In Ahrefs' 20-day tracking of websites cited on day one, 50 percent of citations on Google AI Overviews, 58 percent on AI Mode, and 59 percent on ChatGPT were gone by day two, with Gemini at 81 percent and Copilot at 83 percent. About 75 percent of citations that disappeared at least once returned within the 20-day study, and about half of those returned within three days. The study, published on October 29, is a warning against treating a single day's AI citation as a stable result.
Ahrefs researcher Xibeijia Guan asked five AI tools the same questions daily for 20 days and followed every website cited on day one: 18,009 on Google AI Mode, 11,225 on ChatGPT, 4,830 on Google AI Overviews, 1,994 on Gemini, and 1,447 on Copilot. Ahrefs says it checked the pattern against questions it tracks in its Brand Radar product. The design is simple and powerful: instead of measuring whether a brand is cited on one day, it measures whether that citation survives, disappears, and returns over time.
The headline finding is volatility. A citation that exists today may not exist tomorrow, not because the page changed, but because the AI's answer changed. Different phrasings, different retrieved pages, and different model behaviour can all produce a different citation set for what is essentially the same question. Gemini and Copilot showed higher disappearance rates partly because their next answers often cited no one at all, which is a different kind of volatility from swapping one source for another.
The return rate is the more encouraging half of the finding. Roughly 75 percent of citations that disappeared at least once came back within 20 days, and about half of those came back within three days. That suggests AI citation is not a lottery in which a page is randomly chosen each day. It is a system with memory, in which pages that are relevant, authoritative, and well-matched to a question tend to reappear, even if they are absent on a particular day. A single day's absence is noise. A 20-day absence is a signal.
Ahrefs also found that pages cited for 15 or more days on AI Overviews had a median of 88 estimated monthly organic visits, compared with 8 for one-day citations, though it stresses this is correlation, not cause. Pages that are already visible in organic search, and that already attract visits, tend to be cited more stably. Domain Rating and backlinks, by contrast, made little difference in the study. That points to relevance and usefulness, rather than raw authority, as the stabilising factor for AI citations.
For SEOs, the study invalidates single-day citation reports. A tool that checks AI visibility once, on one day, will produce a snapshot that may look very different the next day, and a client who sees a citation vanish may panic unnecessarily. The right measurement is a rolling window: track citations over weeks, count how often a page appears, and report stability alongside presence. A page cited on 15 of 20 days is a durable asset. A page cited on one day is an anecdote.
That measurement discipline also changes how success is judged. A campaign that lifts a page from zero citations to occasional citations is progress, even if the page is not cited every day. A campaign that makes a page cited every day for a week, only for it to vanish, has not necessarily failed; it may simply be experiencing the normal churn Ahrefs measured. Judge AI visibility by trend and frequency over time, not by a single day's screenshot, and set client expectations accordingly.
For link builders, citation stability reinforces the value of pages that are genuinely useful for a question, rather than pages optimised for a single AI answer. A page that answers a question thoroughly, with original data, clear structure, and up-to-date information, is more likely to be retrieved repeatedly than a page built around a single phrase or a single AI prompt. That is the same advice as for organic search, but with a longer time horizon: build for repeated retrieval, not for a single citation.
Finally, do not chase daily citation changes with daily edits. Changing a page every time it drops out of an AI answer is likely to make it less stable, not more, because the page keeps changing while the AI's retrieval patterns settle. Instead, improve the page substantially, then leave it to be measured over weeks. Stability, in Ahrefs' data, comes from pages that are consistently relevant, and consistency is built by steady quality, not by constant tinkering.
Sources: Ahrefs blog, October 29, 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