You can rank first for a keyword and still be missing from the AI answer sitting above your listing.
seoClarity analyzed 12,011 AI Mode citations across 1,000 transactional queries in September 2025. Only 19% of them came from pages ranking in the top 20 organic results for those queries.
So where did the rest come from? Largely from pages that answered questions the searcher never typed. That process is called query fan-out, and it's the clearest reason topical authority now matters more than any single ranking.
What query fan-out actually is
Google describes it plainly. Its Search Central documentation on AI features says AI Overviews and AI Mode may use a "query fan-out" technique, running multiple related searches across subtopics and data sources to build a response. When Google expanded AI Mode in May 2025, it described the process as breaking a question into subtopics and issuing many queries at once. Its Deep Search feature, according to Google, can issue hundreds.
Here's how that plays out for a small business. Someone asks, "Is a heat pump worth it in a cold climate?" The AI doesn't only look for the best page on that phrase. It may also search for:
- How heat pumps perform below freezing
- Whether cold-climate heat pumps need backup heat
- How long heat pumps last
- How loud the outdoor unit is
- Heat pump vs. gas furnace for winter heating
Then it stitches passages from the pages that answered those pieces into one response.
What it means
An HVAC company with one strong post on heat pumps in cold climates is competing for one citation slot. A company that has clearly answered the backup heat question, the lifespan question and the noise question has a shot at several.
You can't see Google's fan-out list, so you can't target it keyword by keyword. What you can do is cover a topic so thoroughly that your site becomes the natural source for whichever sub-question gets asked.
That's topical authority. Fan-out is simply how AI search tests for it, one sub-question at a time.
Why ranking for the main keyword isn't enough anymore
Surfer studied 10,000 keywords and 173,902 ranking URLs. The team used Gemini to generate roughly 33,000 fan-out queries, then checked which pages ranked for them and which got cited in AI Overviews.
Pages that ranked for the main query plus at least one fan-out query were 161% more likely to be cited than pages ranking for the main query alone. The more fan-out queries a page ranked for, the more likely it was to be cited, with a Spearman correlation of 0.77.
Two other findings are worth sitting with. Pages that ranked only for fan-out queries, and not the main keyword at all, were cited 49% more often than pages ranking only for the main keyword. And 67.82% of cited pages didn't rank in the top 10 for the main query or any of its fan-outs.
Surfer is careful to call this correlation, not proof of cause. Still, the direction is hard to argue with. Covering the surrounding questions beats owning one ranking.
Topical authority, measured: A Wellows study of 9,471 questions across five AI engines (January to May 2026) found that how widely a site was already cited across a topic predicted its future citations better than a domain authority score on four of the five engines. For Google AI Overviews, the correlation was 0.29 for topic coverage versus 0.22 for the authority score.
What to do
Stop judging a topic by whether you rank for its head term. Judge it by how many of its likely sub-questions your site answers well. A topic where you sit at position four but clearly answer eight follow-up questions is in better shape than one where you hold position one and answer none of them.
How to predict the sub-questions behind your topics
Google doesn't publish its fan-out queries, so you have to reconstruct them. Four sources get you most of the way there.
People Also Ask. Search your main query and open a few of the questions. Each one you open loads more. This is Google showing you its own guesses about follow-up intent.
Long queries in Search Console. Question-style queries of six words or more show the sub-questions people already find you for. Often they land on a page that only half answers them.
Your customers. The questions people ask on calls, in emails and at the counter are fan-out queries in disguise. A customer asking "will it still heat my house at minus 20?" is asking the same sub-question an AI would run.
AI assistants themselves. Ask ChatGPT or Gemini: "If you were answering [main question], what related questions would you research first?" The list won't match Google's exactly, but it reliably surfaces the obvious branches.
Quick win: For your three most important topics, list 10 to 15 sub-questions each using these sources. Mark every one as answered well, answered badly, or not answered. That sheet is your fan-out map.
One page or several?
Not every sub-question deserves its own URL. Split too eagerly and you get thin pages competing for the same intent.
Use one rule. If a sub-question shares the main query's intent, answer it as a section on the main page. If people would search it on its own, with a different goal in mind, it gets a separate page.
In the heat pump example, "do cold-climate heat pumps need backup heat?" belongs on the main page. "Heat pump vs. gas furnace" is a comparison with its own audience and its own decision, so it earns a page of its own.
What to do
Build each core topic as a cluster: one pillar page that answers the main question and its closest sub-questions, plus supporting pages for the branches with distinct intent. Link them in both directions. For grouping the questions, start with keyword clustering.
Before you publish a new supporting page, check that an existing page doesn't already target the same intent. Otherwise you're trading a fan-out gap for a keyword cannibalization problem.
Write sections an AI can lift on their own
Fan-out retrieval pulls passages, not whole pages. A section has to make sense when it's lifted out of the article around it. (We covered the mechanics in how LLMs parse content.)
- Lead with the answer. Put the direct answer in the first one or two sentences of each section. Context and nuance come after.
- Write headings the way people ask. "Do heat pumps work below freezing?" gives an AI far more to match than "Performance considerations."
- Use tables for comparisons. Specs, side-by-side options and pros and cons are easy for AI systems to extract accurately.
- Cut cross-references. "As mentioned above" makes a passage useless out of context.
- Add what only you know. Your install numbers, your local climate, the failure you see every winter.
That last one carries the most weight. If your answer reads like the 20 other answers on page one, the AI has no reason to pick yours. It's also the most common reason long guides get passed over in ChatGPT.
Google's documentation says no special optimizations or extra structured data are required for AI Overviews or AI Mode. Fan-out simply applies the usual quality bar to more questions at once.
Audit your fan-out coverage every quarter
Rankings alone won't tell you whether this is working. Run a short check instead.
- Pick your 10 most important queries.
- Run each one through AI Mode and ChatGPT. Note the sub-topics each answer covers and which sites get cited.
- Compare those sub-topics against your fan-out map. Anything the AI covers that you don't is a gap.
- Fill the biggest gaps first, either as new sections on the pillar page or as new supporting pages.
Expect answers to shift between runs, which is why a quarterly pattern tells you more than any single check. For more on tracking this over time, see our guide to AI search visibility.
Start with the pages you already have
Before you write a single new section, make sure your existing pillar pages can be crawled, indexed and read cleanly. A missing canonical, a slow page or a broken internal link can keep your best answer out of the fan-out results no matter how good it is.
Run a site audit at seositecheckup.com to find the technical issues standing between your content and the AI answers your customers are reading.
