November 1, 2026 · ChimpanSEO

What did the Ahrefs study of AI citations actually measure?

The Ahrefs study of roughly 37,500 AI citations changes the unit of measurement for GEO. Because at least half of those citations disappeared by the next day on every tool tested, and about 75% of the ones that vanished came back inside the study window, a single daily check captures churn rather than visibility. For anyone working on SEO and GEO in 2026, the signal worth reporting is citation frequency across a rolling window, segmented by question type, not a yes or no taken today.

What did the Ahrefs study of AI citations actually measure?

The Ahrefs study followed about 37,500 AI citations across five AI tools for 20 days, asking the same questions every day.

According to the Ahrefs blog, the team tracked those citations across Google AI Mode, ChatGPT, Google AI Overviews, Gemini and Copilot, repeating an identical question set daily for 20 days and publishing the analysis on 29 October 2026. The design fixes the prompts and watches the sources that answer them.

That makes the research a persistence test: it shows how long a cited URL holds its place for a given question. Persistence is exactly the variable that a daily GEO report tends to misread as a win or a loss, and it is the variable that separates a real visibility signal from a snapshot of noise.

How quickly do AI citations vanish, and do they come back?

At least half of AI citations vanished by the next day on every tool, ranging from 50% on Google AI Overviews to 80% on Gemini and Copilot.

Ahrefs reported the churn rate per surface, and the spread is wide enough to matter when you compare tools:

AI tool	Citations gone by the next day
Google AI Mode	about 60%
ChatGPT	about 60%
Google AI Overviews	50%
Gemini	80%
Copilot	80%

The return rate is the part that reframes the whole picture. Roughly 75% of the citations that disappeared at least once came back within the 20 days of the study, and citations that had already lasted longer were more likely to return. Around a third of the citations that lasted just one day never came back.

Put together, a missing citation is not an exit. It is one draw from a set that reshuffles every day, and judging a page on a single draw is guesswork dressed as data.

Which signals kept AI citations alive longer?

Google rankings, page titles and descriptions track how long AI citations last, while Domain Rating and page links barely move the needle.

Ahrefs highlighted four patterns that describe citation persistence, and two of them point straight back to classic SEO work:

Google rankings: on AI Overviews and AI Mode, pages that ranked anywhere in Google for the same question were about seven times more likely to keep their citation for at least a week. The gap was smaller on ChatGPT.
Title and description match: pages whose titles and descriptions closely matched the question kept their citations slightly longer.
Domain Rating and page links: made little difference, since long lived and short lived citations came from similarly strong websites.
Question type: questions naming specific brands had the most stable citations, while “best X” questions were among the least stable.

These are correlations inside one data set. Our reading is that rankings and question matched titles are the two levers worth testing first, because they are the ones this data links to persistence.

How should you report GEO visibility after this study?

AI visibility should be reported as citation frequency over a rolling window of days, split by question type, instead of a single daily yes or no.

Ahrefs itself recommends a frequency view, and the data supports it. A score such as “days cited out of the last 7” survives daily churn far better than a binary check, and it stays comparable across tools with very different volatility.

Pick a rolling window and report the share of days with a citation for each prompt and each AI tool.
Segment tracked prompts into branded questions, review questions and “best X” recommendation questions, then read every segment on its own.
Wait two to three weeks before rewriting a page over a missing citation, and check its Google rankings for that question first.
Rewrite titles and descriptions so they answer the question they target, then compare citation frequency before and after the change.
Log the date of every page edit, so a comeback cannot be credited to the wrong week.

What should you avoid when reading AI citation data?

Snapshot GEO reports turn daily citation churn into fake wins and losses, because most AI citations that vanish come back on their own.

The study rewards patience, and it punishes a few common reporting habits:

Do not present the correlations as causes. Rankings and title match are linked to persistence in this data set, and that is the full extent of what it shows.
Do not stretch the percentages. The figures describe five AI tools, one question set and a 20 day window, not the entire web.
Do not read a comeback as proof of an edit. Most lost citations returned within the study without any intervention.
Do not use Domain Rating as a proxy for citation stability. The study examined how long citations lasted, nothing broader.
Do not ignore the context. Ahrefs sells an AI visibility tracker called Brand Radar, and the analysis covers one month, so weigh vendor research with that in mind.

Frequently Asked Questions

Can one AI citation check tell you whether GEO is working?

No. In the Ahrefs data, at least half of citations disappeared by the next day on every tool, and roughly 75% of the lost ones returned. A single check samples daily churn, so track citation frequency across at least a week before drawing any conclusion.

How long should you track a prompt before judging AI visibility?

A rolling window of two to three weeks gives you a stable frequency per prompt and per tool, and it matches the horizon of the Ahrefs study. Shorter windows mostly measure reshuffling between answers rather than a real change in visibility.

Do backlinks and Domain Rating decide whether AI tools cite a page?

The Ahrefs study found that Domain Rating and links to the page made little difference to how long citations lasted, with long lived and short lived citations coming from similarly strong websites. That result covers citation persistence only, not the full set of selection signals.

Why should branded prompts be tracked separately from “best X” prompts?

Ahrefs reported that questions naming specific brands had the most stable citations, while “best X” questions were among the least stable. Mixing both into one average hides that gap, so segment the data before reporting visibility to anyone.

Our reading: the Ahrefs data suggests that a single citation check is a noisy sample rather than a measurement. We think GEO reporting is likely to move toward citation frequency across rolling windows, segmented by question type, and that page edits should be judged on frequency instead of one comeback. If that holds, teams that stay patient and structured in 2026 will read AI search visibility more accurately than teams that refresh a dashboard every morning.