Skip to content
Climb Google Maps: Google Business Profile optimization.

AI SearchGuide

Query Fan-Out: What Google Says It Is, What the Studies Disagree On, and a Script to Find Your Topic’s Sub-Questions

Query fan-out is how AI Mode, AI Overviews and ChatGPT break one question into several searches before they answer. Here is what Google has actually said about it, why two big studies seem to disagree, and a short script that lists the sub-questions for any topic, run on a real example in the US and the UK.

Aerial view of tidal channels branching out across a golden mudflat

Key takeaways

  • Query fan out is the technique Google says AI Mode and AI Overviews may use: "issuing multiple related searches across subtopics and data sources" to build one answer.
  • Google said on Nov. 12, 2026 that Gemini 4 Hydrogen "supercharges" query fan-out in Search, and its documentation still says there are no special optimizations or extra technical requirements for AI features.
  • Ahrefs found fan-out queries in 14.4% of 66.6 million free, logged-out ChatGPT responses (Aug. to Sept. 2026), while AirOps found that pages matching the main question closely beat pages that cover every fan-out subtopic.
  • Google does not publish its fan-out queries, so People also ask and related searches are the closest public stand-in; our script collects them for any topic and country for about $0.008 a run.
  • In our Nov. 13, 2026 run for "google business profile suspended", the script found 22 sub-questions in the US and 25 in the UK, and four of them showed up in both countries.

Short answer: query fan-out is what an AI search engine does when it splits your question into several smaller searches, runs them at the same time, and writes one answer from what comes back. Google says AI Mode and AI Overviews may work this way, and ChatGPT does it too. For a site owner, it means your page can be found through a sub-question you never targeted, not only through the main query.

Google does not show anyone its fan-out queries. Below is what Google has said in its own words, what two large studies found and why they don't actually contradict each other, and a short script that lists the likely sub-questions for any topic from Google's public results. We ran it on Nov. 13, 2026 for one real topic in the US and the UK.

What is query fan-out?

Google's AI features documentation is the plainest statement. It says 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."

Google first described it when AI Mode launched, in a May 20, 2025 post by Elizabeth Reid, Google's head of Search: AI Mode breaks a question "into subtopics," issuing "a multitude of queries simultaneously." Deep Search, the research mode inside AI Mode, uses "the same query fan-out technique," and Google says it "can issue hundreds of searches."

Take a question like "is it worth switching to a heat pump in a cold climate." A fan-out system might search separately for cold-climate heat pump models, running costs, installation prices and local rebates, then combine the results. Each of those searches is a separate chance for a page to be read and linked.

What changed this week?

On Nov. 12, 2026, Google moved Search to its new Gemini 4 Hydrogen model and said Hydrogen "supercharges our query fan-out technique in Search," with "simultaneous queries across all the relevant subtopics." Google says this connects people to "a broader range of helpful links than before." Google says the upgrade is global and applies in every language AI Mode supports. Our Gemini 4 Hydrogen post covers the rest of that announcement.

What did not change: Google's documentation still says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A page has to be indexed and eligible to show in Search with a snippet. There is no fan-out markup, setting or file.

Does ChatGPT use query fan-out?

Yes, but less often than many people assume, according to the largest study we have seen. Ahrefs looked at 66.6 million free, logged-out ChatGPT responses from August and September 2026 and found that "14.4% of the 66.6 million responses contained fan-out queries," with about 1.6 fan-out queries each when they did. 6.8% of those queries used the site: operator to search one website, usually a brand's own site.

Ahrefs states its own limits: one model, free sessions only (which search less than paid or reasoning modes) and broad prompts rather than "best X" shopping questions. It also notes that OpenAI changes how ChatGPT searches all the time.

Is covering every sub-question the way to get cited?

