Lost clicks from AI Overviews are no less valuable than kept ones

The comfortable assumption was that AI Overviews were skimming off the "bad" traffic: the curiosity clicks, the quick lookups, the users who would have bounced anyway. A study reported by Search Engine Journal demolishes that consolation. Visitors who arrived without clicking through an AI Overview showed no measurable difference in bounce rate, time on site, or likelihood to return to search compared with those who did click. The lost traffic, in other words, is not the expendable kind.

The study's method matters here. By comparing engagement metrics across sessions where AI Overviews appeared and sessions where they did not, the researchers isolated the attribution question that most traffic-loss analyses skip: are the clicks that disappear qualitatively different from the clicks that survive? The answer is no. Bounce rates, dwell time, and return-to-search rates, the three proxies most commonly used to infer whether a visitor found what they wanted, were statistically indistinguishable between the two groups.

What this actually means for traffic loss

The conventional defence of AI Overviews ran roughly as follows: yes, publishers lose clicks, but the clicks they lose are from users who were never really engaged. The model satisfies shallow queries; the deeper, more committed users still click through. That story is now empirically unsupported. If the lost visitors were genuinely less engaged, you would expect lower dwell times and higher bounce rates in the AI-Overview-present condition. The data show no such pattern.

This has a specific implication for how brands should account for organic traffic declines. A 20% drop in sessions is often modelled as a 20% drop in a relatively low-quality segment, and therefore a smaller-than-20% drop in business impact. The study's findings make that adjustment unjustifiable. Every percentage point of traffic lost to AI Overviews should be treated as a full-value loss.

For industrial groups, multilaterals, and financial-services brands, this matters in a way that pure-play publishers sometimes miss. These organisations rarely think of their websites as revenue channels in the direct sense, but they do rely on organic traffic to deliver policy documents, technical standards, investor relations content, and procurement information to audiences who act on it. A visitor from a qualified search query, whether they arrived from a blue link or never arrived at all because an AI Overview answered their question in the SERP, represents a real engagement opportunity foregone. The study confirms that the ones who never arrive are not the browsers; they include the decision-makers.

The second-order implication concerns content strategy. If there is no quality gap between retained and lost visitors, then optimising purely for click-through rate in an AI-Overview-heavy environment is a category error. The question is not how to win back the clicks; it is how to remain present in the moments that now happen inside the SERP itself. Citation in AI Overviews is not consolation for lost clicks. For brands whose visitors convert, subscribe, download, or otherwise signal intent on arrival, a citation that keeps the user in Google's interface is a substitution, not a supplement.

Google's standard line on AI Overviews is that they send higher-quality traffic by surfacing more relevant results. The study reported by Search Engine Journal is not saying that is false; it is saying the comparison between those who click with an Overview present and those who click without one reveals nothing to support a quality differential. The quality argument was always a way of making publishers feel better about volume declines. The data no longer support even that.

Brands that have been accepting AI Overviews traffic cannibalization as a tolerable trade-off, on the basis that the losses were low-value, need to revise that position. The losses are representative. Planning should treat them accordingly.