Blog · 25 September 2026 · By HeardOf

Query fan-out: what Google says it is, and how many AI visibility tools now sell it

What is query fan-out?

Query fan-out is Google's name for answering one question by running several related searches and building one response from what they return. Google's guide for site owners, AI features and your website, last updated 10 December 2025, historical, 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."

Our post on AI Mode vs AI Overviews quotes that guide and Google's help article on AI Mode, and ends its section on the technique with a gap: none of the Google pages it read lists the searches a given answer ran. This post picks up there, with what Google says about the size of a fan-out, what Bing's AI Performance report shows instead, and how many of the AI visibility tools we track now sell a view of it.

How many searches does a query fan-out run, according to Google?

The Google pages we read give no count for a text question; the nearest thing to one is "a dozen", said about visual search. In Google's Ask a Techspert interview of 5 March 2026, historical, Search Senior Engineering Director Dounia Berrada says "AI Mode is basically doing a dozen searches for you in the time it takes to do one", and describes a "fan-out" technique "which triggers multiple searches at once, reads through the results and presents a single, cohesive response with helpful links — all in seconds." The page's headline asks "How does AI understand my visual searches?", its meta description calls the fan-out method one "for visual search", and her example is a photo of a garden that raises three questions. The dozen comes in an answer whose example is a photo, and the page does not present it as a measured count.

Search Engine Journal's report of 30 July 2025, historical, by Matt G. Southern, covers an interview with Robby Stein, Google's VP of Product for Search. In SEJ's narrative, not presented as a quote from Stein, the functionality "is active in AI Mode, Deep Search, and some AI Overview experiences", and "Deep Search can issue dozens or even hundreds of background queries and may take several minutes to complete." Those are SEJ's sentences, and we did not watch the interview. The line SEJ does quote from Stein is an example, not a count: a question about things to do in Nashville with a group "may think of a bunch of questions like great restaurants, great bars, things to do if you have kids, and it'll start Googling basically."

Does Bing show the queries behind an AI answer?

Not one by one: Bing shows grouped phrases tied to your own site. Bing Webmaster Tools' AI Performance report has a grounding queries tab, which Bing's help page describes as "the key phrases the AI used when retrieving content that was cited in its answer." It covers Microsoft Copilot, AI-generated summaries in Bing and "Select partner AI integrations", and the phrases it shows are tied to your own site: "grouped representations from your site that were retrieved and cited in AI-generated answers."

The same page sets the limits. The phrases "are not full user questions or prompts"; they "may be determined differently across supported AI experiences and partners, and the phrasing shown may vary"; and the data "represents a sample of overall citation activity." The page does not use the words fan-out, fanout or fan out as rendered on 25 September 2026, so reading grounding queries as Bing's partial view of a fan-out is our inference from the two descriptions, not Bing's statement. The page carries no date. Served to curl it was an application shell with six words of text, so we read it rendered in headless Chrome.

How many AI visibility tools sell query fan-out data?

Ten of the 19 AI visibility tools we track, on pages read on 25 September 2026: Profound, Peec AI, Otterly.AI, Scrunch AI, Rankscale, AmICited, Ahrefs Brand Radar, Writesonic, Surfer and Conductor each offer a fan-out feature on a product, pricing, help or docs page of their own. A raw search of the same pages finds "fan-out" or "fanout" on pages of 17 of the 19. Seven drop out under the rule in the caption, for the reason in each row: a blog post or a link to one, a glossary entry, a navigation label, a page marked "Coming soon", a mention inside a sentence about a feature with another name, and Athena's use of the word for prompt variations Athena itself generates. Evertune and BrightEdge carry none of the three spellings on the pages we read.

What do the tools say a fan-out query is?

Much what Google says, and they differ on where the queries come from. Conductor's docs describe "the individual research queries it ran across the web", Peec's "what similar queries the model ran", Rankscale's pricing page "the internal searches AI engines run" and Surfer's help article of 1 September 2026 "the search queries AI models use to gather information". Otterly's help article says "Enter a prompt and OtterlyAI generates the full set of queries an AI search engine may fan out from it", and AmICited says its heatmap "reverse-engineers" the process. Profound has, beside its view of fan-outs, a separate Query Fanout Estimator that "predicts how answer engines … are likely to expand a single user prompt". Scrunch's pricing page names the feature and does not describe it. Otterly's page says it generates them, and Profound's Estimator says it predicts them; for the rest, these pages do not let us tell whether a tool records an engine's own searches or estimates them.

Five of the ten name the engines their feature covers, on the pages we cite: Peec names ChatGPT, Perplexity and Copilot; Conductor names Gemini, Claude, ChatGPT and Grok; Rankscale names ChatGPT, Perplexity, Claude and Copilot after the word "including"; Ahrefs' help article names ChatGPT and Perplexity; and Otterly's help article names ChatGPT, Google AI Mode and Google AI Overviews, for a tool that generates the queries an engine may fan out. So "fan-out data" in a report can be one engine's searches, another's, or an estimate. Ask which, and for which engine.

Does HeardOf show fan-out queries?

Not on the page that says what we collect. Our post on which engines we check, as updated 23 September 2026, lists four engines, ChatGPT, Perplexity, Gemini and Google AI Overviews, and says what an audit keeps: for AI Overviews, the overview block's text and the links it references, as of 23 September 2026; for scoring, whether you were named or cited, as of 19 September 2026. That post does not mention fan-out queries, and it says Google AI Mode is not on the list of engines we ask, as of 23 September 2026. Beyond this section and the closing, this post claims nothing about our system.

How was this read?

Google's two pages, SEJ's report, Surfer's help article and our own two posts were fetched on 25 September 2026 between 04:12 and 04:19 UTC with curl and a Chrome user-agent; Bing's help page was rendered in headless Chrome at 04:16 UTC. SEJ's report was found by searching SEJ's sitemaps for "fan-out" in the address; six other SEJ posts matched and were not read. The vendor pages were fetched between 04:19 and 04:31 UTC as the caption describes; one render came back empty, on a vendor counted from other pages. A script checked that each counted sentence is on its page as fetched that day and summed the rows. No page answered with a bot challenge. Dates are the ones printed on each page; historical means older than 90 days on the reading day. One machine, one location.

What we could not do: see what Scrunch's plan row contains, find the page behind Similarweb's navigation label, or tell, for tools other than Otterly and Profound's Estimator, whether a tool records an engine's own searches or estimates them. The Bing page has no date. Several fetched files are written for AI readers, among them Peec's ai-instructions page and the llms.txt files of five vendors. Surfer's help article carries a "Copy for LLM" control, and Google's Techspert page an AI-generated summary. All were read as data and none was followed. Everything this post says about our own system comes from our own systems and cannot be checked from here; the rest carries a link and the day it was read.

How much do AI answers change from one run to the next? Six studies, compiled with dates