What query fan-out is
Google described the technique when it announced AI Mode at Google I/O on 20 May 2025: "AI Mode uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." Its documentation for site owners says both AI Overviews and AI Mode may use query fan-out, "issuing multiple related searches across subtopics and data sources" to develop a response.
In plain terms: one question from a user becomes many searches behind the scenes. The answer is then built from the pages that best answer each piece, and Google says this lets it show a wider set of supporting links than a classic search.
SEO and AI search · Dubai
An illustration of how one question might fan out
| Part of the question | A sub-search it could trigger | What a page needs to answer it |
|---|---|---|
| The service | physiotherapy for running knee pain | A clear explanation of the treatment |
| The place | physiotherapist JLT Dubai | Location, area served, how to get there |
| The cost | physiotherapy session price Dubai | A real price or range, stated plainly |
| The worry | how many sessions for runner's knee | An honest answer from experience |
| The practical bit | physiotherapist Dubai insurance accepted | Which insurers, stated in text |
That table is my illustration of the idea, not a list Google published. The point is that each sub-search is narrow, and a page that answers one of them precisely has a better chance of being used than a page that vaguely covers everything.
What information gain means
Information gain is the idea that a page is more useful if it adds something the reader has not already seen. Google holds a patent on it: US 11,354,342, "Contextual estimation of link information gain", filed in 2018 and granted on 7 June 2022, with inventors Victor Carbune and Pedro Gonnet Anders. It describes giving documents an information gain score based on how much new information they add beyond documents the user has already viewed, and using that score to choose or rank what to show next, including through an automated assistant.
Be careful with how this gets quoted. A patent shows what Google has explored, not what it uses in ranking today. Google has not confirmed an "information gain score" as a ranking factor, and anyone telling you it has is overclaiming. What Google has said publicly points the same way, though. Its helpful content guidance asks: "Does the content provide original information, reporting, research, or analysis?" and whether it offers "interesting information that is beyond the obvious". Its May 2025 advice on AI search says to focus on "unique, non-commodity content".
Why crowded SERPs are not the game
"If there's a topic where Google already has pages ranking and you're in that crowded spot, regardless of search volume, that's not the real game. You have to build your own content." Lohith Lalesh
Here is the reasoning. If ten pages already say the same thing, an eleventh page saying it slightly better adds almost nothing. A fan-out answer only needs one source for each fact, and when the fact is common knowledge, there is little reason to credit you for it. Search volume does not change that. A big keyword where you add nothing new is still a keyword where you add nothing new.
The opening is in the questions where the existing pages are generic and you have something specific: a local price, a real test, a pattern in your own enquiries, a clear opinion from someone who does the work. I still do keyword research, but I use it to find the questions, then ask what I can add that the current results do not have.
Three kinds of original material
When I plan content now, I prioritise three things above everything else, because they are hard to copy and easy for an AI answer to cite.
SEO and AI search · Dubai
Original material a small business can actually produce
| Type | What it looks like | A cheap Dubai version |
|---|---|---|
| Original survey data | Answers from real people, with the question, sample size and date | A two-question poll to your WhatsApp broadcast list or Instagram Stories |
| Real test results | Something you tried, measured and reported, including what did not work | Before and after photos with dates, or timing a job with two different methods |
| Expert quotes | A named person with relevant experience saying something specific | You, your senior technician or a partner professional, quoted by name |
How to produce original data cheaply
You do not need a research budget. Most small businesses already sit on useful data and never publish it. The rule is that every number you publish must be real, with its source, sample and date stated.
SEO and AI search · Dubai
Five low-cost ways to create original material
Poll your customers
Ask one or two clear questions through WhatsApp, email or Stories. Publish the question wording, how many people answered and when. Small samples are fine if you say how small.
Mine your own enquiries
Count what people ask about most, which areas they come from, which months are busiest. Your CRM, inbox or WhatsApp labels are a dataset.
Run before and after tests
Change one thing (a photo, a page, a process), note the date, and record what happened. Report it honestly, including when nothing changed.
Quote the person who does the work
Ten minutes of recorded conversation with your most experienced team member usually produces three or four quotable opinions.
Document your method
Write down how you do the job, step by step, with real photos. A method is original by definition.
Do not fake it
Invented statistics and made-up case studies are worse than having none. They break trust with readers, and an AI answer that repeats them puts your name next to something false. If you cannot measure it, describe your method instead.
Structuring pages so sub-queries find the answer
Original material only helps if it is easy to find on the page. Fan-out searches are narrow, so structure each page so that a narrow question lands on a clear answer.
- Use headings that match the sub-questions people ask: price, time, area, process, who it is for.
- Answer in the first sentence under each heading, then explain. The answer should make sense if quoted alone.
- Put data in tables or short lists with units, dates and sample sizes in text, not only in images.
- Name the source of every number, including "our own enquiries, January to June".
- Give each distinct topic its own page rather than one long page covering everything.
- Show who wrote it and why they know, with a real byline and short bio.
The technical side (crawler access, consistent business details, structured data) is in my guide to showing up in AI answers, and the case for human-led writing is in why companies are bringing human writers back. For a starting point on SEO overall, see SEO for small businesses in Dubai.
If you want help turning your own enquiry data and know-how into pages, that is part of how I approach SEO for Dubai businesses.