AI Overviews shipped 14 May 2024; DNR 2024 still separates gateway use from AI authorship

The November 2026 Nieman Lab piece states the split. AI is an information gateway. Audiences still do not want it making their news. Primary records already show both halves. On 14 May 2024 Google documented AI Overviews as a generative layer on Search. Through 2024 OpenAI routed news queries into ChatGPT search. The Nieman text supplied for this assignment contains a headline, a URL, and a Hacker News thread with 20 points and 10 comments. It does not include a new dataset. Claims about audience refusal have to be checked against survey instruments, not against the headline.

Fletcher and Nielsen at the Reuters Institute surveyed Argentina, Denmark, France, Japan, the United Kingdom, and the United States on generative AI in news. Respondents more often expected AI-made news to be less trustworthy than more trustworthy. Acceptance ran higher for spelling, translation, and other assistive tasks than for writing articles. Digital News Report 2024 repeated that split across 47 markets. Comfort rises when a journalist remains the author. It falls when the model produces the story, with the lowest comfort on politics. Weekly chatbot use and stated comfort are separate measures. They should not be reported as one finding.

Coverage that stops at audience fear of bias misses the mechanism. A gateway ranks and compresses. A byline assigns responsibility for facts. Model curation weights training-data frequency. That is a measurable prior, not an intention. Summaries fail in documented ways: dropped caveats, merged attributions, unsupported specifics. Those errors matter when an overview substitutes for the source click. Human expertise remains a control variable in the survey design. Respondents want it on hard news. Publishers that automate the byline are not answering the same question as product teams that put a model in front of search results.

The next measurement point is the Digital News Report cycle, not a product keynote. Two series have to be read together: weekly use of AI answers for news, and stated comfort with news produced mostly by AI. If use rises and comfort does not, byline automation is running against the cross-market instrument. If comfort converges on use, the 2024 rejection was lag. Referral dependence on cited links in overviews is the business constraint. No changelog has made audience authorship preferences a solved deployment problem.