Query fan-out: why one question becomes several

Query fan-out is what happens when one question a person types becomes several searches the system runs on their behalf. Ask which supplier handles case packing for glass jars at low volume, and the system may search separately for case packers, for glass container handling and for low-volume line formats, then assemble a single answer from all three sets of results.

For a supplier this changes what you are competing in. Your name appears in the final answer only if your material surfaced in enough of those hidden searches to be worth including.

A company that has written only about its broad category is absent from most of them, and therefore absent from an answer to a question it would have said it was well qualified to answer.

Source: Optimizing your website for generative AI features on Google Search, Google Search Central.

Why systems do this at all

Because the question as asked is usually too broad to answer from one search.

"Who should I consider for end-of-line packaging" has no single correct set of results. Broken into narrower searches — the format, the product, the rate, the constraint — each one returns something specific, and a useful answer can be built from the combination. It is the same thing a good salesperson does when a buyer asks a vague question: ask four narrower ones, then answer.

The buyer asks one question. The system asks five.

What this means you are actually competing in

Not one contest, but several at once, with different competitors in each.

In the search about case packers you are up against every case packer manufacturer. In the search about glass handling you are up against a different and smaller set. In the search about low-volume lines, smaller again. A company that is mediocre in the first and unmatched in the third can end up named, because the answer is assembled from all three and somebody has to supply the specific part.

This is the single best piece of news in this subject for a mid-sized manufacturer. Scale wins the broad search. Specificity wins the narrow ones, and the narrow ones are where the buyer's actual constraint lives.

Worked through, with one real question

Take a buyer asking which supplier handles case packing for glass jars on a line running 40 units a minute.

Behind it sit searches along the lines of: case packing equipment suppliers; handling glass containers without breakage; case packers suitable for low-speed lines; and often something about the buyer's sector, because glass at that rate usually means food, drink or pharmaceutical.

Now ask what you have published. If the answer is a product page headed "Case packers" and a company page headed "Packaging solutions", you have material for the first search and nothing for the other three. The competitor with a page on glass handling at low speeds is in three of the four.

Why a keyword list is the wrong map

Because a keyword list records what people type, and fan-out is about what the system asks next.

The set of questions that sit behind and around a buyer's question is what we call a query universe, and building one is part of entity mapping — both defined on Map. It is a different exercise from keyword research and produces a different list: longer, more specific, and full of questions with no measurable search volume at all, because nobody types them. The system does.

Search volume is the wrong filter here. A question asked by four buyers a year, each spending a million, is worth more than one asked by four thousand people who are not buying anything.

What to write instead

Pages that answer the narrow questions completely, in the buyer's own terms.

Not "Packaging solutions" but "Case packing glass jars on low-speed lines" — with the format, the rate, the breakage constraint, the changeover time and what it costs to install on an existing line. One question, one page, answered properly.

That is less glamorous than a content calendar and it is the work that actually lands. It also holds up if the mechanism changes, because a page that genuinely answers a specific question has been valuable in every version of search there has ever been.

Can you see the fan-out?

Not directly, and be sceptical of anyone who says otherwise.

The systems do not publish the searches they run behind a question. What you can do is infer the shape of it: ask the same buyer question several times, note which competitors appear in each answer, and watch which aspects of the question the answers keep addressing. If three different answers all discuss breakage, breakage is one of the hidden searches.

That is inference rather than measurement, and we say so on the page rather than presenting it as a readout of something we cannot see.

What this article cannot tell you

It cannot tell you how many searches any question becomes, or what they are. That varies by system, by question and by day, and none of it is published.

It cannot tell you which narrow questions matter most in your category. That is exactly the work of building a query universe, and it is done by looking at your market rather than by reasoning from first principles.

And AI systems are probabilistic and change continuously. The mechanism described here is the shape of the process as its owners describe it, not a rule that will hold unchanged.

What to do next

Write down the five questions your buyers ask most often before they know who to call, then write down the three or four narrower questions hiding inside each one. That list is the beginning of a query universe, and most companies find they have published material for perhaps a fifth of it.

Then check what the systems currently return for the original five. What appearing in those answers actually consists of is set out in what AI visibility actually means, and the discipline built around it in what is generative engine optimisation. If you would rather have the whole set measured properly — agreed questions, all six systems, competitors alongside you — that is the AI Visibility Audit.