There is a mismatch between how quickly tech companies are moving and how fast media organisations can determine the risks and opportunities associated with new AI models and tools.

And there is still very little cost-benefit evaluation and impact measurement when media companies use AI, according to a report published by the European Broadcasting Union (EBU).

I found four news companies — Rappler, Omroep Zwart, Radio Télévision Suisse, and BBC — were interesting case studies from the report:

In the Philippines, investigative news outlet Rappler wanted to hear from a wider, more diverse, range of its audience. But, as a small organisation, found it difficult to scale its focus groups. So, it tested a GenAI-supported solution.

Rappler created a virtual focus-group discussion. An AI system acts as a moderator in this group, asking an initial set of questions. Then it synthesises the participants’ responses — which could be in text and audio — and asks follow-up questions. Finally, it generates summaries of what was discussed.

Rappler found the tool generated many more insights than regular survey answers and was capable of picking up local languages reasonably well. Still, feedback showed more participants found the human-moderated consultations more engaging, meaningful, and trustworthy.  

Dutch public broadcaster Omroep Zwart also struggled with serving and representing viewpoints from its diverse audience, finding focus groups are difficult to organise and expensive to run. Its team created digital twins, which are virtual representations of diverse audience segments. The project used AI-driven digital personas to identify and integrate missing perspectives into the creative process.

“Early findings show that AI-generated feedback can help content makers identify harmful representations, improve inclusivity, and better understand audience sensitivities. The tool fostered more awareness of underrepresented perspectives during ideation and scripting stages,” the report said. 

Some key learnings: Diverse data sources (demographic, behavioural, social media) are essential for creating realistic and representative digital twins that capture a wide range of perspectives, while specific audience data enhances the relevance and usability of feedback from digital twins.

The team at Radio Télévision Suisse wanted to ensure it was meeting its audience’s needs when coming up with different angles to a story. So, RTS developed an AI model trained to operate on a diverse dataset of text, audio, and video while incorporating its journalists’ requirements and the editorial charter. 

This model categorises each piece of content according to the specific audience needs it fulfills, such as updating, diverting, inspiring, connecting, or helping. It automatically analyses incoming content and identifies the primary user need it addresses.

RTS journalists are keen to analyse their content offerings and identify coverage gaps or overproduction in content for meeting specific user needs. “The tool has had a significant impact on the understanding and acceptance of the user needs model within the organisation,” the report said. 

“Tools demonstrate value better than theory: Seeing the framework in action fostered genuine adoption among journalists.

“Workflow integration reduces resistance: By embedding the user needs model directly into existing content analysis processes, journalists naturally incorporate these concepts into their daily work rather than seeing them as a burden.”

The British Broadcasting Corporation realised its audience wanted more live coverage of football matches. Currently, editorial teams listen to and manually transcribe BBC Audio commentaries to feed into live pages that are updated with the latest action in matches through text and stills.  

So, the BBC piloted a tool that transcribes commentaries and delivers quotes and generative summaries based on those transcripts. Its editors look these over before publication.

“While the system-generated outputs contained some errors, with some player names proving particularly challenging, they were broadly accurate and compelling before being subjected to the editing process,” the report said.

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

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