Query Fan-Out: How AI Decides Which Law Firm to Cite
By Brittany Winters, Director of Client Relations
AI search does not answer your question with one search. It breaks the prompt into many sub-questions, runs them at once, and assembles an answer from whatever each one returns. Google calls this query fan-out. It explains why a firm that ranks nowhere for "car accident lawyer" can still get named in an AI answer, and why the unit that gets cited is a passage, not a page. Write to the sub-questions and you become citable without winning the head term.
The most useful thing to understand about AI search is that the question you type is not the question that gets searched. Google's AI Mode takes your prompt, breaks it into subtopics, and issues a multitude of queries simultaneously on your behalf, a technique Google calls query fan-out (Google). The answer you read is assembled from what those hidden searches returned.
Once that clicks, a lot of confusing behavior in AI search stops being confusing.
What fan-out looks like in practice
Say someone types: *"do I need a lawyer after a minor car accident where nobody was hurt?"*
That is one question to a human. To the system it is a cluster. It fans out into something like:
- what counts as a minor car accident
- when injuries show up late after a low-speed collision
- what happens if you accept an insurance settlement early
- do you need a lawyer for a property-damage-only claim
- how contingency fees work when a case is small
- statute of limitations for car accident claims
Each of those runs as its own search against its own set of sources. The model then pulls the most useful chunk from each, merges them, and cites the pages that contributed.
Nobody had to rank for the original sentence. Six different pages, possibly from six different firms, can each supply one piece.
Why this is good news for a firm that does not rank yet
Here is the part that matters strategically.
Winning the head term "car accident lawyer" is a years-long authority fight against firms with far bigger link profiles. Being the best available answer to "when do injuries show up after a low-speed collision" is a writing problem. The competition on that sub-question is thinner, the intent is clearer, and the page that answers it cleanly can get pulled into an answer for the much bigger question sitting above it.
That is a real opening, and it is the practical reason a blog can earn AI citations while the commercial pages are still climbing. It does not mean authority stopped mattering. Citation still weighs whether the source is credible. It means the entry point moved from one impossible term to many reachable ones.
What this changes about how you write
Four things follow directly from the mechanism:
- Answer one question per section, completely. If the retrievable unit is a passage, then every section should stand on its own when lifted out of the page. A section that only makes sense after reading the previous four cannot be cited.
- Use the question as the heading. Sub-queries are phrased as questions. Headings phrased the same way match more cleanly than clever ones.
- Front-load the answer. Give the direct answer in the first sentence or two, then explain. A hedge-first paragraph gives the model nothing quotable.
- Cover the cluster, not the keyword. For any case type you care about, the fan-out will include cost, timeline, what-if, do-I-even-have-a-case, and what-happens-next. Pages that cover the whole neighborhood of questions get pulled in more often than pages that repeat one keyword.
This is why case-type landing pages outperform a single services page, and why FAQ blocks earn their space. They are shaped like the thing being searched.
What it does not change
Three honest caveats, because this topic attracts overclaiming:
- It is not a ranking hack. You cannot enumerate the sub-queries and stuff them. The expansion is generated per prompt and varies.
- Authority still gates it. Being retrievable is necessary and not sufficient. The model still prefers sources it has reason to trust, which is why off-site credibility keeps mattering.
- Citations are not clicks. Getting named in an answer is brand exposure and sometimes a referral. It is not the same as a session, and measuring it as if it were will mislead you.
The short version
Stop writing for the question you wish people asked. Write for the ten smaller questions the machine asks on their behalf, answer each one cleanly enough to be quoted out of context, and let the head term follow.
For the broader picture on earning citations, see how to get cited by AI Overviews and ChatGPT. For the retrieval side, submitting your site to Brave covers the index that a lot of assistants are actually reading.
Frequently asked questions
What is query fan-out?
It is the technique AI search uses to break one prompt into many sub-questions, run them as simultaneous searches across subtopics and data sources, and assemble a single answer from the results. Google has described it as core to how AI Mode works.
Can my firm get cited by AI without ranking on page one?
Yes, and that is the practical significance of fan-out. Citation happens at the level of a passage answering a sub-question, not the page ranking for the original prompt. A page that answers one narrow question well can be pulled into an answer for a much broader question.
Can I optimize for the specific sub-queries directly?
Not reliably. The expansion is generated per prompt and varies between runs and platforms. The durable approach is coverage: answer the full cluster of questions around a case type, one per section, each written to stand on its own.
Does getting cited by AI bring traffic?
Sometimes, and less than ranking would. A citation is brand exposure and a possible referral click, not a guaranteed session. Treat it as a visibility channel worth earning, and keep measuring signed cases rather than citation counts.
Want this run for your firm?
See exactly where your retainers are leaking, then decide. One firm per metro.