Fan-outs tell you what to write. Corroboration is what gets you cited

If you have read the searches an AI runs before it answers you, you know what a fan-out is: before an engine answers a question that depends on anything current, it runs its own background searches, and the ones it discloses are a direct read on how it frames your category. That piece is about what fan-outs are and how to tell a real disclosed query from an invented one. This one is about what to do with the real ones.

The short answer is two jobs, in order, and most teams only do the first. Cover the subjects the fan-out reveals. Then, more importantly, earn the corroboration that turns a covered subject into a citation. Skip the second and you get a tidy content library that AI still does not cite.

Fan-outs are the keyword research of GEO

Traditional keyword research tells you what people typed into a search box. A disclosed fan-out query tells you something sharper: what an AI model decided it needed to search to answer a question about your category. It is the model's own decomposition of the topic, written by the model, for the purpose of actually finding an answer.

That is a different instrument. Where a keyword list is a record of human demand, a fan-out set is a map of machine intent: the sub-questions the engine breaks your category into, the comparisons it reaches for ("vs [competitor]", "for agencies", "pricing 2026"), the modifiers it treats as part of the same question. Each one is a subject you can choose to own, ignore, or lose to whoever already owns it. In SEO you research the query and write the page. In GEO the model hands you its research, and the question becomes whether the web agrees you are the answer.

Job one: cover the subjects, answer-shaped

Once you can see the real fan-out subjects, the first move is the familiar one, done well. For each subject worth owning, publish content built to be lifted into an answer:

• One subject per page, with a heading that matches the question the fan-out query implies.
• The answer up top, stated directly, then supported. Engines lift the sentence that already says the thing; they rarely infer it for you.
• A specific, attributable claim: a dated figure, a named comparison, a concrete definition. Specificity is what gets quoted. "Many buyers use AI" loses to "in [source], X% did."
• Structured and current: lists, tables, schema, and an honest dateModified when you actually update it.

This is real work and it is necessary. It is also where most GEO content strategies stop, and it is why they stall. Covering the subject makes you eligible. It does not make you the answer.

Job two: earn corroboration, which is the part that gets you cited

Here is the belief that should reshape how you spend your effort: AI engines favor claims they can corroborate. A model is not looking for the one page that asserts something. It is looking for the thing that many independent, trusted sources agree on, because agreement is how it estimates what is true and safe to repeat. Consensus reads as authority. A lone page, however well written, reads as a claim.

An AI answer is a consensus engine. It does not cite the page that says it best. It cites the thing the web agrees on, and names a source that states it clearly.

This is the hardest break from classic SEO. In SEO you optimize a page you control. In GEO you also have to move something you do not fully control: what the rest of the web says about your category, and whether your claim shows up, consistently, on the sources the engine already trusts. That is editorial coverage, community discussion, directories and review profiles, analyst and comparison pages. Independent 2026 research puts the large majority of AI citations on earned media rather than on-page work, which is the same lesson from the other direction: corroboration drives citations, not formatting tricks.

So for each fan-out subject you decide to own, the corroboration question is concrete: does the claim you want to be cited for appear, stated the same way, on more than one source the engine trusts for this topic? If the only place it lives is your own page, you have covered the subject and lost the citation. The work is to get it echoed: pitch the data to a publication, answer the question where your buyers actually discuss it, make sure your directory and review profiles state the same facts, earn the comparison page that names you. Consistency across sources is the signal. Contradiction across sources is what makes a model hedge and omit you.

The workflow, end to end

Put together, the tactical loop is six steps, and it is a loop, not a launch:

• 1. Pull the real fan-out queries for your category, using only genuinely disclosed ones. Treat an inflated count for an engine that discloses nothing as a guess, not data.
• 2. Map branded vs unbranded, and check which brand each query actually surfaces. The most valuable subjects are usually unbranded ones where a rival owns the framing.
• 3. Prioritize the subjects where you are absent and a competitor is named. Those are addressable targets, not a vague "rank higher."
• 4. Cover them, answer-shaped, one subject per page, owning a specific claim.
• 5. Earn corroboration for those claims across the sources the engine trusts for the topic. This is the job that actually moves the citation.
• 6. Track the rate of your brand-naming share on those fan-out queries over time, per engine, with the sample size behind it, so you can tell whether any of it worked.

