The New York Times starts answering queries the way the answer engines do

Retrieval over a closed corpus is the easy version of this build. What Semafor describes is several sentences of existing Times journalism, drawn only from the paper's own work ^[12], which removes the failure mode where a model wanders off into a stranger's blog post. It does not remove summarization error, and the accounting on that error changes. The paper runs copy through layers of fact-checking and editing and appends corrections to what still slips past ^[13]. A summary assembled from Times material with no editor between it and the reader has no outside party to carry the blame, so any mistake is a Times mistake inside the Times' own corrections machinery ^[1].

The building has had machine text in it for a while. Journalists have had internal AI tools for some time ^[16], and Wirecutter has been testing Wirecutter Finder, which points readers to AI-generated summaries and tips built from its recommendations ^[4]. Neither crossed the line this does. Several Times employees told Semafor the search work is necessary because the paper's internal search has long lagged its otherwise best-in-class publishing products ^[11]. Put those together and the first unedited machine writing the news desk shows readers is being spent on repairing the weakest product in the portfolio, not on anything the paper sells ^[2].

That is a modest ambition, and Semafor says as much: this is not the most ambitious implementation of AI in news media, and the Times is playing catch-up to summarization features already common across sites, apps, inboxes and rival newsrooms ^[18]. Catch-up has a track record. The Washington Post trialed an AI-generated podcast feature with less-than-accurate results ^[17]. Jim Luttrell, the Guild unit chair and a senior staff editor, told Semafor that every time AI summaries have been deployed they have made embarrassing mistakes, which is his argument for writing guardrails into the contract rather than leaving them in policy ^[10].

Management's framing is narrower than either the union's or the critics'. Spokesperson Graham James called it an experimental feature and an example of using the latest technology to create a better search experience ^[3]. That is a fair description of a search box. It is a thin description of the precedent, because the thing being tested is not retrieval quality but whether reader-facing prose can leave the building without an editor's name on it. The test cohort is small; the rule it establishes is not.