Will AI be accepted by journalists in the newsroom? Can it perform to the standards of the news media industry? And will the models deliver high-quality material across all the languages our customers work in?

These were some of the questions we asked ourselves before starting to build AI into the editorial workflow. They were also the same questions many of our customers were raising.

When we first introduced our AI-powered authoring tool in late 2024, we didn’t know what the adoption curve would look like. What we did know was that AI had to earn its place in everyday work. It had to be useful, trustworthy, and fit into established newsroom routines.

Over the past 10 months, we’ve seen encouraging signs. The number of monthly requests made to our AI service has more than doubled — from around 5,000 last autumn to more than 10,000 this summer. Peaks above 13,000 show moments of intense use, but perhaps more important is the steady, ongoing reliance that signals integration into daily work rather than one-off trials.

Building step by step

We’ve often described our approach as a “skateboard approach.” Instead of waiting to deliver the perfect solution, we introduced something small and useful, then kept improving it. Like starting with a skateboard, adding wheels, and eventually shaping something closer to a bike or car, the idea is to learn along the way and keep momentum.

The challenge is not just starting small, but also continuing to move forward. Each improvement we’ve made — whether in speed, usability, or language coverage — has come from working closely with customers and understanding what actually helps in practice.

What shaped our decisions

A few lessons stand out from this process:

• Flexibility matters: Different tasks call for different AI models. Allowing customers to choose the right one has been key.
• Language is critical: Supporting multiple languages has been more than a technical feature; it’s central to making the tool relevant and inclusive.
• Human oversight is non-negotiable: Journalists need to stay in control. We’ve focused on clear interfaces that make it obvious when AI is involved and easy to adjust or reject outputs.

Working with customers, not just for them

Most of the progress we’ve made has come from conversations, experiments, and feedback loops with customers.

Small details, like latency in responses or how prompts are structured, turned out to make a big difference. By addressing those, we’ve helped make AI a tool that supports workflows rather than interrupting them.

Looking back — and ahead

What started as a set of open questions has turned into real, measurable use in newsrooms. AI is not replacing journalists, but it is becoming part of how stories are planned, shaped, and published.

The usage data is only one signal, but it tells us AI is no longer just an experiment in editorial workflows. It’s beginning to play a role in everyday work. And like any tool in journalism, its value will depend on how thoughtfully we continue to develop it together with the people who use it.

For news companies more broadly, these lessons extend beyond AI in the newsroom. Any organisation working to build stronger relationships with their own customers can benefit from the same principles: Start small, learn quickly, and improve in collaboration with users.

By doing so, they not only introduce new tools but also strengthen trust, adaptability, and long-term engagement.

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