Being read by ChatGPT is not the same as being cited by it. That is the finding of an Ahrefs study that examined 1.4 million ChatGPT prompts and tens of millions of URL retrieval and citation outcomes. It reported roughly 23.4 million cited URLs, but those citations represented only 49.98 percent of the URLs ChatGPT retrieved. In other words, finding a URL did not ensure that it appeared as a visible citation. The gap between retrieval and citation is the most important number in AI search that most SEOs have never seen, because it separates two wins that are often confused.
The study separates two steps that happen inside every AI answer. Retrieval is what the system accesses: the URLs it fetches, reads, and considers. Citation is what the system exposes: the URLs it actually shows to the user as sources. Ahrefs counted both, across 1.4 million prompts, and found that about half of what was retrieved was never cited. The system read it, used it or discarded it, and moved on without telling the user it was there.
The distinction matters because the two steps have different audiences. Retrieval is about the machine: did the crawler find the page, could it read it, was it relevant to the question. Citation is about the person: is this page worth showing as a source, is it trustworthy enough to attach to the answer, is it the clearest version of the fact. A page can pass the first test and fail the second, and half of retrieved URLs do.
The reasons are the same ones that separate a good source from a merely relevant one. A page that is thin, derivative, or hard to attribute will be read and then passed over in favour of a page that states the fact clearly, carries a real author, and comes from a source the system trusts. A page that buries its answer in the tenth paragraph will lose to a page that states it in the first. A page that contradicts itself, or that disagrees with stronger sources, will be read and then set aside.
OpenAI's own crawler documentation draws a related distinction. It separates OAI-SearchBot, used for search functionality, from GPTBot, used for training data collection. The separation means that being crawled for training and being fetched for search are different processes with different purposes, and a site's robots.txt choices can affect them differently. A page that is available to one is not automatically useful to the other.
The practical conclusion is to optimise for citation, not just for retrieval. Retrieval is a technical achievement: the page is crawlable, fast, and relevant. Citation is an editorial achievement: the page is clear, quotable, authoritative, and trustworthy. Most SEO work already targets retrieval, because that is what classic search rewarded. The new work is the citation layer, and it is the work that determines whether a brand appears in the answer or merely contributed to it invisibly.
For link builders, the finding reframes what a good placement does. A link from a page that gets retrieved but never cited still helps rankings, but a link from a page that gets cited puts the brand in front of the buyer at the moment of decision. The pages worth earning links from are the ones that AI systems treat as sources, not just as reading material. Track citations as well as rankings, and treat a retrieved-but-not-cited page as a page that needs a clearer answer, a stronger author, or a fact worth quoting. The study's real lesson is that visibility now has two gates, and passing the first no longer guarantees the second. Build for both, and measure both, because a page that is read but never shown is doing invisible work.
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