Query fan-out keyword research maps the smaller questions an AI assistant searches for when it answers one large question from a buyer. Instead of starting from single keywords, you start from a complete buyer question, break it into its facets (cost, fit, comparison, requirements, risks, location), collect the sub-questions for each facet from real sources, and assign each one to a page or section. The result is a content plan that matches how AI Mode, ChatGPT and Perplexity actually retrieve information.
Why keyword lists are no longer enough
When Google introduced AI Mode in 2025, it explained that the system breaks a question into subtopics and runs many searches at once, a technique it calls query fan-out. Other assistants rewrite questions in similar ways. A buyer who asks “Which payroll software should a 50-person company in Pune use if we already use Tally?” never types “payroll Tally integration”, but the assistant may search for exactly that. Keyword tools, which report what people type, miss most of these sub-queries.
Step 1: write the full buyer questions
Collect 10 to 20 complete questions from sales calls, enquiry forms and support emails. Keep the conditions buyers mention: size, budget, location, existing tools, deadlines. These conditions are what the assistant turns into sub-searches.
Step 2: break each question into facets
| Facet | Typical sub-questions |
|---|---|
| Definition | What is it? How does it work? |
| Fit | Is it right for a company of our size, industry or location? |
| Cost | What does it cost? What drives the price? |
| Comparison | How does it compare with the alternatives? |
| Requirements | What do we need before starting? What does it integrate with? |
| Process and time | How long does it take? What are the steps? |
| Risks | What can go wrong? What are the limits? |
| Provider choice | Who offers it? How do we choose? |
Step 3: collect real sub-questions for each facet
- “People also ask” and related searches for the main question and each facet
- Search Console queries with five or more words, which often show sub-questions you already appear for
- AI assistants: ask the full question, note which aspects the answer covers and which sources it cites for each
- Ask an assistant directly what searches it would run to answer the question, and treat the list as ideas to verify, not data
- Sales and support teams: the follow-up questions buyers ask after the first call
Step 4: map sub-questions to pages
Put everything in one sheet with these columns:
Buyer question | Facet | Sub-question | Source | Existing page? | Action | Target page or section
For each sub-question, decide whether it needs its own page, a section in an existing page, or nothing because it is already answered well. Questions with the same intent share one page.
A worked example
For the question “How should a B2B company in India start with generative engine optimisation?”, the facets and the pages on this site that answer them look like this:
| Facet | Sub-question | Where it is answered |
|---|---|---|
| Definition | What is GEO? | GEO services |
| Comparison | How is GEO different from SEO? | GEO vs SEO |
| Requirements | Can AI crawlers reach our site? | AI crawler checker |
| Process and time | How long does GEO take? | GEO timeline |
| Risks | Should we block GPTBot? | Decision guide |
| Measurement | How do we know it works? | Tracking mentions |
| Provider choice | How do we assess where we stand? | GEO Readiness Score |
Each facet has one clear owner page, and the pages link to each other, so an assistant searching any part of the question finds a direct answer.
Step 5: write for the sub-question
- Use the sub-question, or a close version, as the heading.
- Answer it in the first sentence, with the condition that makes the answer true.
- Include the buyer’s conditions where relevant: size, budget, location, tools.
Step 6: check coverage
Run the full buyer questions through ChatGPT, Perplexity and Google every month. For each facet, note whether one of your pages is cited. Gaps show which sub-questions still need a better answer.
Frequently asked questions
Can I see Google’s actual fan-out queries?
No. Google has not published them, and Search Console does not label them. The method above approximates them from real sources.
Does this replace keyword research?
It adds to it. Keyword data still shows demand and wording. Fan-out mapping shows the structure of the questions behind the keywords.
How many sub-questions should one page answer?
As many as share the same intent, usually three to eight. More than that often means the page is covering several topics.
Sources
- Google, introduction of AI Mode and query fan-out, Google I/O, May 2025
- Google Search Central, “AI features and your website”