Ahrefs Examines Volatility in AI Citations

AI citations can change over time, and marketers should treat that volatility as normal rather than expect every mention to last, according to Ahrefs. In an article on citation stability, the company describes 10 lessons from tracking 37,500 citations and recommends focusing on how often a source appears over time.

The distinction is important for anyone assessing visibility in AI-generated answers. A citation appearing in one answer should not be treated as a permanent placement. Equally, the disappearance of an individual citation does not, by itself, establish that a source will remain absent. Ahrefs presents fluctuations as an apparent built-in characteristic of AI answers, rather than an exceptional event.

The company also notes that the degree of volatility may change substantially when models are updated. That adds another reason to be cautious about treating a snapshot as a lasting measure of performance. A review taken at one point can describe what appeared then, but it cannot guarantee that the same citations will appear without interruption later.

For marketers, the recommendation shifts the emphasis from defending each individual mention to evaluating the broader pattern. The question becomes whether a source is being cited more often over time, rather than whether every citation remains in place. This framing allows for movement in individual answers while keeping attention on the longer-term goal of being referenced.

It also affects how citation reports are interpreted. A change in an individual mention can be recorded without automatically being treated as a definitive result. When comparing observations, marketers can distinguish the presence or absence of a particular citation from the wider frequency with which their sources appear. Those are related measures, but they do not answer exactly the same question.

What this means for marketers

Track citation visibility over time and report recurring patterns alongside individual changes. Avoid making uninterrupted placement the sole measure of success, and keep the possibility of model-related volatility in mind when reviewing results.