep 2026).

He pulls the query fan-outs for every LLM. Gemini, ChatGPT, Perplexity, Claude. Then he writes the articles the fan-outs are looking for. That’s it. That’s the trick.

Follow-ups in the same thread: backlinks and listicles don’t save you if you’re missing from the hidden searches; three content buckets; tool disclosure — Rank OS is Kasra’s (getrankos.com). This page is the mechanism + how we’d write to it, not a paste of his product.

One-sentence TL;DR

Treat the model’s follow-up questions as the brief. Cover the entity, attributes, comparisons, price, caveats, “best for X,” and this year’s facts on the same cluster — because those are the searches it actually runs.

What fan-out is (under the hood)

It is not a fancy keyword trick. It is the retrieval step that decides whether ChatGPT, Gemini, Perplexity, Claude, or Google AI Overviews even see your page.

When someone types one question, the model usually does not search that exact phrase once. It splits the question into related sub-queries, runs those searches in parallel, pulls passages from the results, then synthesizes one answer and cites the sources that covered the cluster best. Those hidden sub-queries are the fan-outs.

Google named the technique in public for AI Mode at I/O 2025 : “breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf.” Deep Search is the same idea taken further — hundreds of searches, then a cited report. The same pattern shows up in ChatGPT search, Gemini, Perplexity, and others, with different volume and style.

Not the same word as Postbridge fan-out (one tweet → many socials). That is distribution. This is retrieval.

A simple example

User prompt: “best CRM for a small agency.” Typical fan-outs the model invents (you never see them):

• CRM pricing for small teams
• CRM with project management
• agency CRM integrations
• HubSpot vs Pipedrive for agencies
• CRM reviews 2026
• easiest CRM for non-technical staff

It retrieves pages for each of those, then builds one answer. If your article only targets the head term, you miss most of the retrieval surface. Pages that cover several of those angles get cited more often.

Volume (rough, published ranges)

The exact strings are unstable from run to run. The types of questions are much more stable: comparisons, reviews, pricing, “is it legit,” current-year freshness, implied follow-ups. Don’t memorize one Rank OS export. Own the types.

What the tweet is saying

Dooley’s 20-second answer, as Kasra summarized it:

1. Pull the query fan-outs from each LLM (Gemini, ChatGPT, Perplexity, Claude).
2. Write the articles those fan-outs are looking for.

The line that stung: you can have the best backlinks and be on every listicle. If you’re not in the fan-outs, the model still isn’t sure you exist.

Three buckets. Every fan-out lands in one:

1. Articles on your own site
2. Self-corroboration — your site says it, a second source repeats it
3. Hybrid — lives on your site and gets picked up elsewhere

Listicles are the cherry, not the cake. Traditional backlinks still matter for classic SERPs; they help less if the hidden searches never retrieve you.

The +161% number (source, not folklore)

Surfer SEO (reported Search Engine Land, Dec 2025): ~10k keywords, ~33k Gemini-extracted fan-outs, ~174k URLs.

• Pages ranking for the head term and at least one fan-out: 161% more likely to be cited in AI Overviews than pages ranking only for the head term.
• Spearman 0.77 between number of fan-out rankings and citation likelihood.
• Those “both” pages were ~51% of AIO citations that also ranked organically; head-term-only was ~20%.
• Fan-out-only pages were cited more often (~29%) than head-term-only (~20%).

Correlation, not a guarantee. Surfer’s own writeup: don’t chase every synthetic string — own the topic. That’s the keep.

Classic SEO vs fan-out / GEO

How to write for it (without a new SaaS bill)

Tools people use: Rank OS, Surfer, Semrush-style fan-out features, browser extensions that expose ChatGPT’s searches, or a custom prompt/API capture. Kasra disclosed Rank OS is his. We don’t have that pack. Day one is a sitting, not a subscription.

1. Pick one money prompt (e.g. “best CRM for a small agency,” or for us: “SEO agent for GSC,” “headless domains for AI agents”).
2. Ask 2–3 models: “What follow-up searches would you run to answer this well?” Save the cluster. Repeat next week — strings drift, types shouldn’t.
3. Map types, not phrases: definition, attributes, comparison, price, caveats / “is it legit,” “best for X,” current year, how-to, vs.
4. Cover several types on one URL (or a tight cluster with internal links). Thin head-term posts lose the retrieval surface.
5. Self-corroboration: a second domain (partner, docs, GSC-visible property) repeating the same entity facts. Not a 50-site PBN.
6. Judgment still belongs in seoloop: changelog, 14-day lookback, reverse losers. Fan-out is the brief. The loop is still the grader.

How this maps here

What we keep vs skip

• Keep: model follow-ups as the brief; types over strings; one URL covering several angles; self-corroboration; seoloop Memory/Judgment around the cluster.
• Verify: whether RankingSolution can store a fan-out cluster per domain; whether GSC already shows the long modifiers (often zero MSV — Seer: ~95% of Gemini 3 fan-outs had no search volume).
• Skip: buying Rank OS on day one; chasing every unstable string; treating listicles as the strategy; confusing this with social fan-out; Always Allow a bot to mass-publish “fan-out articles.”

Related on this site

• Agentic SEO loop → RankingSolution — Access / Action / Memory / Judgment / Reversal. This page is the retrieval brief that loop should grade.
• GrokBot as an SEO agent — named bot, page-2 GSC sprint, Require Approval on publish
• Backlinks + RS brain — still the classic layer; Dooley’s sting is “not sufficient”
• SEO audit as a product · DataForSEO · IndexNow
• 23 agent growth plays — GEO-adjacent distribution

Primary: @Kasra_Dash — Dooley’s 20-second answer · Google I/O 2025 AI Mode · Surfer fan-out study