Query fan-out: there is a default brand inside AI search, and it may not be you
Query fan-out is how AI search really works. Instead of answering your question directly, the model expands it into a web of sub-queries, researches each one, then combines the results into a single answer. We studied 289,850 total prompts and found something surprising happening inside that step.
The prompts were the kind of neutral questions customers ask every day: which option is best for a given need, how something works, what to look for in a category. No brand in any of them. Just the problem the customer wants solved.
When someone asks about your category with no brand in it, the model quietly names one anyway. We watched this play out across brokerage, insurance, travel, and luxury fashion, and one finding stood out above everything else.
You ask something ordinary, no brand names, just intent.
The model splits your prompt into a tree of research sub-queries you never see.
Some sub-queries name specific, real-world brands. That is your shot at being the answer.
One prompt did this. Across the study we captured 668,020 fan-out queries in total.
The brand-free question is a myth
That is the shift in one sentence. Search used to hand people a page of links and let them choose. AI hands them an answer, and the shortlist gets written before the customer has a single brand in mind.
There is no blank slate anymore. The question that feels neutral to your customer is already half answered before they finish typing.
| Industry ↕ | Total prompts asked ↕ | Fan-out queries ↕ | Branded fan-out ↕ | % branded ↓ |
|---|---|---|---|---|
| Travel / airlines | 10,750 | 19,710 | 3,404 | 17.3% |
| Brokerage / trading | 250,800 | 567,937 | 39,637 | 7.0% |
| Insurance | 12,700 | 30,843 | 1,744 | 5.7% |
| Luxury fashion | 6,600 | 24,247 | 1,261 | 5.2% |
| GEO / AI-search tools | 9,000 | 25,283 | 703 | 2.8% |
| Total | 289,850 | 668,020 | 46,749 | 7.0% |
Every category already has a favorite
To keep the results clean, we ran every prompt in a fresh temporary chat with no account, no history, and no personalization. So what we saw is not one person's tailored answer. It is the model's baseline instinct for the category, with nothing nudging it.
Look at which brands the model reaches for, and a second pattern appears. It does not spread attention evenly. It leans on a few names.
On brokerage prompts that named anyone, one broker was the brand the model put forward 56% of the time. The top three names covered close to 90% between them. The rest of the field split what was left.
The model is not weighing the whole field on every question. It has learned a default, and it returns to that default again and again. In brokerage, once it names a brand, it circles back to that same small set of names dozens of times across its internal research before it writes a word.
If your brand is the default, this is the best distribution you have ever had. If it is not, you are not losing the deal at the comparison stage. You are losing it before the customer knows a comparison is happening.
Your how-to content is brand territory too
Here is the part that surprises most teams. The most brand-loaded questions are not the shopping questions. They are the ones where people are still learning.
Take a simple how-to, like how a loyalty program works. In our data, that kind of question pulled a single brand into the model's reasoning more than 8 times out of 10, higher than any "best of" comparison. The reason is obvious once you see it: the model cannot explain how the program works without naming whose program it is. Educational questions are not neutral ground. They are where the brand gets decided first. We go deeper on this in why your educational content is brand territory.
Teams spend years building helpful, top-of-funnel content and assume it sits above the brand fight. It does not. The moment a topic has a clear real-world owner, the model folds that owner into the explanation.
The fan-out layer is not hidden. Almost no one is measuring it.
The fan-out layer is not hidden. Anyone can open a model and watch it think. The problem is that almost no one is paying attention to it, and even fewer are measuring it at scale. That is the point.
The brands winning AI search are not guessing. They can see which questions pull a competitor into the model's thinking and leave them out, they know which of their pages the model actually reads, and they can measure what changes when they close the gap. The pattern also shifts by country, which we cover in AI search visibility is market by market.
The default is already set for your category. The work is finding out whether it is you, and doing something about it if it is not.
FAQs
What is query fan-out?
Query fan-out is the technique where an AI model expands a single question into several smaller sub-queries, researches each one, then combines the results into one answer. Our research found that models often name specific brands inside those sub-queries, even when the original question named none.
Why does AI name brands I did not ask about?
Because the model has learned a default set of brands for each category. When it fans out a neutral question, it reaches for those defaults to build its answer, even when the original prompt named no one. In our data this happened in up to 71% of brand-free prompts.
How do I know if my brand shows up in AI search?
You need visibility into the fan-out layer: which questions in your category pull which brands into the model’s reasoning, and whether yours is one of them. A tool like Limy measures this directly and shows where you are missing.
Is AI search visibility the same as SEO?
No. Traditional SEO optimizes for ranked links on a results page. AI search visibility, sometimes called answer engine optimization, is about being named inside a single AI-generated answer, which is decided during query fan-out.
Most Viewed Articles
