Query Fan-Out: Why One Page Per Keyword Stopped Working

Most content plans still work the way they did in 2016. Pick a keyword, write a page targeting it, repeat. That approach is inefficient now, and the reason is a mechanism Google has been open about: query fan-out.

Understanding it changes what a content plan should look like more than any other single thing in search right now.

What query fan-out is

When someone asks Google’s AI features a question, the system does not just retrieve results for that question. It generates a set of related queries behind the scenes, retrieves results for each of them, and synthesises an answer across all of it.

Google’s own documentation uses a question about fixing a weed-filled lawn, and names the fan-out queries it might generate: best herbicides for lawns, removing weeds without chemicals, and preventing weeds in a lawn. The user asked one question. The system ran three searches behind it.

The consequence is that being the best page for the original query is no longer sufficient, and being a decent page for several of the fan-out queries can be more valuable than being the best page for one.

Why single-keyword pages lose under this model

You are only eligible for one of the retrievals. If a question fans out into several subqueries and your site can answer one, you enter the synthesis once. A competitor covering four of them enters four times, and is far more likely to be the source the answer leans on.

Semantic matching killed the long-tail variant strategy. The old approach of building near-identical pages for slight phrasing differences assumed the engine was matching strings. It is not. It understands synonyms and equivalence, so those pages now compete with each other rather than covering distinct ground.

Thin coverage reads as thin. A page that answers one narrow question and nothing around it looks, to a retrieval system assessing depth, like a page that does not know very much.

How to plan a cluster instead

The process is not complicated. It just requires doing something before writing that most people skip.

1. Start with the parent topic, not the keyword. Write down what the reader is trying to accomplish. Not the phrase you want to rank for. The job they are trying to get done.
2. List the fan-out questions. Sit down and write the five to ten questions a system would plausibly generate from that parent query. What comes before this question, what comes after it, what are the objections, what are the alternatives, what does it cost, what goes wrong. Ten minutes with a blank page gets you most of the way.
3. Decide what is one page and what is several. This is the judgement call. Questions that a reader would want answered in the same sitting belong in the same piece. Questions that represent a different moment in the journey belong in separate pieces that link to each other.
4. Link them deliberately. Internal links are how a retrieval system understands that these pages are one body of work rather than scattered posts. Link from the specific pieces to the parent and from the parent out to each specific piece.

Getting this wrong in either direction is costly. Split too finely and you have thin pages competing with each other. Combine too aggressively and you have one enormous page that answers everything shallowly.

One email platform, 4 pieces

Say the parent topic is choosing an email marketing platform for a small business.

The lazy version is one post titled Best Email Marketing Platforms, targeting that phrase.

The fan-out version starts by listing what the system would also want. What do these platforms cost? How do they differ? Does a small business need one at all? What happens to your list if you switch, what do the free tiers restrict, how do you migrate, and where does deliverability go wrong.

That yields a cluster. A comparison piece covering the main options. A pricing breakdown with real numbers. A piece on migration, which is the question people ask second and almost nobody writes about. A piece on deliverability, which is where the expensive mistakes happen.

Four pieces, each answering a distinct question properly, linked to each other. That structure gets retrieved across multiple fan-out variants. The single post does not.

Where the cluster strategy stops

Mechanical clusters read as scaled content abuse

The failure mode is producing the cluster mechanically, which is a real risk given how tidy the framework looks.

If you take a topic and generate fifteen near-identical pages to blanket the fan-out space, you are doing the thing Google’s spam policies describe as scaled content abuse. The guidance on this is explicit: generating at volume for ranking purposes, rather than because each page has something to say, is the problem regardless of how it was produced.

The test I use is whether each page in a cluster could stand alone and be worth reading by someone who arrived at it directly, with no interest in the rest. If the answer is no, it should not be its own page.

Getting retrieved is not getting chosen

Clusters help you get retrieved. They do not help you get chosen, which is most of why posts go uncited in AI search.

Once a system has retrieved six candidate sources for a subquery, it selects among them on other grounds: specificity, whether there is anything concrete to cite, whether the page has an author and a date, whether the claims are verifiable.

Which means the cluster strategy only pays off if each piece in it contains something particular. Numbers, dates, named examples, first-hand detail. A cluster of six generic pages is six chances to not be picked.

More on how AI answers pick their sources is in AI Search and Getting Cited.

The strategy pays off for people with first-hand experience as the content moat, and much less for people assembling content from other content. The structural advantage of the cluster only converts if there is something inside it worth quoting.

Start with your 3 best pages

Take your three best-performing pages. For each, write the fan-out list. Then check which of those questions your site already answers.

The gaps you find are the highest-value content you could produce right now, because they sit next to something that already works and they complete a cluster that is currently incomplete. That is a better prioritisation method than a keyword tool, and it costs half an hour. It sits ahead of SEO tests that cannot fail because the answer can come back empty.