The New York Times has quietly started testing its first generative AI tool directed at readers: a redesigned search page shown to a small subset of visitors that answers queries with excerpts, links to relevant Times stories, and AI-generated summaries based solely on Times reporting. According to reporting, this is the news division’s first trial of AI-generated text that is not mediated by a Times journalist or editor.
Times spokesperson Graham James told Semafor the tool is intended to improve reader queries. “We are always testing new ways for our users to discover and engage with Times journalism,” he said, adding that the experimental feature demonstrates how the organization is using modern technology to create a better search experience.
In recent months the paper has also piloted AI functionality in non-news sections. Wirecutter, the Times’ recommendation site, has been testing an AI search called Wirecutter Finder that delivers AI-generated summaries and tips based on the site’s recommendations.
Pushback from newsroom staff
The introduction of AI-generated summaries in search has drawn skepticism from some rank-and-file journalists at The New York Times. Media newsletter Breaker reported earlier this month that management told the paper’s union during a July bargaining session that AI summaries would be tested in search. That disclosure caused frustration within the Times Guild, which argues the paper is violating its own editorial standards on AI usage.
The union has proposed rules for AI use that include creating a worker payment pool into which 22.5% of AI licensing deals for training would be placed and distributed among unionized employees. The union is also pushing for provisions requiring human oversight of AI outputs — a clause that, if enforced as written, would make the new search tool impractical in its current form.
Jim Luttrell, chair of the Times Guild unit and a senior staff editor, told Semafor the paper’s current AI policies are “meaningless.” He said management can change or violate the company’s policy at will, as they have with the summarization tool, and noted that past deployments of AI summaries have produced embarrassing mistakes. That, he argued, is why strong contractual guardrails are needed to protect readers’ trust in the paper’s journalism.
Why this matters for the Times
The Times is trying to catch up with search and summarization features that have become common across websites, apps, inboxes and other news organizations. Delivering a short, several-sentence summary of existing Times journalism based solely on the paper’s own reporting is not the most ambitious use of AI in news, but implementing generative AI presents specific challenges for an institution that emphasizes accuracy and careful editing.
Generative AI introduces clear accuracy risks at a time when the Times’ political adversaries could seek to exploit any errors. There is also a reputational concern: as news organizations position themselves as bulwarks against disinformation in an era of deepfakes and hallucinations, removing human oversight from even small parts of the journalism process can undercut that role.
Related experiments in the industry
The Times has already made internal AI tools available to its journalists for some time. Other outlets have also trialed generative AI features: The Washington Post tested an AI-generated podcast element that produced results later criticized for inaccuracies.
Overall, the Times’ small-scale search experiment shows the organization wrestling with how to incorporate generative AI while balancing product improvements, editorial standards, and newsroom concerns about accuracy and accountability.
