Most ChatGPT answers never touch the web at all. That is the headline finding of an Ahrefs study from October 2026, which analysed 66.6 million free, logged-out ChatGPT responses from August and September, along with about 15 million fan-out queries. Only 14.4 percent of the responses triggered a web search, and when ChatGPT did search, it generated about 1.6 fan-out queries per response on average. The study adds important nuance to the idea that every prompt triggers a bundle of searches, and it changes how SEOs should think about being found inside a chatbot.
Query fan-out is the process by which a single prompt triggers several searches behind the scenes. A person asks one question, and the chatbot breaks it into sub-questions, searches for each, and synthesises the results into a single answer. The concept has become a useful lens for LLM SEO, because it suggests that a brand can be found through a sub-query it never targeted, as long as it is the best answer to one of the questions the chatbot asks on the way to the answer.
Ahrefs' numbers put a scale on that idea. With 14.4 percent of responses triggering a search at all, most prompts are answered from the model's own knowledge, and only a minority reach the web. When a search does happen, the 1.6 fan-out queries per response mean the chatbot is usually asking one or two follow-up questions, not a dozen. The opportunity to be found is real, but it is narrower than the fan-out concept can sound in the abstract.
The study's most practical finding concerns site searches. In the study's sample of roughly 160 most-searched sites, 75.9 percent of site searches targeted brand-owned websites. When ChatGPT searches for a specific site, it is usually looking for the brand's own pages, not a third-party page about the brand. That means the brand's own site is the primary source for the questions the chatbot asks about it, and the quality of that site directly shapes what the chatbot says.
The implication is straightforward. Keep the brand's own pages accurate, complete, and easy for a machine to read, because when the chatbot comes looking, it comes to you first. A product page with stale prices, a help page with outdated instructions, or an about page that has not been updated in years will be the source the chatbot quotes, and the answer it gives will carry the error. The brand site is not just a destination anymore. It is the reference the AI checks.
The numbers describe a specific model and access state over a specific period, and they are not a universal rate for every ChatGPT user or other AI platform. Logged-in users, paying users, and different models will search at different rates, and the figures will move as the products change. But the shape of the finding is likely to hold: most answers come from the model, a minority come from the web, and the web searches that do happen are concentrated on a small number of questions and sites.
For SEOs, the strategy follows from the shape. Optimise for the minority of prompts that do search, by being the clearest, most authoritative source for the questions in your space. Make sure the brand's own site is the best source for questions about the brand, because that is where the chatbot looks first. And do not judge an AI visibility programme by how often the model searches. Judge it by how often the brand is cited when it does, and by whether the citation is accurate. The fan-out is the mechanism. The citation is the prize.
Sources: Ahrefs study, October 2026.
Linkbaits helps you create content that earns citations and backlinks. Start with the free tools and build your first linkable asset this week.
Start free trial