October 16, 2026 · ChimpanSEO

What did the Ahrefs study of 15 million ChatGPT fan out queries find?

Ahrefs analyzed 66.6 million ChatGPT responses and 15 million fan out queries, and the headline result reframes how GEO work should be prioritized. According to Ahrefs, only 14.4% of those responses triggered a web search, which means most prompts are answered from what the model already knows about a brand. When searches do happen, brand sites dominate: among the roughly 160 most searched sites, three in four searches targeted a brand site. For SEO and GEO teams, the practical shift is straightforward. Publish firsthand facts, keep them crawlable, and build brand memory over time.

What did the Ahrefs study of 15 million ChatGPT fan out queries find?

Ahrefs recorded that only 14.4% of 66.6 million tracked ChatGPT responses triggered a web search, out of 15 million fan out queries.

The study covers a large sample, and the shape of the data is more interesting than any single number. According to Ahrefs, searches that do fire tend to concentrate on brand properties, and the mix of fan out queries is shifting toward commercial topics. Price related fan outs climbed from 6.04% to 8.88% of all fan outs in one month, a relative rise of 47%.

Metric	Ahrefs finding
Responses that triggered a web search	14.4% of 66.6 million tracked responses
Most searched sites that were brand sites	Three in four, among about 160 most searched sites
Price related fan out share	6.04% to 8.88% in one month, a 47% relative rise
Reddit citation share in ChatGPT answers	About 1 in 5 in March, about 1 in 43 in late September

Why does a 14.4% web search rate matter for SEO and GEO?

A low web search rate means most ChatGPT answers rely on what the model already learned about a brand during training and prior exposure.

That single figure changes the order of operations for anyone doing generative engine optimization. If roughly seven out of eight answers never open a retrieval step, then brand familiarity accumulated before the prompt matters as much as page level optimization for the prompt itself. Google AI Overviews and Perplexity lean harder on live retrieval, so a portfolio approach still makes sense.

What does not change is the value of structured, extractable content. Retrieval augmented generation pipelines still parse the pages they find, and a clear answer on a crawlable page helps whenever a search does fire.

Why do brand sites dominate the most searched queries?

Ahrefs found that three in four of the most searched site: queries targeted a brand site, so ChatGPT checks official pages for facts.

The site: operator narrows a search to one domain, which suggests the engine is verifying something specific rather than browsing the open web. A pricing question about your product, for example, is best answered on your pricing page, not on a third party comparison post.

That has a direct consequence for technical SEO work: prices, specifications, supported formats and support policies need to sit in plain text on indexable URLs.

State the price figure in text, not inside an image or a hover widget.
Keep specifications in HTML tables that crawlers can parse.
Confirm that schema markup describes the same values users see.

Why did Reddit citations in ChatGPT answers drop so sharply?

Ahrefs reports that Reddit citations in ChatGPT answers fell from about 1 in 5 in March to about 1 in 43 in late September.

Alongside that drop, Ahrefs interprets the wider citation shifts as ChatGPT favoring original sources over sites that summarize them. Read together, the two observations describe a preference for the place where a fact was first published.

For content marketing, this raises the cost of being a middleman. An article that restates a manufacturer’s specification competes against the manufacturer. An article that runs its own test, collects its own data or publishes its own measurements holds something no summarizer can copy.

What should you do today to stay citable in AI answers?

Publishing prices, specifications and firsthand test results in plain text on indexable pages keeps a brand citable in 2026 AI answers.

The work splits into four habits that reinforce each other, and none of them requires abandoning classic SEO fundamentals.

Put prices, specifications and support details in plain text on indexable pages, and mirror the same values in schema markup.
Map the factual questions people ask about your brand, then check which page answers each one first.
Add original data, hands on testing or first party facts to articles that currently only summarize other sources.
Track how often AI answers cite you for price and review style queries, using a fixed prompt panel you rerun monthly.

Internal linking supports this work by connecting related facts. When a pricing page and a specification page link to each other, a retrieval system has an easier path from question to answer.

What should you avoid when you read a vendor study?

One vendor dataset from Ahrefs describes that sample, so SEO and GEO teams should not treat the numbers as a confirmed ranking rule.

Vendor research is useful because it quantifies patterns that practitioners otherwise sense but cannot measure. The numbers still describe one dataset, one platform and one observation window.

Do not turn a single dataset into a permanent rule about how an answer engine ranks sources.
Do not assume that a site: search guarantees a citation for the page it finds.
Do not pull budget out of community platforms because one citation share moved in one month.
Do not rewrite a content strategy before the pattern repeats in a second observation window.

Frequently Asked Questions

How often does ChatGPT search the web?

According to Ahrefs, only 14.4% of the 66.6 million ChatGPT responses it tracked triggered a web search. The remaining answers came from knowledge the model already held, which puts a premium on long term brand presence rather than live retrieval for every prompt.

What is a fan out query?

A fan out query is a secondary search that an answer engine generates while working on a user prompt. Ahrefs tracked 15 million of them, and its data shows price related fan outs rising from 6.04% to 8.88% of all fan outs in one month.

Does a site: query mean ChatGPT will cite my page?

No. A site: query shows that an engine is checking a specific domain for a factual answer, but citation still depends on whether the page is crawlable, indexable and written in clear plain text. Treat the operator as a signal of intent, not a guarantee.

Should I stop investing in Reddit and other community platforms?

No. Ahrefs reports that Reddit citations in ChatGPT answers fell from about 1 in 5 in March to about 1 in 43 in late September, but one vendor dataset describes one sample. Community platforms still shape brand perception and still appear in classic search results.

How can I track whether AI answers cite my site?

Run a fixed panel of prompts every month covering price, specification and review questions about your brand, then log which sources the answers cite. Over several months the pattern shows whether your pages are being pulled into AI answers or passed over.

Our reading

Our reading is that the Ahrefs numbers point to a split discipline rather than a single tactic. If these patterns hold, generative engine optimization is likely to reward brands that publish prices, specifications and original test data in plain text on pages crawlers can reach, while thin summarization loses ground. We would treat the study as a directional signal, not a rulebook, and we would watch how citation shares move over the next few months before rebuilding a content strategy around one dataset.