# What is query fan-out, and why does it change which pages AI answers quote? Source: https://www.ahalevel.com/query-fan-out/ Last updated 2026-09-25 · By Lonzo Sheffield, AHA Query fan-out is how AI search answers a question: it splits the question into several narrower searches, runs them, and builds one answer from what they return. Google says AI Overviews and AI Mode may use it. OpenAI says ChatGPT search rewrites a question into one or more targeted queries. So a page can be quoted for a narrow question it answers clearly, even if it never ranks for the question the customer typed. ## How does query fan-out work? The AI system does not search for your question as typed. It rewrites it into narrower searches, sends those to a search index, reads the results, and often sends follow-up searches based on what came back. The final answer is assembled from the pages those searches returned, not from one results page. Google describes AI Mode this way: it “uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf” (Google, May 2025). OpenAI gives a local example in its help center: a question like “What are some good restaurants near me?” can be rewritten, using the general location of the person asking, into the search “top restaurants San Francisco.” After reviewing the first results, ChatGPT search may send more specific queries. ## Which AI systems use query fan-out? Google and OpenAI both say so in their own documentation. Google says AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources to build a response. OpenAI says ChatGPT search typically rewrites a question into one or more targeted queries. Perplexity and Microsoft have not published an equivalent statement. _What each company has said, in its own documentation_ | System | What the company says | Source | | --- | --- | --- | | Google AI Overviews and AI Mode | May use a “query fan-out” technique: multiple related searches across subtopics and data sources | Google Search Central | | Google Deep Search (in AI Mode) | Issues hundreds of searches for one research request | Google blog | | ChatGPT search | Typically rewrites a question into one or more targeted queries sent to search providers, then may send more specific ones | OpenAI Help Center | | Microsoft 365 Copilot Chat | Generates a short query from the prompt and sends it to Bing; no statement of multiple queries | Microsoft Support | | Perplexity | No official statement found | — | ## How many searches does one question turn into? It varies by system and by question. In one third-party study of 102,018 searches generated from 38,418 prompts between September and November 2025, ChatGPT averaged 3.51 searches per prompt and ran more than one search 67.3% of the time. Perplexity averaged 2.24, and 70.5% of its prompts ran a single search. The study is Qwairy’s Q3 2025 query fan-out study. It measured what the systems did through their search tools during those months; the numbers will move as the products change. Google has not published a per-question average for AI Mode. It says only that Deep Search issues hundreds of searches. ## Why can a page that ranks on page three still get quoted? Because the AI system did not search for the customer's question. It searched for narrower questions, and a page that answers one of those clearly can be pulled in whatever it ranks for the original. That is consistent with Semrush's finding that almost 90% of the time, the pages ChatGPT search cites rank at position 21 or lower for related queries. Google’s own AI answers show the same drift. In Ahrefs’ early-2026 study, 38% of AI Overview citations came from pages in Google’s top 10 for the query, down from 76% the previous July (Ahrefs changed how it detects citations between the two runs, so part of the drop is measurement). The Semrush study is from July 2025. Neither study measured fan-out directly; fan-out is the mechanism both companies’ documentation describes. ## What does query fan-out mean for a local business? A customer's one question becomes several searches, and each can return a different source about the same business: the Google Business Profile, a directory listing, a review site, the business's own website. The answer is assembled from all of them. If those sources disagree on the phone number, service area, or license, the answer can carry the wrong one. An illustration, not an observed log: a question like “who’s a good roofer near me?” could be rewritten with the asker’s city, then followed by narrower searches about licensing, reviews, or roof replacement cost in that city. The business that states those facts clearly, and the same way everywhere, gives each of those searches something accurate to find. This is why AI SEO is the condition of everything published about a business, not a tactic applied to one page. ## Can you optimize for query fan-out? Not directly. Nobody outside the AI company sees the searches a question becomes, and they change from one run to the next. Google says the best practices for SEO continue to be relevant to its AI features, and that no special schema.org markup is required. What a business can control is whether each narrow question has a clear, accurate, consistent answer somewhere a search can reach. Google’s wording is in its AI optimization guide. In practice that means one question per page with the answer at the top, facts that match across every profile and listing, and measuring which sources the AI systems actually cite, on a dated schedule, so changes can be checked rather than assumed. That measurement is what the AHA Level records. ## Who is AHA? AHA Level — Anti-Hallucination Architects. AHA Level (Anti-Hallucination Architects) measures what AI systems and search channels tell customers about local businesses — scored as the AHA Level across Google, AI search, reviews, social, directories, and the website itself — and fixes what's wrong. Based in Wesley Chapel, Florida; serving Tampa Bay in person and US businesses remotely. Founded 2026. Contact: hello@ahalevel.com.