Not on its own, says the other big study. AirOps, working with Kevin Indig, ran 16,851 queries through ChatGPT three times each and scored 353,799 pages it retrieved. Their findings:

  • Retrieval rank mattered most. A page in first position in ChatGPT's search results was cited 58% of the time; at position 10, 14%.
  • Matching the main question beat breadth. Pages whose headings closely matched the user's query were cited 41% of the time, against 29% for weak matches.
  • Covering every sub-question did not help. With query match held constant, pages covering 26 to 50% of ChatGPT's fan-out sub-queries were cited more often (38.2%) than pages covering all of them (34.0%).

AirOps sells AI search software, the findings are correlations rather than proof, and the study covers ChatGPT only. Weigh it that way.

Why both studies can be true

They measure different things. Ahrefs counts how many of all ChatGPT responses fan out at all, including the many answers that never search the web. AirOps starts only from answers that did search and asks which retrieved pages got cited.

Put together: fan-out happens on a minority of ChatGPT answers, but when ChatGPT does search, the fan-out queries decide which pages it even sees. Once a page is retrieved, a clear, direct match to the question matters more than having a section for every possible angle. Google's own numbers on how often AI Mode fans out are not public, so nobody can say yet whether the same holds there.

Our view: write focused pages that answer one question clearly, then make sure each likely sub-question has a good answer somewhere on your site, often on its own page, linked from the main one. Don't stuff every angle into one "ultimate guide."

How to find the fan-out questions for your topic

Google doesn't publish the searches AI Mode runs. The closest public signal is what Google already shows people who search the topic: the People also ask box and related searches. They are not the fan-out queries themselves, and we don't present them as such, but they are Google's own record of the follow-up questions people ask, which is what fan-out is built to cover.

The script below collects them. It searches the topic once, then searches the top three related searches one level deeper (a small fan-out of its own), and lists every question it found, with where it saw each one. Questions that show up in several places come first.

It needs Node 18 or later and a DataForSEO account (pay as you go; each of our runs cost $0.0080 for four searches). Set DATAFORSEO_KEY to the Base64 of login:password, then run node fanout.mjs "your topic" 2826, where the number is the country: 2840 US, 2826 UK, 2356 India, 2036 Australia, 2124 Canada.

// fanout.mjs: list the sub-questions a topic is likely to fan out into,
// from Google's own "People also ask" boxes and related searches (DataForSEO).
// Usage: DATAFORSEO_KEY=<base64 of login:password> node fanout.mjs "<topic>" [location_code]
// location_code: 2840 United States (default), 2826 United Kingdom, 2356 India,
// 2036 Australia, 2124 Canada
const [topic, loc = '2840'] = process.argv.slice(2);
if (!topic) {
  console.error('Usage: node fanout.mjs "<topic>" [location_code]');
  process.exit(1);
}
const KEY = process.env.DATAFORSEO_KEY;
 
async function serp(keyword) {
  const res = await fetch('https://api.dataforseo.com/v3/serp/google/organic/live/advanced', {
    method: 'POST',
    headers: { ...(KEY ? { Authorization: `Basic ${KEY}` } : {}), 'Content-Type': 'application/json' },
    body: JSON.stringify([
      { keyword, location_code: Number(loc), language_code: 'en', device: 'desktop', depth: 10 },
    ]),
  });
  const json = await res.json();
  const task = json.tasks?.[0];
  if (task?.status_code !== 20000) throw new Error(`${keyword}: ${task?.status_message ?? res.status}`);
  return { items: task.result?.[0]?.items ?? [], cost: task.cost ?? 0 };
}
 
const found = new Map(); // normalized question -> { text, from }
function add(text, from) {
  const key = text.trim().toLowerCase().replace(/[?.]+$/, '');
  if (!key || key === topic.toLowerCase()) return;
  if (!found.has(key)) found.set(key, { text: text.trim(), from: new Set() });
  found.get(key).from.add(from);
}
function collect(items, from) {
  const related = [];
  for (const it of items) {
    if (it.type === 'people_also_ask') for (const q of it.items ?? []) add(q.title, `asked: ${from}`);
    if (it.type === 'related_searches') {
      for (const r of it.items ?? []) {
        add(r, `related: ${from}`);
        related.push(r);
      }
    }
  }
  return related;
}
 
