When an AI summary answers the question well enough, do people still bother clicking through to the original?

This explores whether AI summaries that sit between readers and original sources reduce how often people click through to the source, and what follows when they do. The corpus has no direct study of search-engine AI overviews and publisher traffic. It does have a clean experiment on the underlying mechanism, plus several notes on what happens to an ecosystem when AI content captures attention that used to go to human sources.

The sharpest evidence comes from Nextdoor. LLM-written notification summaries were measurably more informative than the old ones, yet click-through rates fell Does better summary writing actually increase user engagement?. The reason is simple once stated: if the summary already answers the reader's question, there's nothing left to click for. So a summary that does its job well works against engagement with the original. A platform that rewards summary quality and a platform that rewards traffic to the source are aiming at targets that pull in opposite directions.

The social media notes show the same pattern one level up. AI-generated posts win engagement by being comprehensive, saying everything in one confident, complete package Does AI content displace human influencers on social media?. That same completeness suppresses replies, because nothing is left open to argue with or add to Why do AI posts get likes without inviting conversation?. Swap 'influencer' for 'original content site' and you get a likely story: attention flows to the synthesizing layer, while the people who produced the underlying material lose the visibility and reputation that kept them producing. A related point is that human writing actively asks for the reader's attention, and AI text inherits visibility without making that ask Does AI writing lack the internal appeal to attention that humans use?. That may be part of why summaries feel like an endpoint rather than an invitation to read further.

There's also a quieter twist: engagement numbers can rise while the experience gets worse. In a 680-person experiment, AI commenting tools increased participation, but readers rated the discussions as more generic and less authentic. The drop in perceived quality even affected conversations among people who never used the tools Do AI writing tools improve online discussion or degrade it?. So 'does engagement go down?' may be the wrong single metric. The more useful question is which engagement moves, and where to.

The longer-term risk is a feedback loop. About 35% of new websites were AI-generated or AI-assisted by mid-2025, and the range of topics and ideas across the web has narrowed How much of the internet is AI-generated now?. Writers accept AI text largely unedited (only 23% of paragraphs get changed at all) Do writers actually edit AI-generated text before publishing?. If summaries pull traffic away from original sites, and the sites that remain are increasingly AI-written, then summaries end up summarizing summaries. To go further on traffic effects for publishers specifically, you'd need sources outside this collection. What the corpus does show is why the effect happens and what it costs beyond the click count.