Americans have mixed feelings about AI summaries in search results (www.pewresearch.org)

A new Pew Research Center survey finds AI-generated summary panels from search engines (like Google and Bing) are already a common part of U.S. web searches: 65% of adults encounter them at least sometimes and 45% see them extremely often or often, while 17% rarely or never do and 13% are unsure. Use and exposure skew young and educated — 62% of those under 30 report frequent exposure versus 23% of people 65+ (a 39-point gap), and 57% of college grads see them often compared with 38% with less education. Political differences are modest: Democrats and Democratic-leaners (49%) report slightly higher exposure than Republicans (42%), driven by liberal Democrats (56%). Despite wide exposure, Americans are lukewarm about value and trust. Only 20% find the summaries extremely or very useful (52% somewhat useful, 28% not very or not at all), and while 53% have at least some trust in the summaries, just 6% trust them a lot and 46% have little or no trust. Higher incomes correlate with more trust (63% upper vs 53% middle vs 50% lower), but trust and perceived usefulness don’t vary much by party. For the AI/ML community, these results spotlight rapid user-facing deployment and uneven adoption: search-generated LLM outputs are shaping discovery but face significant trust and usefulness gaps, underscoring needs for better transparency, attribution, evaluation of hallucination risk, and UX that helps users calibrate confidence.