ChatGPT skips the web for six in seven prompts, Ahrefs finds

ChatGPT only runs a web search for about one in seven prompts, according to a new Ahrefs study of 66.6 million responses. When it does search, it increasingly goes straight to the original source of a fact rather than to forums or summary sites, with Reddit's citation share in US answers falling from 19.1% in March 2026 to 2.3% in September. For founders, this means your own website and brand pages now matter more to ChatGPT visibility than ranking well on third-party community sites.

What did Ahrefs find?

Ahrefs published its query fan-out study on 12 October 2026, built on two datasets. Ahrefs analysed 66.6 million GPT-5.6 Luna responses and their 15 million fan-out queries from August and September 2026, plus separate citation data for US ChatGPT answers from March to September.

The headline number is about how rarely ChatGPT bothers to search the web at all. Only 14.4% of the 66.6 million responses involved a web search. When ChatGPT did search, it ran about 1.6 fan-out queries per searching response on average. In plain terms, most ChatGPT answers come straight from the model's training, not a live lookup, which makes the minority of searches that do happen disproportionately important for anyone trying to get cited.

The second part of the study tracks where those searches land. Ahrefs looked at the 200 domains most frequently cited by ChatGPT in US answers, based on 2.4 million answers in March and 3.8 million in September. The shift away from community and reference sites is stark: Reddit was cited in 19.1% of US ChatGPT answers in March and 2.3% in September, while Wikipedia went from 14.1% to 2.9% over the same period. Ahrefs is careful to note that both sites remain among ChatGPT's most-cited domains overall, just at a far smaller share of total answers than earlier in the year.

What's filling the gap is first-party and brand content. According to Relevant Audience's summary of the study, 76% of top site: searches hit brand sites, and PubMed rose to 4th in US citations, pointing to a model that is leaning harder on primary, authoritative sources rather than sites that aggregate or summarise other people's content.

A related finance-sector study backs this up with sector-specific numbers. Green Flag Digital tracked 215 personal finance pages ChatGPT cited most in October 2025 and found that a year later, 107 of those pages get no citations in September 2026, and the group kept just 14% of its citations overall. In that same period, bank, card issuer and government pages rose from 2% to 28% of citations, while Wikipedia and Reddit fell from 18% of top-500 citations to none.

Reddit's decline specifically has been tracked by several outlets with wildly different absolute numbers, which Ahrefs itself flags as a limitation. Search Engine Journal reported that despite the drop, in Ahrefs' separate worldwide sample, reddit.com was still the ninth domain ChatGPT searched most often with site: searches, which limit a search to one website.

Why does this matter for your business?

This study confirms a pattern that's been building all year: ChatGPT is quietly moving away from being a "wisdom of the crowd" tool that leans on Reddit threads and Wikipedia pages, and towards citing the people and companies the facts actually belong to. If you sell a product or run a service, that's an opening. Brand pages, pricing pages and official documentation are exactly the kind of primary source ChatGPT now seems to prefer when it does bother to check the web.

But the low search rate is the bigger structural point. If only 14.4% of responses trigger a web search at all, then for the vast majority of questions about your industry, ChatGPT is answering from what it already "knows" from training data, not from anything happening on the live web today. That means your visibility in ChatGPT depends heavily on whether your brand, your positioning and your facts were well represented in the data the model was trained on, not just on today's SEO.

This also raises the stakes for the searches that do happen. With brand sites capturing 76% of top site: searches, a company with a clear, well-structured, easily-parsed official website has a real shot at being the source ChatGPT reaches for, while sites that depend on being summarised or discussed by third parties (forums, comparison blogs, roundup articles) are losing ground fast.

The volatility itself is also a warning. Multiple trackers measuring the same period have produced very different Reddit numbers: one analysis, reported by The State of Brand, noted that Promptwatch put Reddit's pre-drop baseline at 3.83%, while Ahrefs' July 2026 dataset had Reddit as the single most-cited domain in ChatGPT at 16.7% of citations. If the measurement itself is this unstable, betting your entire AI visibility strategy on ranking well in one place, whether that's a forum, a review site or a single platform, is risky. A share of citations can evaporate within weeks for reasons that are hard to diagnose from outside.

What should you do now?

Start by treating your own website as the primary source of truth about your brand, products and pricing, since that's increasingly what ChatGPT reaches for during the minority of prompts where it does search. Make sure facts that matter, such as pricing, specifications, policies and comparisons, live clearly on pages you control, rather than only in forum threads or third-party listicles about you.

Second, don't assume a strong presence on Reddit, Wikipedia or Quora is a durable AI visibility strategy on its own. These sources remain part of the mix, but their citation share has proven to swing sharply in short periods, so community-platform visibility should be one channel among several, not the whole plan.

Third, remember that most ChatGPT answers never touch the live web at all. That means long-term brand reputation, accurate public information and consistent messaging across the sources that do get crawled during model training matter just as much as this week's SEO. If you haven't checked how your brand actually shows up across ChatGPT, Google AI Overviews and other AI systems, running a free audit with Sited is a quick way to see where you currently stand before deciding what to fix.

Frequently asked questions

Does this mean SEO for ChatGPT is pointless?

No, but it shifts the emphasis. Since only 14.4% of the 66.6 million responses involved a web search, the priority becomes making sure your brand's facts are clear and consistent wherever the model might have learned about you, with live-web optimisation mattering most for the smaller set of prompts that do trigger a search.

Why did Reddit's citation share fall so much?

Multiple trackers have measured the drop differently, but Ahrefs' US data shows Reddit was cited in 19.1% of US ChatGPT answers in March and 2.3% in September. The exact cause isn't confirmed publicly, but it coincides with ChatGPT favouring primary sources over sites that aggregate opinions.

Should I stop trying to get mentioned on Reddit or Wikipedia?

No. Both remain among ChatGPT's most-cited domains even after the decline, according to Ahrefs. The lesson is diversification: don't rely on a single platform when citation shares can shift this quickly.

What's a "fan-out query" in this context?

It's a follow-up search ChatGPT runs internally to gather more information before answering. Ahrefs found that when ChatGPT did search, it ran about 1.6 fan-out queries per searching response on average, a relatively narrow net compared with how AI Overviews or AI Mode tend to search.

How is this different from the Semrush study on brand mentions versus citations?

Semrush's AI Visibility Index looked at how often brands are mentioned versus cited across platforms generally. Ahrefs' study specifically isolates ChatGPT's search behaviour and shows it rarely searches at all, and when it does, it's moving towards primary and brand sources rather than aggregator sites.