By Mahesh Ramachandran
Chief Technology Officer, Reuters
Journalism has always been a race: to understand what happened, verify it, add context and reach the people who need to know. That mission hasn鈥檛 changed. What has changed is the terrain 鈥� and the tools that help journalists navigate it.
The latest tool, generative AI, buys journalists the one thing they need most in a race: time. By fundamentally slashing the time required to send news alerts from a press release, translating a story into Japanese or German or turning a transcript into a first drafts from hours into seconds, it creates space for higher-value work. That reclaimed time is where the real advantage lies.
At , we have spent the last two years building tools that close that gap in ways that preserve what a 175-year-old news organization cannot afford to compromise: accuracy, trust and editorial accountability.
When building AI-assisted editorial tools, the temptation is to optimize for speed: get the alert out faster, and get the summary live before the competition. Speed is fundamental and incredibly valuable at Reuters 鈥� and our clients depend on it.
But we understood that speed unmoored from accuracy is the opposite of what our customers need. A fast wrong answer is worse than a slow right one. Our tools had to be fast and correct, or they were not Reuters tools at all.
Two kinds of AI tools
The editorial tools we have built fall into two broad categories: tools that reduce friction in production, and tools that expand what is possible.
In the first category: our AI summarization skill, which now generates bullet-point summaries for 94 percent of major news stories, up from 33 percent before the tool existed. CheckMATE, our generative AI tool for enhancing machine translation quality, is used by more than 100 Reuters translators across French, Spanish, Portuguese, Italian, Korean, Chinese and Japanese. These tools do not produce content that goes straight to readers. They produce better starting points for journalists who then edit, verify and are accountable for the result.
In that same category sits LAMP, our packaging platform. Pairing a story with the right images, video and related items used to be careful manual work. LAMP uses a large-language model to match assets to stories with markedly high accuracy. The journalist still owns the final package. The machine just removes the assembly work.
In the second category: AI-voiced video packages in Spanish and Portuguese, which let us serve global markets at a scale that was previously out of reach. And Super Summaries, our AI-driven earnings intelligence product launched in July 2025, has expanded Reuters earnings coverage with more than 3,600 additional companies thus far. These tools create new value, not just faster versions of things we already did.
Video is the next frontier for that kind of reach. Live events generate hours of raw footage that someone must watch, log and cut before a single clip reaches a client. For the World Cup, we built an AI video-enrichment pipeline that helps turn the relentless volume of press conference footage into edited, ready-to-use raw video, at scale. It is the difference between covering a single moment and covering everything that is happening around it.
Why journalists have to be in the room
The most important lesson from this work is one that any technologist building for knowledge workers will recognize: the people closest to the work know things the technology does not.
Every tool we have deployed went through multiple rounds of development with journalists in the room 鈥� not as test users at the end of the process, but as co-designers from the beginning. When our alert suggestion and text annotation tool FactGenie got a news judgment call wrong, it was our journalists who could articulate exactly why it was wrong and what the model had missed. When CheckMATE flagged a translation error, it was our translators who knew whether the fix preserved the right register and tone. You cannot build that kind of editorial intelligence into a system without editorial intelligence built into the process itself.
The journalists who use our tools are not the only beneficiaries. The clients who depend on Reuters are as well.
When a financial professional opens a Super Summary on a company Reuters previously could not cover, that is a direct consequence of AI-powered scale applied with editorial judgment. When a media client receives a news package in Spanish that would previously have required a separate production workflow, that is AI-enabled reach. When an alert reaches a client鈥檚 screen ahead of the competition, that is AI-assisted speed, backed by the same editorial standards that have defined Reuters since 1851.
We are also building AI tools into the platforms our clients use directly. Reuters Connect has integrated AI-driven search. Reuters Imagen is deploying new features to speed up and enhance video editing. We launched the Reuters AI Suite in April 2025, starting with transcription and translation tools tuned specifically for news use. The goal is not just to give clients better content. It is to give them better workflows.
Setting the bar for what comes next
The professionals Reuters serves 鈥� journalists, financial analysts, media organizations, tech companies and consumers 鈥� share something with every changemaker: their work carries consequence. The decisions they make reach further than the task in front of them.
That is the opportunity AI presents for us: not to replace the judgment that makes Reuters valuable, but to give the people who exercise that judgment more time, more reach and better tools. AI helps them get to the story faster, understand more of the world and serve their audiences with depth and accuracy.
The technology that earns a place in this newsroom is technology that meets our standards. That is a high bar. And it is the same bar we are setting for everything we build next.
The right technology can change everything
The best don't just do their work. They change what is possible.
See how 鈫�
