VisibilityOS
notableReviewed Sat Oct 31

Randomized publisher experiment finds AI answers reshape citation exposure and reading behavior

An original randomized field experiment with 37,561 readers of The Washington Post archive compared conventional archive search with an interface also showing cited AI answers. The authors report more overlap in consumed topics, shifts toward cited articles and follow-up searches, and a small increase in article consumption per reader despite reduced article consumption per search.

Context

An original randomized field experiment with 37,561 readers of The Washington Post archive compared conventional archive search with an interface also showing cited AI answers. The authors report more overlap in consumed topics, shifts toward cited articles and follow-up searches, and a small increase in article consumption per reader despite reduced article consumption per search.

What changed

The study supplies randomized evidence about the effects of a cited AI-answer layer on search behavior, exposure to articles and shared topic attention in one publisher archive.

Why it matters

AI answer visibility and citation counts do not directly measure article clicks, reading behavior or downstream business outcomes. More searches and cited-article visits may offset lower per-search article consumption in specific settings.

Visibility OS interpretation

Visibility OS treats this as strong setting-specific evidence that answer exposure, source clicks, repeat searches and total article consumption should be measured separately, not as a general guarantee of referral lift from commercial AI search.

What remains uncertain

Would effects replicate on open web search, commercial queries, other publishers and other models? How much of the shift is caused by the interface, ranking or citation design? What are revenue effects?

Evidence claims

Emerging Evidenceintroduces

In a randomized field experiment with 37,561 Washington Post archive readers, Lee and colleagues found that adding cited AI answers above conventional results shifted search and article-consumption behavior, increased shared topic exposure, and was associated with a small increase in articles consumed per reader.