How llemmy tracks and powers this

llemmy is built to run exactly this loop honestly, end to end.

The Real Fan-outs view surfaces the genuine, provider-disclosed background searches AI engines ran while answering your tracked prompts. It labels every engine by what that engine actually discloses rather than blending real and invented queries into one impressive-looking number. Your brand-naming share of the fan-out set is reported as a proportion, so it carries a sample size and a 95% confidence interval, never a bare percentage: a handful of disclosed queries is a different claim than a few hundred, and the view says which you have. Each real query is one click from rank tracking, so a fan-out finding becomes a tracked subject with its organic rank and its AI Overview citation status, not a screenshot.

The corroboration check: the Citations view shows the actual source mix behind your answers, owned pages versus third-party and community sources, so you can see whether the consensus is forming around you or a rival. Illustrative data.

Then the Citations view closes the loop on corroboration. It shows the sources AI engines actually cite about you, and the citation source mix (your own pages versus competitor, community and earned sources), so you can see whether the corroboration you are chasing is landing, or whether the consensus is still forming around someone else. Share of voice, tracked per engine and refreshed daily, tells you whether covering the subjects and earning the corroboration is moving your slice of the answer over weeks, which is the only timescale that means anything here.

Read it honestly

One discipline holds the whole thing together. When you roll fan-out coverage up into a rate, "we are named in 38% of the fan-out searches about this topic," that is a proportion, and it needs its sample size to mean anything. Read it as a daily-refreshed signal you watch move over weeks, not a number you screenshot once and declare victory on. The same goes for corroboration: earning a mention is a step, and whether it changed your citation share is a question you answer by watching the rate, not by assuming. For how we keep every number honest, see how we measure.

How llemmy helps

llemmy shows you the real searches AI runs about your category, tells you which subjects a rival is winning, and then measures whether your content and your corroboration are actually moving your share of the answer, across ChatGPT, Claude, Gemini, Perplexity and Google AI, with a sample size and a 95% confidence interval on every rate and a daily refresh. It turns fan-outs from a curiosity into a content and PR plan you can hold accountable. Run a free GEO audit or start tracking free to see what your category's models are searching for, and who they agree on.

FAQ

How are fan-out queries different from keyword research?

A keyword tool tells you what people typed into a search box. A disclosed fan-out query tells you what an AI model decided it needed to search to answer a question about your category. It is the model's own decomposition of the topic, so it maps the sub-questions, comparisons and modifiers the engine treats as part of the answer. Cover those subjects and you are writing for the questions the engine is actually asking, not the ones you guessed.

Is publishing content for a fan-out subject enough to get cited?

No. Covering the subject with a clear, answer-shaped page is necessary but not sufficient. AI engines favor claims they can corroborate across independent sources. A single page asserting something rarely wins the citation on its own; the same fact stated consistently across sources the engine already trusts builds the consensus and authority that gets you named. Content is job one, corroboration is the bigger job two.

How do I turn fan-out queries into a content plan?

Pull the disclosed fan-out queries for your category, map which mention your brand versus a rival, and prioritize the unbranded subjects where a competitor currently owns the framing. Write answer-shaped content for those subjects, then earn corroboration for the same claims on sources the engines trust. Finally, track whether your brand-naming share on those fan-out queries moves over time, per engine, so you know the work landed.

How does llemmy help with fan-out queries?

llemmy's Real Fan-outs view surfaces the genuine, provider-disclosed background searches AI engines ran while answering your tracked prompts, labels each engine honestly by what it actually discloses, and reports your brand-naming share of the fan-out set as a proportion with a sample size and a 95% confidence interval. Each real query is one click from rank tracking, and the Citations view shows the source mix behind your answers so you can see whether the corroboration is landing, all refreshed daily.

By the llemmy team, July 2026. The corroboration point draws on 2026 research attributing the large majority of AI citations to earned media; figures vary by source and method and should be read as directional. Related reading: Query fan-out: the searches an AI runs before it answers you, Earned media drives most AI citations, and The AI Citation Gap.