Analysis
Google has started auto-expanding its AI-generated answers in search results rather than requiring a click to open them, The Verge reported. The practical effect is that the answer occupies the screen and the ten blue links move further down.
This is the latest step in a three-year progression, not a new policy. Google launched Search Generative Experience as an opt-in Labs product in May 2023, made AI Overviews generally available in the US in May 2024, and introduced AI Mode as a dedicated tab in 2025. Each step moved the generated answer closer to the default. Auto-expansion removes the last friction.
Publishers have documented the consequence throughout. News organizations and independent sites have reported material declines in Google referral traffic since AI Overviews became general, and the structural argument is simple: if the answer is on the results page, the click that funded the answer's source does not happen. Several publishers have litigated or filed complaints in the EU on exactly this point.
“This is the latest step in a three-year progression, not a new policy.”
The competitive context makes it rational from Google's side. OpenAI just opened ad inventory inside ChatGPT in India after launching US ads in February 2026, and Perplexity has monetized answer pages since 2024. If users are going to get answers without clicking somewhere, Google would rather that somewhere be Google. Alphabet's search advertising business generated the majority of its revenue last year; conceding the answer layer is not an option.
The counterweight worth stating plainly: nobody outside Google has clean data on the net effect. Google has consistently disputed publisher traffic claims and argues clicks from AI surfaces are higher quality. Aggregate referral data from third-party analytics firms is noisy and confounded by algorithm updates, seasonality and the general shift of discovery to social video. "Traffic is down" is well established; "traffic is down because of AI Overviews specifically, by X percent" is not.
For operators the planning assumption should change regardless. If your growth model assumes organic search compounds, model a version where it does not, and check what fraction of your pipeline survives. Content that exists to rank is depreciating; content someone would subscribe to is not.
