Ahrefs has analyzed 66.6 million ChatGPT answers to examine when the assistant triggers web searches and how it chooses queries and sources. In its study published on 12 October 2026, Ahrefs reports that only 14.4% of the analyzed answers triggered a web search. When ChatGPT did search, it ran an average of 1.6 queries. The findings offer a useful look at query fan-out, but they describe Ahrefs' dataset rather than all ChatGPT usage.
The study also reports that 75.9% of the most frequent searches using the site: operator targeted official brand websites. Separately, Ahrefs reports an 88% decline in Reddit citations over six months in another dataset. These findings concern different measurements and should not be combined as if they came from one identical sample.
What Ahrefs found
- 66.6 million answers analyzed: the study examines ChatGPT answers to investigate query fan-out and web-search behaviour.
- 14.4% triggered web search: most analyzed answers did not trigger a web search in the study's measurement.
- 1.6 queries on average: when ChatGPT searched the web, it ran an average of 1.6 queries.
- 75.9% brand-site targeting: 75.9% of the most frequent
site:searches targeted official brand websites. - 88% fewer Reddit citations: a separate Ahrefs dataset showed Reddit citations declining by 88% over six months.
What query fan-out means
Query fan-out describes how an AI system may turn a user's request into one or more additional searches to gather information before composing an answer. Instead of treating every prompt as a single conventional search, the system can issue targeted queries to retrieve supporting material.
Ahrefs' reported average of 1.6 queries applies to answers in its dataset that triggered web search. It does not mean every ChatGPT prompt produces 1.6 searches, because the study reports that only 14.4% of analyzed answers triggered web search in the first place.
Why the 14.4% figure matters
The finding suggests that, within the analyzed sample, web retrieval was not triggered for most answers. That distinction matters for AI visibility work: a brand's opportunity to appear as a web source depends in part on whether a response invokes search and which sources are retrieved when it does.
However, the result should not be interpreted as a universal rate for every ChatGPT model, user, prompt category, product surface or time period. The study's sampling and measurement methods determine what the figure can tell us. The headline number is a dataset-specific finding, not proof that 85.6% of all ChatGPT answers never use external information in any form.
Official brand websites in site: searches
Ahrefs reports that 75.9% of the most frequent site: searches it identified targeted official brand websites. This is relevant to brands investing in first-party information: official pages may be directly targeted by some retrieval queries, making clear, accurate and accessible brand information valuable.
The statistic is specifically about the most frequent site: searches in the study. It should not be restated as meaning that 75.9% of all ChatGPT citations, all web searches or all retrieved sources are official brand websites.
Reddit citations reportedly declined 88%
Ahrefs also reports an 88% decline in Reddit citations over six months, based on a separate dataset. That result may be worth monitoring for teams that track how AI answers cite community discussions, but it is not the same measurement as the query fan-out analysis.
The decline should not automatically be read as evidence that Reddit has become unimportant across all AI systems or that every topic has seen the same change. Citation patterns can vary by model, query type, source set and observation period. The dataset and methodology matter when interpreting the trend.
Implications for SEO and AI visibility
- Strengthen first-party information. Keep official brand pages clear, useful, crawlable and consistent, especially for product, service, policy and factual information.
- Measure citations separately from traffic. Being cited, being retrieved as a source and receiving a visit are different outcomes; track them as separate signals where possible.
- Study prompts and query intent. Analyze the questions that matter to your audience and how AI answers may branch into more specific searches.
- Do not optimize for one source in isolation. The Reddit result comes from a separate dataset and does not establish that community sources are no longer relevant.
- Keep study boundaries visible. Use the 14.4%, 1.6 and 75.9% figures with their original definitions instead of generalizing them to all AI search behaviour.
What the study does not prove
The findings do not establish a universal web-search rate across ChatGPT, nor do they prove that official brand websites dominate all AI citations. The reported Reddit decline is a separate observation and cannot by itself explain why citation patterns changed. These are proprietary research results, useful for generating hypotheses and guiding further measurement, rather than a complete map of how every AI system retrieves information.
The takeaway
Ahrefs' analysis of 66.6 million ChatGPT answers reports that 14.4% triggered web search, with an average of 1.6 queries when search occurred. The study also finds that 75.9% of the most frequent site: searches targeted official brand websites, while a separate dataset showed Reddit citations down 88% over six months. For SEO and AI visibility teams, the key lesson is to strengthen first-party sources and measure retrieval and citations carefully—without treating one proprietary dataset as representative of every ChatGPT answer.
Source
Ahrefs — How ChatGPT Does Query Fan-Out (15 Million Queries Studied)