BBC News is testing dozens of artificial intelligence pilots as it looks to speed up production, repackage content and create new formats without compromising its public service remit. The work is being led by Olle Zacharison, head of News AI, who spoke about the corporation’s approach on the Newsroom Robots podcast.
At the heart of the strategy is a push to understand where AI can strengthen rather than supplant BBC journalism. The corporation is organising its newsroom efforts around four themes – productivity, reformatting, augmenting reporting and improving user experiences – which Zachariasen said has helped teams see how individual pilots tie into long-term goals.
The productivity drive focuses on tasks where AI can cut repetitive work. Translation is the most advanced example, with BBC World Service teams testing side-by-side tools that speed up work across its 42 languages while preserving editorial control. Other experiments include transcription at scale, automated image descriptions and support for newsletter production.
Reformatting content is becoming a major priority as the BBC publishes text, audio and video across multiple platforms. News video is often produced several times for different aspect ratios, and Zacharison said AI-powered resizing and templating tools could remove duplication. The corporation is also piloting short daily podcasts for football fans created from website articles and voiced synthetically.
The BBC is equally focused on how AI can help reporters find better stories. Journalists are using tools such as Copilot and customised GPTs to interrogate documents, sift large datasets and spot angles that may otherwise be missed. Adoption varies across teams, but Zacharison said the biggest barrier is time rather than scepticism. He argued that journalists need space to experiment and that much of the value will come from peer-to-peer learning rather than directives from above.
The corporation is also exploring how AI can change the way audiences consume its output. Synthetic voices reading articles are already being tested. More ambitious conversational interfaces – similar to beta projects in his native Sweden – are under consideration, though Zacharison stressed that the BBC’s goal is to improve the news experience rather than mimic general-purpose assistants.
One of the organisation’s most significant bets is a fine-tuned large language model trained on BBC material. Its first use case, called “style assist”, rewrites stories from the Local Democracy Reporting Service to match BBC style while preserving meaning. Zacharison said the tool performs well but still requires close human oversight, particularly on political nuance and impartiality.
Coordinating AI work at the BBC’s scale remains a challenge. With 5,500 staff across the world, similar ideas often emerge simultaneously in different teams. Zacharison said grouping projects under the four strategic themes helps identify clusters of needs and avoid one-off pilots. Early involvement from product and tech teams is now seen as essential to avoiding dead-end ideas.
The BBC has blocked some AI assistants from scraping its content after research showed misattributions and distortions, though Zacharison said the organisation remains open to partnerships where its journalism is represented fairly. He also emphasised that the corporation should not underestimate its strengths – trusted reporting, deep archives, a global audience and a strong brand – as it navigates an era dominated by large tech companies.
