Washington Post study finds AI search eases filter-bubble fears, raises shared reading

A randomized field experiment with 37,561 Washington Post readers found that generative AI search can diversify what people read while also increasing what they have in common. Researchers Heeseung Andrew Lee, Dokyun Lee, Gwanhoo Lee and Dongwon Lee report that readers who got AI-generated answers with article citations above the conventional results consumed a wider range of topics, both individually and as a group, while shared exposure to the same information rose too — mostly through the AI answers themselves, without requiring a click into an article.