Fanout Queries: The Complete Guide to AI Visibility & GEO

Languages / Diller: English (this page) · Türkçe · Deutsch

Summary (AI TL;DR)

Fanout queries are the hidden sub-questions an AI model (ChatGPT, Gemini, Perplexity, Google AI Overview) searches behind the scenes while answering a prompt. If your brand doesn't show up in those sub-questions, it won't appear in the model's final answer either. GEO (Generative Engine Optimization) is the practice of seeing those sub-questions and owning the most frequent — and cross-model shared — ones with content. This guide walks through how to find fanout queries, turn them into a content roadmap, and make AI visibility a measurable channel.

Search is no longer a list of ten blue links. When a user asks ChatGPT "what's the best family holiday tour company?", the model doesn't answer directly — it breaks that question into dozens of sub-questions, researches each one, and builds an answer from the evidence it gathers. Those hidden sub-questions are fanout queries, and they are the real signal that decides whether your brand appears in an AI answer.

This guide covers GEO (Generative Engine Optimization), AEO (Answer Engine Optimization), and AI Visibility, explains why fanout queries are the next layer of SEO, and shows step by step how to turn that data into a concrete content strategy.

This guide is based on Maya's GEO field experience and fanout methodology. Customer data is confidential; brand names in examples are anonymized. The methodology applies as-is.

Why did visibility change in the AI-search era?
What's the difference between GEO, AEO, and LLM SEO?
What is a fanout query and why does it matter?
How does an AI model "fan out" a question?
Why are fanout queries replacing classic keyword research?
Fanout dynamics by market: local search behavior
How do you find fanout queries? (Step by step with Maya)
From fanout data to content strategy: a 7-step playbook
"Shared across models" — the questions models agree on
Worked example: fanout analysis for a travel brand
The fanout → brief → publish → measure loop
The 10 most common GEO mistakes
Frequently asked questions (FAQ)
Conclusion: measurable AI visibility

1. Why did visibility change in the AI-search era?

For a decade the definition of digital visibility was clear: rank on Google's first page. A user searches, sees ten blue links, clicks one. The entire SEO discipline was built on that behavior.

AI assistants broke the chain. Today, when a user asks ChatGPT, Gemini, Perplexity, or Google AI Overview a question, they:

Don't see a list of links — they see one synthesized answer.
Often don't click — the answer itself is enough (zero-click).
Only hear about the brands the model chose — a brand not named in the answer effectively doesn't exist for that user.

The implication is clear: the question is no longer "what's my Google rank?" but "does AI recommend me?" The name for this new discipline is GEO — Generative Engine Optimization.

How big is AI search now?

AI-assisted search is growing fastest in high-intent categories — travel planning, product comparison, financial and health questions. Commercial-intent queries like "best X brand", "X vs Y", "is X reliable" are rising quickly. For businesses that means both risk and opportunity: if you're absent from the AI answer, a competitor is recommended in your place.

2. What's the difference between GEO, AEO, and LLM SEO?

These terms are often conflated. Briefly:

SEO (Search Engine Optimization): Content and technical optimization for organic ranking in classic search engines (Google, Bing).
AEO (Answer Engine Optimization): Making content "answerable" so answer engines (AI Overview, featured snippets, voice assistants) take their single, clear answer from you.
GEO (Generative Engine Optimization): Optimizing so generative AI models (ChatGPT, Gemini, Claude, Perplexity) cite and recommend your brand in their answers.
LLM SEO: Largely synonymous with GEO; emphasizes visibility inside large language models specifically.

GEO's biggest difference from SEO is the measurement layer. In SEO you see rank; in GEO you have to see the model's answer, whether it mentions you, which sources it cites, and — crucially — which sub-questions it searched behind the scenes. That's exactly where fanout queries come in.

3. What is a fanout query and why does it matter?

A fanout query is a sub-question an AI model generates and searches while answering a user's prompt. The model "fans out" a broad question into narrower, searchable pieces and grounds its answer in what those searches return.

A simple example. User prompt:

"Can you recommend a family-friendly beach holiday for us?"

Fanout queries the model might generate behind the scenes:

"family-friendly seaside hotels"
"all-inclusive holidays for families"
"hotels with kids' club"
"baby-friendly resorts for ages 0–6"
"beachfront family hotel early booking"

The user never typed these sub-questions. But the model searched them and grounded its answer in the content those searches surfaced. If your content doesn't appear for those sub-questions, you're absent from the final answer.

Why is a fanout query so critical?

Because fanout queries reveal the model's real search intent. Classic keyword research tells you what people type into a search box. Fanout queries tell you what the AI searches on your behalf — the information it needs to build an answer. It's the only direct signal that makes a "content gap" visible:

If a model searched a sub-question 20 times and never found you, that miss appears in no ranking report — yet it's exactly why you aren't cited.

