This article was originally published on January 27, 2025 in the AI in News Media Newsletter. Get AI in News Media first and direct to your inbox by signing up here.
Hello and welcome back,
In this issue, we interview Head of Reuters AI Strategy Jane Barrett. You don’t have to follow AI in the media and publishing space particularly closely to know that Jane is an expert on the subject and has been at the front of the wave for a while now. Jane gives brilliant advice based on what she’s learned so far including how she pivoted from traditional journalism to video and then on to AI. She also gives tips on what to do if you’re feeling behind when it comes to AI and of course tells us what innovative and exciting AI initiatives we can expect this year from Reuters. We’re always looking for ways to improve, so if there’s something you want to see here you can reply to this email or get in touch on LinkedIn.
Thanks for reading,
Beth Ashton
Chief Growth Officer, Bright Sites
Interview with Jane Barrett, Head of Reuters AI Strategy
How’s your newsroom currently using AI, and can you discuss any specific initiatives?
Reuters has been using AI for a long time, but just older, “unsexy” AI and machine learning. The beauty of the big generative AI moment at the end of 2022 and early 2023 was that it really democratised access to AI.
Reuters was well ahead of the curve on investing in AI, so it was more of a turn in the road for us.
Now, with gen AI we have a general approach and a specific approach.
The general approach to AI is that we have our own sandbox, using Microsoft Azure, called Open Arena. We have all of the major models in there for our journalists to play, learn, test, experiment, and solve their own problems.
We’ve done a huge amount of training and had a mandatory training course last year. Then we ran workshops for people to learn about prompt engineering and encouraged them to come with a problem—something they wish AI would do for them. We aim to teach them to prompt their way into that solution.
On Open Arena, we have dozens of prompts that people have saved and keep honing to solve their problems. We did that because we wanted to lower the barrier to entry and help everyone to help themselves.
We have 2,600 journalists around the world. So we used an ease-impact analysis to determine where to build specific tools. We thought about which ideas would impact the greatest populations within Reuters. In those cases, we’ve built specialised tools.
Last year, we launched two tools.
One takes in all the corporate news releases worldwide. We already had AI handling corporate results, but this handles more unstructured data. The tool is called Fact Genie and sits in the CMS. It brings in all the press releases and returns alerts a journalist might want to send. Critically, it shows you exactly where the proposed alert came from in the press release so journalists can quickly check for accuracy before sending.
From over 100 journalists in different geographies who do this work, we’re seeing adoption rates of over 90%, which I think is amazing. I think it proves that if you’re trying to use AI in the newsroom, solve a real problem because that’s the way you’ll tell if it’s going to help and how much.
We also launched an AI assistant in our main text-editing CMS. We’ve already got live headline helping and bullet point summarisation. Soon, we’re launching editing tools. That’s where the generic and specific meet because a lot of generic tools have been built by individual editors. We’re honing those for more widespread use and moving them into our CMS.
This year, we’re launching our AI video script writer and AI content packager.
What are your priorities for this year?
If last year was about proving we could do things, this year is about scaling.
Last year was about text. This year, we’ll do a lot more in video. Many of our clients use video heavily so it’s an area where we can have a lot of impact.
Our first video tool is in beta. It transcribes all video audio from any language, translates it, identifies people in the shots if they’re in a public database and creates shot lists. This should help us speed up production and get videos to clients faster.
What is a sandbox, and why should publishers use it?
As a big company, we have a huge amount of IP and strong structures around protecting our data.
We don’t want our content training models unless we are being paid for it. By providing the controlled environment of a sandbox, we know our data is safe.
We also train our journalists to be very careful with anything that’s pre-published. If you haven’t yet published your story, we don’t want you putting it near a public tool even to get a headline or summarise it. Once it’s published, we’re calmer, but IP issues still come into consideration.
You have to make your own decisions about how comfortable you are using open models, given the terms and conditions.
Do you build everything in-house, or do you buy tools?
We do a mixture. In Thomson Reuters as a whole, we have a buy/build/partner approach to AI.
We have built a lot of our specific tools on the big well-known LLMs and are now looking at building small language models for some of those tasks.
On some less specific tasks, we should partner and not build. For example, we’ve been trialling synthetic voices from ElevenLabs, who are real experts in the text-to-speech field. They’ve built an amazing library of voices which get better by the month.
It comes back to knowing your use cases, your newsroom’s needs, your resources, your budget, how important latency is. Depending on the answer, you build or buy.
What was your journey from traditional publishing to Head of AI Strategy, and what advice would you give others?
This is my third transition. The first was from text into video, the second from traditional publishing to digital, and now from that into AI.
The key in this transition was throwing myself in even though I’m not a technologist. Instead I treated it like a reporting assignment and gave myself the task of learning about it. I started playing with ChatGPT and quickly realised this was the start of a major change. Then we made it practical.
A colleague and I chose two important newsroom tasks we thought GenAI could help with. We worked with a few amazing data scientists and did proofs of concept. We learned what was possible and wasn’t.
We shared the results and other people got excited – and scared – thinking, “We have to get on this quickly.” That’s when we started formalising my personal interest into a job, to bring structure to the next wave of transformation.
As part of that, I was invited to do a three-month secondment at TR Labs, our AI research unit. For the first three weeks, I didn’t understand anything, but the team was incredibly generous and just living through sprints, scrum ceremonies and the like taught me a huge amount both about AI and how to deploy it into production.
I worked with people half my age but twice as intelligent. It was brilliant—an opportunity to learn from people with different expertise and combine my knowledge of news with theirs on AI.
What would you say to someone who feels scared or left behind by AI in the newsroom?
Just jump in. This world of AI is moving incredibly fast. Even if your job is to lead AI strategy, new models or techniques come out constantly so you can’t always be at the front of the wave.
Give yourself permission not to have to know everything. Go back to first principles: how might AI help me or us? What should we try? And then get on and try it.