Brief IA
›

Ahrefs: Three-Quarters of Lost AI Citations Are Returning

AI Tools·Tom Levy·

Ahrefs: Three-Quarters of Lost AI Citations Are Returning

Ahrefs: Three-Quarters of Lost AI Citations Are Returning
Key Takeaways
1According to Ahrefs, about three-quarters of lost citations returned during a 20-day follow-up.
2Gemini cited no source in about 64% of the sample responses, compared to about 53% for Copilot.
3Ahrefs recommends counting citation days over the last seven days.
Why it matters : An isolated return after a page modification is not enough to establish that this intervention is the cause.

According to Ahrefs, about three-quarters of lost citations reappeared during a 20-day follow-up covering five AI products. A one-time absence is not enough to establish that a page needs to be modified, nor that a subsequent correction has had an effect.

A brand can remain visible without a link

The disappearance of a link does not necessarily mean the company's name is gone. A citation provides access to a site, while a mention refers to the presence of a brand in the text, with or without a link. Maintaining the name does not compensate for the lost citation: the indicators describe different objects.

In Ahrefs' sample, about 50% to 70% of the cited sites, depending on the tool, were only cited for one day. This limited presence for just one day concerned about one-fifth of the named brands. Mateusz Makosiewicz, an SEO and marketing educator at Ahrefs, suggests that mentions could be a better tracking indicator.

Tools can also use long or short forms to refer to the same company. Checking these variants helps avoid mistakenly counting a mention as lost.

Makosiewicz advises against considering a single reappearance as proof that a modification worked and recommends observing a progression in frequency over a longer duration. Conversely, losing a citation the day after a modification is a weak indicator of a potential negative effect.

Gemini omits sources in 64% of responses

In Gemini, about 64% of responses in the sample contained no citations, compared to about 53% in Copilot, according to Ahrefs. In a tracking table, a response without any source makes the monitored site appear absent. For these products, the loss often corresponded to the absence of any citation in the response, rather than the choice of a competitor.

A significant portion of the reappearances observed in Gemini and Copilot came from the restoration of sources after a response that provided none. Once these responses were excluded, their behavior resembled that of AI Mode and ChatGPT. For Gemini and Copilot, tracking can distinguish the frequency of citation of the monitored site from that of references to other sources.

AI Overviews, AI Mode, and ChatGPT presented a different profile: less than 1% of their responses were without citations. Makosiewicz estimates that, for these products, the disappearance of a site almost always means that another has replaced it. Reading the response allows for spotting this replacement or, conversely, noting the general absence of links; renewing the check in the following days is also part of the proposed verifications.

The tracking covers about 37,500 citations

Ahrefs examined about 37,500 citations from AI Overviews, AI Mode, ChatGPT, Gemini, and Copilot. Its protocol is based on a group of questions asked each day and represents only the perspective of one organization. Xibeijia Guan, a colleague of Makosiewicz at Ahrefs, recorded the sites present on the first day and then tracked their references until the twentieth; Makosiewicz documented the results.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

According to Ahrefs, about half of the lost citations returned within three days across all tools, with the timeframes counted from the first disappearance. However, the category without return only describes this observation window, and the site count does not indicate whether the recovered page is the one that was initially cited.

A separate follow-up conducted in Brand Radar showed a comparable movement but less frequent returns: about two-thirds of the lost citations eventually reappeared. Ahrefs markets this tool.

These daily observations complement research on variations between multiple responses to the same prompt. Results from SparkToro reported in January indicated that ChatGPT and the AI integrated into Google Search reproduced an identical list of brands in less than 1% of repetitions. The frequency of a brand's appearance was generally more stable than its rank. SparkToro thus favored a frequency per execution for brands, while Ahrefs examines the presence of sites over the days.

The duration of presence matters before a disappearance

The age of a citation before its first disappearance was associated with its likelihood of return. For AI Overviews, AI Mode, and ChatGPT, Ahrefs grouped citations according to this prior duration. Among those present for only one day, 34% to 41% had not returned during the follow-up. For those that remained for eight to fourteen days, the proportion was 4% or less.

A record on a given date does not capture this history: it does not distinguish a unique appearance from a continuous presence. The contrast can thus oppose a site cited once to another present for 10 days before disappearing. Ahrefs recommends counting, for each question, the days with a citation over the last seven days. This frequency measures the regularity of recourse to a site, without being confused with the duration of uninterrupted presence.

By the second day, about half of the initial references had disappeared from AI Overviews, and even more so in other products. Rewriting a page during this absence can lead to mistakenly attributing its return to the modification.

Makosiewicz recommends waiting a few days before considering a new citation as durable. For a disappearance, his rule is not to modify the page immediately but to re-examine it if the absence persists for two or three weeks.

Pages present in Google retain their citations longer

A separate aspect, conducted this time at the page level across 230 questions, compares references based on their presence in Google results for the corresponding question. On AI Overviews and AI Mode, pages also ranked in Google retained their citations longer, with more often a presence of a week or more. The gap was less pronounced for ChatGPT.

According to Ahrefs, presence in the results was more correlated with this longevity than the exact rank occupied by the page. This is a correlation, not a condition for accessing citations. The majority of pages cited by AI Mode and ChatGPT did not appear in the Google results for the relevant question.

In the face of a persistent absence, Ahrefs recommends examining both the visibility of the page in Google and its ability to answer the question accurately. If the ranking remains stable, the pages now cited in its place constitute a line of inquiry. If a loss of ranking precedes the disappearance in AI Overviews, the diagnosis may begin with the reasons for this decline in Google.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.