4. How does an AI model "fan out" a question?

It varies by model, but the general flow is:

Intent parsing: The model interprets the user's true goal (informational, comparison, purchase?).
Sub-question generation: It splits the question into the narrower searches needed to ground an answer.
Retrieval / grounding: It researches those sub-questions via a search layer (web search, internal index, retrieval).
Source selection: It picks the results it finds trustworthy and relevant.
Synthesis: It builds a single answer from the selected sources, often with citations.

Perplexity exposes its "related questions" explicitly; Gemini carries search queries in grounding metadata; Google AI Overview runs multiple Google searches behind the scenes. The common thread: every answer has a sub-question set beneath it, and that set is the real battleground for your visibility.

Is fanout the same as RAG or query expansion?

Close relatives, but not the same. Query expansion widens a single query with synonyms in classic search. RAG (retrieval-augmented generation) is the architecture by which a model pulls in external knowledge. Fanout is the behavior of splitting one user question into multiple distinct sub-questions, each searched separately. Fanout is the layer inside RAG that is most actionable for GEO.

5. Why are fanout queries replacing classic keyword research?

Classic keyword tools (volume, difficulty, CPC) model human behavior. But machines now do a large share of searching on your behalf. The two worlds compared:

Dimension	Classic keyword	Fanout query
Source	Query typed by a human	Sub-question generated by the model
Metric	Monthly search volume	Model lookup frequency
Intent	General	Answer-specific, narrow
Action	Page targeting	Closing content gaps
Visibility	SERP position	Being mentioned in the answer

Fanout queries are the AI-era equivalent of keyword research — but grounded in the model's actual behavior, not estimates. That lets you prioritize content with far higher precision.

6. Fanout dynamics by market: local search behavior

Fanout analysis has market-specific nuances. Some universal ones worth calling out:

Language and brand-matching traps

Brand names are often written multiple ways — split into two words, joined, or with suffixes/inflections. If a model's answer mentions the brand in its spaced form but your system only searches the joined form, you miss the match and your mention rate falsely reads 0%. Defining brand aliases (all spelling variants) is critical for accurate fanout and mention tracking.

Local intent and seasonality

Fanout queries carry strong seasonal and local intent: "early booking", "last minute", "installment payment", "cancellation terms", "reviews". These patterns are what the model searches most often when answering local users and should sit at the center of your content strategy.

Provider diversity

Users lean on different assistants — ChatGPT, Gemini, Perplexity, Google AI Overview. Their fanout behavior differs; appearing in one but not another is common. That makes the provider breakdown and the queries models agree on especially important.

7. How do you find fanout queries? (Step by step with Maya)

Collecting fanout queries by hand is effectively impossible: the data lives in each model's grounding metadata, scattered across hundreds of runs and providers. Maya collects it automatically, normalizes it, and turns it into a single ranked list.

Steps:

Define your prompts. Add the real user questions about your brand (discovery prompts): "best X", "is X reliable", "X vs Y".
Maya runs the prompts across every model. It asks the same questions to ChatGPT, Gemini, Perplexity, and Google AI Overview.
Fanout is captured automatically. The sub-questions each model searched are extracted, normalized, and deduplicated.
Results in the Fanout Queries view. Sub-questions arrive ranked by lookup frequency; each row shows how many times it was searched (×N), how many distinct prompts triggered it, and the provider breakdown.
Filter and export. Narrow by provider, by search term, or with the "Shared across models" filter (sub-questions 2+ models searched); export as CSV and hand your content team a ready-made brief.

The Fanout Queries feature lives in the Monitoring section, and the same data is available programmatically via the Maya MCP tool get_fanout_queries.

8. From fanout data to content strategy: a 7-step playbook

The fanout list is interesting on its own; the real value is turning it into a content roadmap. Seven field-tested steps:

Step 1 — Flag high-frequency, low-coverage queries

Sub-questions the model searches often (high ×N) but where you don't appear are your most urgent content gaps. They top your priority list.

Step 2 — Find the core with "Shared across models"

Questions searched by 2+ models are the consensus core. Win those first: publish one page and you're more likely to gain visibility across ChatGPT, Gemini, and Perplexity at once.

Step 3 — Cluster sub-questions into themes

Don't work hundreds of sub-questions one by one — group them into themes: "price/payment", "trust/reviews", "comparison", "returns/cancellation", "early booking". Each cluster is a pillar-content candidate.

Step 4 — Write "answerable" content for each cluster

In GEO, content must be clear enough for the model to quote directly: heading questions (H2/H3 = user question), a short, direct answer paragraph, then detail. Tables, lists, and definition boxes raise the odds of being quoted.

Step 5 — Strengthen brand aliases and entity signals

Mention your brand in all spelling variants, alongside your category and related entities. Structured data (Organization, FAQPage) and consistent NAP info help the model recognize you as an entity.

Step 6 — Publish and re-measure the fanout

After publishing, re-run the prompts. Do you now appear for those sub-questions? Did mention rate and citations rise? GEO is an iterative loop.

Step 7 — Track competitor fanout

Which competitors appear in the sub-questions? Competitor names surfaced in fanout text are direct input for your comparison and differentiation content.

