Discussion
  • AI Search

If query fan-out means AI Mode cites pages ranking for sub-queries, should content plans target fan-out queries directly, or is that long-tail SEO renamed?

Published 1 reply

Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    Mostly, it is long-tail and topic-coverage SEO under a new name, and I would not build a content plan around a list of fan-out queries. The underlying skill is the same one good content strategists already use: break a broad task into the specific questions a person needs answered, then make sure each question has a strong answer somewhere on the site.

    There are two real differences, though, and they change how you plan rather than whether you plan. Fan-out queries are written by the system, not typed by users, so they have no search volume, they vary between sessions, and Google does not report them. And the outcome you are competing for is being retrieved as a supporting source for part of a synthesised answer, which is not the same as ranking a page for a keyword. Use fan-out as a lens for checking coverage and structure. Keep real demand as the basis for deciding what gets its own page.

    What Google has actually documented about fan-out

    Google’s documentation says both AI Overviews and AI Mode “may use a ‘query fan-out’ technique”, issuing “multiple related searches across subtopics and data sources” to develop a response. It also says the models identify more supporting pages while generating the answer, producing “a wider and more diverse set of helpful links” than a classic web search.

    Three details in that page matter for planning. There are “no additional requirements” to appear in these features. A page must be indexed and eligible to show a snippet to be used as a supporting link. And the fan-out reaches “data sources”, not just web pages, which on commercial topics can include structured sources such as product feeds. Google’s May 2026 guide to generative AI search goes further and says you do not need artificial chunking, special markup or AI-specific rewording, and that optimising for generative AI search is still SEO.

    The third-party evidence is consistent with fan-out but does not prove how it weights pages. Ahrefs reported that the share of AI Overview cited pages ranking in the top 10 for the main query fell from 76% in July 2025 to 38% in March 2026, while noting its detection method changed, and BrightEdge reported only 17% overlap using a different method. Pages cited without ranking for the head query fit the idea that they were retrieved for a sub-query. They also fit other explanations, such as passage-level retrieval or personalisation, so treat this as supporting context rather than a measured mechanism.

    Where fan-out and long-tail really are the same thing

    Both reward sites that answer specific intents well rather than sites that repeat a head term. A page that already ranks for “payroll software with contractor payments” in classic search is plausibly well placed to be retrieved when a model generates something close to that query. Topical depth, clear page purpose, internal links between related answers and genuine expertise are doing the work in both cases.

    A useful way to picture fan-out: a customer asks a research assistant one big question, and the assistant phones five specialists with narrower questions before writing the answer. You want to be the specialist who gets called. The catch is that you never hear the exact questions, and the assistant phrases them a little differently each time. Technically, the model decomposes the prompt into sub-queries, runs retrieval for each against Google’s index and other sources, then selects passages to support parts of the generated response.

    Where they differ enough to change the plan

    Dimension Long-tail keyword Fan-out sub-query
    Who writes it A user The model, from the user’s prompt and context
    Volume data Available from Search Console and keyword tools None; the query may never have been typed by anyone
    Stability Fairly stable phrasing over months Varies by prompt wording, conversation history, model version and personal context
    Reporting Query-level data in Search Console Not exposed; Search Console’s generative AI reports show impressions by page, with no queries or clicks
    Unit that competes A page ranking for the query A passage or page supporting one part of a synthesised answer
    Success looks like A ranking and a click A citation or mention, often with no click

    The reporting row is the practical constraint. Search Console’s generative AI performance reports, which reached all eligible sites on 31 August 2026, give impressions for AI Overviews, AI Mode and AI features in Discover, filterable by page, country and device, but no queries. You can see which pages appear. You cannot see which sub-question pulled them in.