let cost = 0;
// Round 1: the topic itself.
const first = await serp(topic);
cost += first.cost;
const related = collect(first.items, 'topic');
const aio = first.items.some((i) => i.type === 'ai_overview');
// Round 2: the top three related searches, one level deeper.
const deeper = related.slice(0, 3);
for (const r of deeper) {
  const next = await serp(r);
  cost += next.cost;
  collect(next.items, r);
}
 
console.log(`Likely fan-out questions for "${topic}" (location ${loc})`);
console.log(`AI Overview on the topic itself: ${aio ? 'yes' : 'no'}\n`);
const rows = [...found.values()].sort((a, b) => b.from.size - a.from.size);
for (const q of rows) console.log(`- ${q.text}  [${[...q.from].join('; ')}]`);
console.log(`\n${rows.length} questions from ${1 + deeper.length} searches. Cost: $${cost.toFixed(4)}`);

What we found when we ran it

We ran the script on Nov. 13, 2026 for "google business profile suspended", a topic we wrote a full guide on that morning, in the US (2840) and the UK (2826). Google showed an AI Overview for the topic in both countries.

  • US: 22 questions from 4 searches. The script went one level deeper on "Google Business Profile suspended due to policy violation", "Google business profile suspended how long" and "Google business profile suspended reddit".
  • UK: 25 questions from 4 searches. Google offered different related searches here, so the second round went into "Google business profile suspended reddit", "how to fix" and "how long".
  • In both countries, four questions appeared: "How do I get my Google account unsuspended?", "Why does Google keep saying I have suspicious activity?", "Why did my account suddenly get suspended?" and "How long does it take to verify Google business?"

Then we checked those against our own guide's subheads. It answers why profiles get suspended, the policy violation and appeals tool questions, how long an appeal takes, what to do when an appeal is denied, and what to do when the whole Google Account is restricted, which is what "how do I get my Google account unsuspended" is really asking. It does not answer the "suspicious activity" warning, how to contact Google Business Profile support, or how long verification takes. Those are the gaps we would fill first, each with a short, direct answer.

Two things stand out. Several of the most repeated questions are about the Google Account, not the Business Profile, so a page that only talks about the profile misses them. And Google offered different follow-up searches in each country ("how to fix" in the UK, "due to policy violation" in the US), so if you publish for more than one country, run the script for each.

What to do with the list

  1. Group the questions. Merge near-duplicates ("how to recover", "how to fix", "appeal") into one sub-topic.
  2. Answer each group in one or two sentences, under a subhead phrased the way people ask. That is the part a fan-out search can lift and link.
  3. Give big groups their own page. AirOps' data suggests a focused page that matches its question beats one page that tries to cover everything. Link the pages to each other.
  4. Leave navigational searches alone. Results like "My Business" or "Google Business Profile Manager" are people looking for Google's own tool, not an answer.
  5. Check you can be retrieved at all. Google needs the page indexed and eligible for a snippet; ChatGPT needs OAI-SearchBot allowed in robots.txt. Our ChatGPT guide covers the crawler side, and our post on AI Overviews controls explains how nosnippet limits what Google can show.

What we don't know yet

  • Google's real fan-out queries. Google doesn't publish them, in Search Console or anywhere else. People also ask and related searches are a stand-in, not the list itself.
  • How often AI Mode fans out. Ahrefs' 14.4% is for free ChatGPT. Google has given no comparable number, before or after Hydrogen.
  • Whether "broader range of links" means more clicks. Google's Hydrogen post gave no traffic figures. AI Mode clicks are counted in Search Console's Performance report under the Web search type, mixed with regular results, so watch your pages there over the coming weeks.

Sources

Photo of Kiran Dasgupta

About the author

Kiran Dasgupta, Editor

Kiran runs Kihan Marketing, a web design, SEO and ads agency, and manages dozens of websites across real estate, home services and healthcare. He writes about SEO, AI search and paid ads news, and what it means for the people who run websites.