Google AI Overviews Cut Clicks, and Citations Barely Help

The citation consolation prize

For two years, publishers and marketers have operated on a hopeful premise: if AI-generated answers are going to sit on top of search results, the next best thing is to be the source those answers cite. A growing body of research suggests that this consolation prize is worth far less than many assumed.

New experimental work examining Google AI Overviews finds that these summaries change how people click, and that the source links attached to them recover only a small share of the traffic that would otherwise have flowed to publishers 1. The conclusion from that research is direct. Publishers need audience-growth strategies that do not depend on search referrals 1.

What the numbers show

The most rigorous evidence comes from studies built to separate the effect of AI answers from everything else changing in search. In a randomized field experiment, Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University tracked the browsing of more than 1,000 Google users 2. When an AI Overview appeared, outbound organic clicks dropped 39.8%, and searches ending with no click at all rose 34.5% 2. Pew Research Center has also analyzed a large sample of Google searches, adding observational weight to the experimental findings 2.

The counterargument that standalone AI platforms will make up the difference does not hold yet. Referrals from tools like chatbots are growing, but they remain far too small to replace what publishers are losing from traditional search 2. The pattern holds across search behavior, click-through rates, and publisher traffic data, and the evidence consistently points the same way 2.

The industry keeps chasing visibility anyway

While publisher-focused research measures losses, much of the optimization industry is focused on how to win AI citations. The guidance it offers is shifting. Ahrefs' study of 75,000 brands found that branded web mentions correlated with AI Overview visibility at 0.664, while raw backlink counts came in at just 0.218 3. Branded anchors (0.527), branded search volume (0.392), Domain Rating (0.326), and referring domains (0.295) fell in between 3.

The picture is not entirely consistent. SE Ranking's analysis of 129,000 domains found referring domains to be the single strongest predictor of ChatGPT citations 3. One way to reconcile the two is to treat backlinks as a threshold that must be cleared rather than a lever to keep pulling 3. Practitioners have not caught up with any of this. In a survey of 500 SEO professionals, 74% said backlinks affect AI visibility, but only 24% track AI visibility at all and just 19% have changed their link-building approach 3.

Marketing-analytics firm Mentionlytics pushes the brand-mention thesis further, citing a 0.737 correlation between mentions and AI visibility. It argues that third-party mentions now outweigh backlinks 4. It frames the 2026 landscape around "entity consensus" rather than keywords 4. Under this model, claims on a brand's own site need validation from multiple outside sources, and sentiment on community platforms such as Reddit and YouTube shapes how large language models present a brand 4. It also points to "LLM tribes," meaning loyal user bases for ChatGPT, Gemini, and Claude, and to agentic commerce as reasons to tailor content for machine readability 4.

Where the two conversations collide

Put side by side, these findings show a gap. The optimization playbook is increasingly sophisticated about how to earn a citation. The traffic research suggests that earning one does relatively little for the cited site 12.

The brand-mention findings also hint at a further problem for brand credit. If AI systems favor brands that are widely discussed elsewhere, as the correlation data suggest 34, then visibility goes to entities that already have broad recognition. Being named or linked inside an answer is not the same as a reader visiting, subscribing, or remembering where the information came from. The traffic data show that most users simply don't click 2.

One caveat applies. The correlation figures from Ahrefs and Mentionlytics measure association, not cause, and the two studies on backlinks partly disagree 3. The click-loss evidence rests on firmer ground because it includes a randomized experiment 2.

The reading

The sensible conclusion is that answer engine optimization is a defensive measure rather than a growth engine. Being absent from AI answers is probably worse than being cited. But for publishers in particular, a citation should not be counted on to restore lost referrals. The research points toward direct relationships with readers, such as newsletters, subscriptions, and owned communities, as the more durable bet 1. Brands chasing AI visibility should measure what citations actually deliver, something most of the industry is not yet doing 3.