Study

What Gets Cited by ChatGPT: Ahrefs' 1.4 Million-Prompt Study

ChatGPT reads far more than it quotes. Ahrefs analysed 1.4 million ChatGPT prompts, covering roughly 23.4 million cited URLs and 23.4 million non-cited ones, and the headline finding is how selective the engine is: it retrieves about 33 pages per prompt and cites 49.98 percent of them. Roughly half of what gets fetched never makes it into the answer. For SEOs and link builders, that selectivity is the whole game — getting retrieved is only the first hurdle, and getting cited is a second, harder one.

Half of what ChatGPT retrieves never gets cited

The scale of the study is what makes the finding hard to dismiss. With 1.4 million prompts and tens of millions of URLs on each side of the cited and non-cited divide, the 49.98 percent citation rate is not a quirk of a small sample. ChatGPT pulls about 33 pages for a typical prompt, reads them, and then quotes barely half. The other half did the work of being found and fetched, and still contributed nothing visible to the answer. Visibility in AI search is therefore a two-stage contest, and most content loses at the second stage.

That has a direct consequence for how teams should measure their AI search performance. Ranking in the retrieval set is invisible to the user, so a page can be fetched on thousands of prompts and never once appear in an answer. The metric that matters is the citation itself, and the gap between retrieval and citation is where most of the opportunity sits. Content that is merely relevant enough to be fetched is not enough; it has to be the page the model reaches for when it writes the answer.

Relevance decides which pages make the cut

The study gives a clear signal about what separates cited pages from ignored ones. Cited URLs scored 0.602 on similarity to the prompt, against 0.484 for non-cited ones, which means the pages that get quoted are the ones that match the question most closely. This is not about authority in the abstract or about having the biggest backlink profile. It is about answering the specific question in front of the model, in language close to the question itself. A tightly focused page beats a broad one.

That finding should be reassuring to smaller sites and to link builders working with niche clients. You do not need to outrank the biggest publisher on every topic; you need the closest, clearest answer to the exact prompt a user typed. Pages that wander, hedge or bury the answer under marketing copy are the ones that get fetched and forgotten. Pages that answer directly, in the terms the user used, are the ones that get quoted — and a quoted page is a page that can earn the click and the link.

The search channel is where citations are won

One channel dominates the citation picture. Pages pulled from the search channel were cited 88.46 percent of the time, which means that when ChatGPT goes to search results for its material, it quotes what it finds there almost nine times out of ten. That is a very different success rate from the overall figure, and it tells SEOs exactly where to focus. Traditional search visibility is not a separate discipline from AI visibility; it is the front door. If your page is the one search surfaces, it is very likely the one ChatGPT quotes.

The practical lesson is to keep doing the fundamentals and to stop treating AI search as a mystery. Strong search rankings put a page into the pool ChatGPT draws from, and close relevance to the query turns retrieval into citation. Link building supports both stages: links help the page rank in search, and they signal to the model that other people found the page worth referencing. The study does not describe a new kind of SEO. It describes the old kind, with a new scoreboard attached.

Sources

Sources: Ahrefs study, reported November 3, 2026.

More from Linkbaits

Turn this playbook into links

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