The New York Times is experimenting with AI-generated summaries for its readers.

      The New York Times has begun presenting readers with text crafted by a machine. A new search page responds to inquiries with excerpts, links to Times articles, and AI-generated summaries of Times journalism. Max Tani, the media editor at Semafor, reported on this development on Sunday.

      The newspaper quietly introduced this feature to a limited number of visitors over the past few weeks. It marks the first instance of the news division testing AI-generated text that hasn't been reviewed by a Times journalist or editor.

      The Times aims to improve a search function that Semafor describes as problematic, though some journalists within the newspaper are concerned about this initiative.

      Statements from the Times

      Graham James, a spokesperson for the Times, told Semafor, "We are continuously testing new methods for our users to discover and engage with Times journalism. This experimental feature exemplifies how we are utilizing the latest technology to enhance the search experience."

      Limitations of the tool

      The summaries are based solely on Times journalism and do not include information from other sources. This is not a chatbot providing answers about external events, nor does it access the wider web. This limitation does reduce potential issues, but it does not eliminate them. A summary that accurately reflects reporting can still misrepresent the original content.

      Wirecutter’s early adoption

      The Times has been implementing AI features in its non-news sections for several months. For example, Wirecutter, the Times' review site, recently began testing a search tool called Wirecutter Finder, which directs readers to AI-generated summaries and advice based on the site’s own recommendations. These tests have remained separate from the newsroom, unlike the new search summaries.

      Internally, the Times has provided its journalists with AI tools for their own use for some time. However, these tools are designed to support reporting rather than to be presented directly to the audience.

      Union discovery during negotiations

      During contract negotiations in July, Times management informed its union about the planned testing of AI summaries in search. A media newsletter called Breaker, authored by Lachlan Cartwright, reported on August 7 that the session became heated. Semafor referenced this report this week.

      The announcement led to frustration within the editorial union, which argues that the Times is violating its own editorial standards regarding AI usage.

      Union demands

      The Times Guild has proposed a set of AI regulations. One proposed rule would establish a payment pool for employees, distributing 22.5% of the newspaper’s AI licensing earnings for training data among unionized staff. Another rule would mandate human oversight of any AI implementation, potentially making the new search tool challenging to operate.

      Perceptions of insignificance

      Jim Luttrell, chair of the Guild unit and a senior staff editor, commented that the current AI policies of the paper are "meaningless." He stated, "Management can alter and even disregard the company’s policy however they please, as they are doing with the summarization tool." He added, "Every time AI summaries have been used, they have resulted in embarrassing errors. This is why we need strong regulations on AI in the contract to uphold readers' trust in our journalism."

      Rationale behind the initiative

      Search and summarization tools have become commonplace on various websites, applications, email platforms, and competing news outlets. Semafor characterized the Times as attempting to catch up. Several employees expressed to Semafor that the change is necessary, explaining that search has lagged behind the paper's other high-quality publishing products.

      Negotiations are ongoing, with management and the Guild returning to discussions this month over unresolved AI clauses.

      The accuracy challenge

      Summarizing the newspaper's own reporting in a few sentences doesn't represent an ambitious application of the technology. According to Semafor, this poses a greater challenge for the Times than for many other publishers since the organization relies heavily on multiple layers of fact-checking and editing. Even with these layers, the Times still makes mistakes and often issues corrections. It maintains a system intended to catch errors before publication and has built its reputation on this framework.

      Moreover, there is a risk from malicious actors. Semafor mentioned the possibility of someone trying to provoke the summarizer into making false statements, especially since the paper's political adversaries are searching for ways to undermine its journalism.

      Previous missteps in the industry

      The Washington Post experimented with an AI-generated podcast feature that produced mistakes and fabricated quotes, Semafor reported in December. Other publishers have relied more heavily on the technology and even paid for it; for instance, BuzzFeed slashed another third of its workforce after betting on AI.

      The spread of AI-generated text

      In July, Pew found evidence of AI authorship in approximately 10% of the sampled webpages. This context is where the Times has aimed to position itself against throughout the AI era. The rise of deepfakes, bots, and hallucinated content has complicated the fundamental question of what constitutes reality. News organizations have responded by promoting themselves as places where human oversight ensures