The New York Times has begun showing AI-generated summaries to some readers who use its site search, a small experiment that puts machine-written text directly in front of the paper’s audience.
The test has rolled out in recent weeks to a small subset of visitors, according to Semafor’s Max Tani, who reported on the experiment Sunday after speaking with several Times employees last week. The new search page responds to queries with excerpts and links to Times stories, along with several-sentence summaries generated from the paper’s own journalism.
For the Times’ news operation, that is a notable line to cross. Semafor says it is the first test of AI-generated text for readers that has not been reviewed by a Times journalist or editor before it appears.
The paper describes the project more narrowly: as a search improvement.
“We are always testing new ways for our users to discover and engage with Times journalism,” Times spokesperson Graham James told Semafor. He called the feature an experiment using new technology to create a better search experience.
Several Times employees told Tani the paper’s internal search tools have long lagged behind its other publishing products. AI-generated summaries are one attempt to improve them.
The Times has already experimented with generative AI elsewhere in the company. Wirecutter, its product-review site, has been testing Wirecutter Finder, which gives users AI-generated summaries and tips based on its recommendations. But the new search experiment brings generated text into the news side of the Times and removes a human editor from the final step.
That has made some journalists inside the newsroom uneasy.
What do 1,000 journalists and PR pros know about AI that you don't? They took AI Quick Start, a 1-hour live class from The Media Copilot. 94% satisfaction. Find out how to work smarter with AI in just 60 minutes. Get 20% off with the code AIPRO: https://mediacopilot.ai/
Breaker reported earlier this month that management told the Times Guild during contract bargaining in July that it planned to test AI summaries in search. The union has been pushing for rules that would require human oversight of AI use, among other provisions.
Jim Luttrell, the Times Guild unit chair and a senior staff editor, told Semafor that the union wants stronger contractual guardrails because AI summaries can make mistakes and could damage readers’ trust.
Accuracy is the central tension in the experiment. The summaries are based only on Times journalism, but Semafor still points to the risk that generated text could get something wrong. The source material may be controlled; the output is not edited by a person before a reader sees it.
Tani also raises a more adversarial possibility. Someone deliberately trying to make the system produce a bad answer, then circulating the result as evidence against the Times. For a newsroom built around layers of reporting, editing and corrections, even a relatively modest search feature creates a new question about who is responsible when automated text gets something wrong.
Semafor notes that other news organizations have also experimented with reader-facing AI. The Washington Post, for example, trialed an AI-generated podcast feature that produced accuracy problems.
The Times has not announced a broader rollout. Semafor’s report also does not specify how the summaries are labeled or what happens when one contains an error.
The Times is starting small, with a search tool meant to help readers find more of its journalism. What happens next may depend on whether readers find the summaries useful — and whether the technology can earn the same trust as the reporting underneath it.
