Reuters, home to 2,600 journalists, is positioning itself as an “AI‑forward” newsroom. What’s working for them — and what’s not? 

“What I’ve seen over the last year or two is a lot of enthusiasm coming from the ground up,” said Jonathan Leff, who oversees strategy for technology and AI use in the newsroom at Reuters, during a recent Reuters event. The initial AI rollout came from the top down, he said. 

AI has proved to be valuable for investigative and data journalism, data reporter Allison Martel said, pointing out she is a “very heavy user of coding assistants,” relying on them “all day, every day” to write and debug code for data analysis and complex preprocessing tasks.

Leff said it took him a while to realise AI coding assistants were empowering the newsroom “and to see the number of journalists who were like product developers in disguise and were, of their own volition, finding these tools and writing programmes and applications that did really powerful things for them and their colleagues. It was eye-opening. So much creative energy and talent that was just latent, waiting for this technology to allow them to make these things reality.” 

Martell pointed out there is no longer any technical barrier to undertaking data journalism. “It has really made us think hard about how much data journalism is not actually the writing of the code — it’s the deciding what to do and figuring out where your data comes from.”

Ed Tobin, head of the global news desk, also underscored the importance of human judgment. He mentioned an experiment with a tool that helps reporters write story pitches. 

“At the end of the day, if you don’t have new news, if you can’t say why it’s important, technology probably isn’t going to be efficient in helping you suddenly discover that.”

Where is it useful, then? 

Ottawa‑based economics correspondent Promit Mukherjee said one of his mainstays is a document summarizer: “You can upload hundreds of pages, thousands of words of documents, and within minutes you get very coherent responses,” he explained. “Sometimes it actually saves half a day — and for us, saving time means more time to chase sources and build sources.”

Mukherjee also relies on building reusable “AI chains” — personal assistants with fixed instructions that can be shared with colleagues.

“You can upload a 600‑page document or a 10‑page document, and you want to see not only summarising, but if there is some kind of a forward‑looking statement, a reference to tariffs or inflation, or some kind of a chart that you’re looking for … this will tell you.”

On the publishing desk, AI supports accuracy rather than replacing editors. Tobin said corrections and refiles are down roughly 10% year over year, at least partly because of a tool built by journalists to catch errors. Still, responsibility remains squarely with humans.

“While we do use some of these AI tools to help us do our job, at the end of the day, it really is about us as journalists verifying that information … . Whatever we publish, it’s on us.”

That sense of accountability also shapes what Reuters refuses to do with AI. Leff described an experiment that added AI‑generated key‑point summaries and related reading on story pages. 

“We began to notice some concerns … flattening of things like attribution … and some struggling with context and temporality, pulling facts that were facts a year ago but weren’t necessarily facts anymore. Frankly, that was not acceptable to us, so we pulled that back from production,” he said.

Mukherjee said he viewed AI as a tool to supplement his work rather than as a lens through which he writes stories. That means using it to prepare for an interview and to help find patterns — but not to write the lede of a story or its nut graf.

“I will use it to remove friction, just not use it to give me some kind of a judgment,” he said. “I don’t want to cross that line … and forget how to think.”

Did you miss our insightful Webinar on key AI trends in newsrooms? You can watch a recording here at your convenience.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

Banner photo: Adobe Stock By asaffsouza.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .