Let me explain the most important idea in AI search with Google's own example. Someone asks: "how to fix a lawn that's full of weeds". Before answering, the model sends several related searches at the same time, like "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn". That's query fan-out.
What query fan-out is, in Google’s words
Google defines it as a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query. It works together with grounding: AI Overviews and AI Mode retrieve pages from Google's index with the core ranking systems, read them, and write an answer with links to the pages that support it.
The result is a wider and more varied set of links than a classic results page. That's an opportunity: a page that answers one of those follow-up questions really well can be cited even if it would never rank first for the original query.
The trap: one page per sub-question
The obvious reaction is to create a page for every fan-out query. Don't. Google says that creating separate content for every possible way people might search, fan-out queries included, primarily to manipulate rankings or AI responses, violates its scaled content abuse policy. It also doesn't work: AI systems understand synonyms and several topics on one page, and more pages don't make a site more relevant.
User first: plan for the next question
The way I use fan-out is the opposite of keyword stuffing. I start from the person. If they ask the first question, what will they need to know next to actually solve their problem? Those follow-up needs are, very often, the same ones the model fans out to.
- Write down the main question your customer brings to you.
- List the five to eight follow-up questions you hear in real conversations: costs, risks, alternatives, timing, how to start.
- Answer them on one deep, well-organized page, with clear headings, in the order a person would ask them.
- Add what only you know: a case, a number from your own work, a mistake you have seen clients make.
- Split into another page only when the topic really is different, not when the wording is.
This is how I built my own FAQ page: real questions business owners ask me, grouped by topic on one page, instead of dozens of thin pages.
How to know if it works
Search Console now has a Generative AI performance report that shows how often links to your site appear in AI Overviews and AI Mode, by page, country and device. Watch which pages enter and which don't. Usually the ones that make it are the ones with the most first-hand detail.
An example from a service business
Let's say you run a small accounting firm in Alicante, and a foreign founder asks AI Mode how to set up a company in Spain. The model could fan out to related questions about the types of company, the costs, the identification number a foreigner needs, how long it takes and how to open a bank account. A page that only explains the legal forms answers one of those. A page that walks the founder through the whole journey answers most of them.
And here's where your experience wins: what usually delays the process, which documents clients forget, what you would do differently if it were your own company. No model can make that up, and it's exactly what makes a page worth citing.
Fan-out works with meaning, not keywords
Google says AI systems understand synonyms and general meaning, so you don't need to write a section for every way a question can be phrased. It also says you don't need to chunk content into small pieces, and that there is no ideal length. In practice I write naturally: one clear heading for each real sub-question, the answer right below it, and an example from real work.
How I map fan-out for a client
- Collect real questions from sales calls, emails, WhatsApp messages and support tickets.
- Group them by what the person is trying to do: understand, compare, decide or act.
- Check what already exists on the site, so we improve a page instead of duplicating it.
- Decide one page per real difference in the topic, never per wording.
- Add proprietary input to every section: a case, a number, a decision.
- Review, publish at a human pace and watch the Search Console report.
Fan-out and local businesses
Google says your Google Business Profile and Merchant Center feeds also feed AI responses, so keep them up to date. And be careful with city pages: a template that only swaps the name of the city is a classic doorway pattern and scaled content abuse. One strong page per real location, with real details about that place, is fine.
Where AI Mode shines, and where you can too
Google describes AI Mode as built for questions that need further exploration, reasoning or complex comparisons, the nuanced questions that used to take several searches. Those are exactly the questions where an expert's comparison is worth the most. So write the comparison you'd give a client across the table: the options, the trade-offs, what you'd choose and why.
One question to check your page
When a page is finished, I read it as the person who asked the first question and ask myself: would they need to search again? Google lists "leaves readers feeling they need to search again" among the warning signs of search engine-first content. If the answer is yes, I find the missing follow-up question and answer it on the page.
When a follow-up really is a different topic, I link to the page that covers it with a normal, crawlable link. That way both the reader and Google's crawler can follow the same journey the person would follow in their head.
If you want help mapping the real questions behind your service and turning them into content that AI search can use, let's build it together.
Sources
- Optimizing your website for generative AI features on Google Search · Google Search Central, updated July 10, 2026
- Spam policies for Google web search · Google Search Central, updated August 28, 2026
- Generative AI performance report (Search) · Search Console Help
How I made this article: I researched and drafted it with the help of Claude, then checked every fact myself against the official documentation listed in the sources.