Researchers from the Washington Post ran a randomized test of a new generative‑AI search feature. The feature showed AI‑written answers that included citations to articles, alongside the usual list of results. Participants used the same archive as a control group that only saw the traditional list. The study was posted on arXiv on 1 Oct 2026.
What the test showed
Readers who received AI answers opened fewer articles directly from the result list but clicked more on the cited articles inside the AI response. The AI answers accounted for most of the increase in shared information because they delivered content without requiring a click. Overall, users searched a bit more often, which balanced the slightly lower number of articles read per search, leading to a small rise in total article consumption per reader and per minute.
Why it matters
For everyday news readers, AI‑enhanced search could expose them to a broader mix of topics, including less‑popular stories they might otherwise miss. This diversification may help people stay better informed about a wider range of issues while still seeing some of the same core stories as others.
Why it matters
Regular news readers could see a more varied set of stories, reducing the chance of only seeing the most popular headlines. At the same time, the shared core of information remains, which may support common public discourse. The exact long‑term impact on reading habits is still unknown.