    Why targeting fan-out queries directly tends to go wrong

    • You would be targeting guesses. Fan-out simulators prompt a language model to produce plausible sub-queries. That is a reasonable brainstorming input, but public evidence that simulated lists match what AI Mode actually issues for a given user is thin.
    • It pushes towards page-per-query production. A plan with 400 synthetic sub-queries tempts teams into 400 thin pages. Google’s scaled content abuse policy targets large volumes of low-value pages made mainly to capture search, whether written by people or AI.
    • Many sub-queries share one best answer. “Payroll software pricing per employee” and “how much does payroll software cost for a small business” usually deserve one strong section, not two competing pages.
    • Some sub-queries are answered by data, not prose. Where Google fans out to structured sources, the fix may be a complete Merchant Center feed or Business Profile rather than another article.

    Watch for this:

    Rewriting pages into disconnected question-and-answer fragments “for fan-out” is the artificial chunking Google’s AI guide says is unnecessary. Clear headings and self-contained answers help readers and retrieval alike. Fragmenting a coherent explanation helps neither.

    A planning framework that uses fan-out without chasing it

    1. Start from real demand. Search Console queries, query groups in Search Console Insights, People Also Ask, sales and support questions. This decides what deserves a page.
    2. Decompose each priority topic into sub-questions. Ask what a competent expert would need to answer to handle the full task: criteria, constraints, costs, steps, alternatives and risks. An LLM simulation is one input here, alongside your own subject knowledge.
    3. Map each sub-question to an existing URL, a section or a gap. Give a sub-question its own page only when it has independent demand and a distinct intent. Otherwise, add a clearly headed section to the parent page.
    4. Check classic rankings for the explicit versions of the sub-questions. Google says fan-out issues searches and that supporting pages must be indexed and snippet-eligible. It is a reasonable inference, not a documented weighting, that pages which rank for the typed version of a sub-query are better placed when the model generates something similar.
    5. Measure at page level. Track AI impressions by page in Search Console, classic clicks and rankings for the sub-question set, and repeated manual prompts for your priority topics.

    Hypothetical example:

    An Australian payroll software company wants visibility for prompts like “best payroll software for a 30-person business with contractors”. Decomposing that task produces sub-questions about Single Touch Payroll compliance, paying contractors, award interpretation, pricing per employee and Xero integration.

    The audit shows pricing, STP and Xero already have strong pages. Contractor payments is a paragraph buried in an FAQ, so it becomes a proper section on the pay-runs page. Award interpretation has real search demand in Search Console and no page at all, so it gets one. The team does not create “payroll software for 30-person businesses” or “for 50-person businesses” pages, because those are prompt variations, not distinct intents.

    Bing is currently more open than Google here. Bing Webmaster Tools’ AI Performance report (public preview since February 2026) includes a sample of grounding queries alongside page-level citations for Copilot and Bing AI summaries. Those samples are the closest thing most sites have to seeing real sub-queries, and they are a useful check on whether your simulated lists resemble anything a live system issued. They come from Microsoft’s pipeline, though, not Google’s.

    How to test whether gap-filling actually changes AI visibility

    Pick 20 comparable topic clusters. In 10, fill the sub-question gaps you identified, using sections before new pages. Leave the other 10 unchanged. Over 8 to 12 weeks, compare AI impressions by page in Search Console, classic impressions and clicks for the sub-question queries, and citation rates from a fixed set of prompts run several times each (AI answers vary between runs, so single checks are unreliable). Watch for core updates and seasonal demand in the window.

    If the treated clusters gain AI impressions and classic rankings for sub-questions together, that supports treating fan-out as coverage work. If they gain classic rankings but no AI visibility, the retrieval link is weaker than the inference assumes, and you would look harder at page-level factors such as extractable answers and freshness. If Google ever exposes fan-out queries with any volume signal, they become one more keyword source and this whole debate gets simpler.

    My working position

    Plan from real demand, audit coverage using fan-out-style decomposition, and measure by page. Treat “fan-out keywords” as a hypothesis generator, not a publishing queue. The sites that do well here are the ones that were already answering the full task behind a query, with each sub-question handled once, clearly, in the place a reader would expect to find it.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.