9. "Shared across models" — the questions models agree on

Not every sub-question in the fanout list is equal. Some are one-off tangents a single model searched once; others are core questions every major model searches for the same topic.

The "Shared across models" filter isolates exactly these: only sub-questions two or more models searched together. In practice the list is short — maybe 10–20 of hundreds. But that short list is gold: winning it means owning the "must-answer" topics every assistant agrees on, not chasing one model's quirk.

These consensus questions typically cluster around trust ("is X reliable", "X reviews"), price/payment ("installments", "returns"), and comparison ("X vs Y").

10. Worked example: fanout analysis for a travel brand

Let's walk the whole flow with the brand name kept confidential. Say you're an online tour/holiday brand.

Input prompts:

"What are the best holiday tours?"
"Best tour company for a family holiday?"
"Is company X reliable?"

Fanout queries Maya collected (lookup-ranked, brand anonymized):

"early-booking hotel deals" — ×51 (ChatGPT, Gemini, Perplexity)
"all-inclusive family-friendly hotels" — ×30 (Gemini, Perplexity)
"company X reviews complaints" — ×24 (ChatGPT, Perplexity)
"tour cancellation and refund terms" — ×20 (Perplexity)
"installment holiday payment options" — ×15 (ChatGPT)

Interpretation:

"Early booking" is both high-frequency and shared across three models → top-priority pillar content.
"Reviews/complaints" is a reputation signal → needs a transparent, trust-building reviews/FAQ page.
If you don't appear for "cancellation/refund terms", a strong cancellation-guarantee page earns direct visibility.

Action: these five sub-questions become five clear content briefs, each targeting a topic the model actually searches — driven by data, not guesswork.

11. The fanout → brief → publish → measure loop

GEO isn't a one-off; it's a continuous loop:

Discover: Extract the sub-questions models search with Fanout Queries.
Prioritize: Pick the high-frequency, shared, low-coverage ones.
Brief: A content brief per sub-question cluster (CSV export is direct input).
Create & publish: Clear, quotable content with structured data.
Re-scan: Re-run prompts, measure the change in mention/citation.
Iterate: Target new fanout queries as gaps close.

Brands that operationalize this loop turn AI visibility from luck into a measurable channel.

12. The 10 most common GEO mistakes

Not defining brand aliases — the brand appears in its spaced form but you search the joined form, so the match is missed and mention reads 0%.
Focusing only on prompts, ignoring fanout — the real intent hides in the sub-questions.
Looking at a single model — appearing in ChatGPT but not Gemini/AI Overview is very common.
Not writing "answerable" content — long, unstructured text isn't quoted; you need a clear question-heading + concise answer.
Missing structured data — no Organization/FAQPage schema makes it harder for the model to recognize you.
Ignoring reputation/review questions — "is it reliable" is among the highest-intent fanout queries.
Publishing without measurement — not re-scanning fanout after publishing to see the impact.
Neglecting llms.txt and robots settings — unknowingly blocking AI crawlers.
Not tracking competitor fanout — failing to turn competitors surfaced in sub-questions into comparison content.
Copying content across markets — local intent (installments, holidays, early booking) doesn't exist in generic templates.

13. Frequently asked questions (FAQ)

Is a fanout query the same as a keyword?

No. A keyword is a phrase a human types into a search box. A fanout query is a sub-question an AI model searches on your behalf while answering. Fanout is the AI-era version of keyword research — grounded in the model's real behavior, not estimates.

Which models do fanout queries come from?

Every provider that exposes fanout: chiefly ChatGPT, Gemini, Perplexity, and Google AI Overview. Provider names are normalized, so the same engine collected different ways is shown as one model.

What does "shared across models" mean?

Sub-questions two or more models searched for the same topic. These are the core topics to prioritize, because every major assistant tries to answer them — win those pages and you show up everywhere.

Is GEO replacing SEO?

No, it's layered on top. Solid technical SEO and quality content remain the foundation; GEO is the new layer that turns them into visibility inside AI answers. The two work together.

What's the first step for AI visibility?

Measure first. Run your real brand questions across every model and see your mention rate, citation status, and fanout queries. You can't optimize what you can't measure.

How do I export the fanout data?

Any filtered view (provider, search term, or "shared across models") can be exported as CSV and turned directly into a content brief. Programmatic access is also available via the Maya MCP tool get_fanout_queries.

14. Conclusion: measurable AI visibility

Visibility in AI search is no longer a guessing game. Between the user's question and the model's answer there's a hidden layer — the sub-questions the model searches behind the scenes — and that layer decides whether your brand gets mentioned.

Seeing your fanout queries moves GEO from "let's publish content we like" to "let's target the topics the model actually searches, with data." A fanout strategy built on correct brand aliases, local intent, and cross-model consensus turns AI visibility into a measurable, repeatable growth channel.

The next step is simple: run your brand's questions across every AI model, see what's searched behind the scenes, and own the topics the model is looking for.