RevScope reviews how Google’s generated answers change the bargain between search engines and publishers. Its strongest evidence comes from an experiment that randomly hid AI Overviews for some US desktop users. It also examines whether getting cited in AI answers can make up for lost visits.
- The headline’s denominator matters: on searches eligible for an Overview, users made 0.37 outbound clicks when it appeared versus 0.62 when hidden. The working paper reports a 39.8% reduction, rather than a 40% fall in every website’s traffic.
- Citations offer uncertain compensation: Ahrefs reports AI assistants sending 0.28% of traffic across 74,752 sites in March 2026. High conversion rates at Ahrefs’ own site are a separate, single-site observation.
- Optimization claims are weaker: links and clear, sourced pages correlate with citations. The review finds little evidence for special AI files or guaranteed citation tricks.
The experiment covers two weeks of desktop Chrome use and remains a working paper. Broader industry panels describe changing search results, but cannot isolate AI’s contribution to every traffic decline. RevScope sells analytics, and several of its sources sell search-optimization tools.
The 530-comment thread on Hacker News adds opposing experiences from publishers and search users.
What the thread adds
- nicbou — “My traffic is down over 80%. The information I create is really easy to steal, but it’s as expensive as ever to create.” This is a personal report.
- robeym — “It’s true our search traffic took a dip, but our sign up rate and conversions are significantly higher.” A counterexample, without published measurements.
- pcarolan — “The best product for most questions is to just give you the answer.”
- reticulates — “A great interface that leads to less information being published is not a good thing.”
The unanswered question: uptrails, reticulates, _rwo, and nicbou all raise the future supply of original information. The article measures traffic loss; it leaves a sustainable replacement for that publishing bargain unresolved.
Handles are pseudonymous; HN publishes no per-comment scores. Ordering follows HN’s ranking, and this selection is a slice of the thread, not a consensus.