ChatGPT Fan-Out Queries Are Turning SEO Into Evidence Design
Lily Ray’s Search Engine Journal piece points to a useful shift: AI search may reward pages that prove experience, expertise, authority, and trust across many related subqueries, not just one keyword target.
TL;DR: If ChatGPT search fans out a user prompt into related queries, the practical SEO move is to build pages that answer the whole evidence trail, not just the head keyword.
What are ChatGPT fan-out queries really testing?
Lily Ray’s Search Engine Journal piece, “What We Can Learn From Evolving ChatGPT Fan-Out Queries,” argues that ChatGPT’s search behavior is starting to look like E-E-A-T: experience, expertise, authoritativeness, and trust.
That is a useful frame, with one caveat. E-E-A-T is not a scoring knob you can optimize directly. It is a quality pattern. A way to describe what trustworthy information tends to look like.
Fan-out queries make that pattern more operational.
Instead of treating a user prompt as one query, an AI search system can break it into related searches. A broad question becomes a cluster of checks: definitions, comparisons, examples, risks, pricing, alternatives, current claims, first-hand reviews, and contradictory viewpoints.
That changes the content target. The old SEO move was often, “Can this page rank for the phrase?” The AI search move is closer to, “Can this page survive the follow-up investigation?”
A page that makes a claim but does not show experience may get exposed. A page with generic advice but no named author, no original examples, and no citations may be easy to skip. A page that answers only the obvious query may be less useful than a page that anticipates what the assistant needs to verify next.
Why does this feel like E-E-A-T?
Because fan-out rewards the same signals humans use when they are unsure.
If I ask for the best way to migrate a site, I do not only want a checklist. I want to know who has done it, what broke, what tradeoffs they made, and whether the advice matches my case. If I ask whether a medical, legal, financial, or technical claim is true, I want stronger proof than polished copy.
That is where E-E-A-T becomes less abstract. Experience is the screenshot, teardown, field note, failure mode, benchmark, support ticket, customer pattern, or code sample. Expertise is the correct framing and the absence of lazy overclaims. Authority is other credible people or institutions pointing to you. Trust is the boring stuff: dates, corrections, disclosures, citations, clear authorship, and claims that do not outrun the evidence.
The catch is that AI search can remix these checks at query time. You may not know which supporting question gets asked. That means thin “SEO pages” are in a worse spot than they look. They can still answer the main phrase, but fail the surrounding checks.
What should builders and publishers change?
I would not read Ray’s argument as “write longer posts.” Length is not the point. Coverage is not the same as completeness.
The better move is evidence design. Start with the user’s real decision, then map the doubts an assistant would need to clear before giving an answer. What would it verify? What would it compare? What claim would need a first-party source? What would a skeptical reader ask next?
For product pages, that means clearer docs, real constraints, changelogs, integrations, security notes, and examples. For service businesses, it means named people, real work, process detail, client context where allowed, and proof that the advice came from practice. For publishers, it means fewer generic explainers and more original reporting, testing, expert review, and explicit sourcing.
This also changes internal linking. The job is not only to move PageRank around. It is to help a machine and a person follow the evidence path. A strong page should connect to the proof behind its claims, not just adjacent content written for another keyword.
Practitioner’s take: take one page that matters and run a fan-out audit. Write the main user question at the top, then list the five to ten questions an assistant would need answered before trusting your page. Add first-hand evidence where you have it, cite primary sources where you do not, and remove claims you cannot support. The missed catch is that AI search does not only want an answer. It wants confidence in the answer.