VRT tests voice prototype that keeps AI news answers within its reporting
The Froomle-powered system lets users move through VRT NWS coverage and pose follow-up questions, while VRT stresses editorial context and reliability.
VRT Innovation is testing a voice-driven news prototype with the AI company Froomle that lets users navigate the latest VRT NWS coverage, skip articles and ask follow-up questions. The system is designed to adhere strictly to the broadcaster’s latest reporting, rather than draw answers from a wider pool of material.
Froomle supplies the voice technology, navigation and personalisation. VRT presents the conversational format as a way to make news more accessible for people who find reading difficult, as well as for those who want to understand the essentials quickly.
The experiment reflects VRT’s effort to make news more interactive than conventional articles, video reports and audio clips allow. But the broadcaster also argues that editorial news selection cannot be determined solely by user requests or by an algorithm’s predictions of what an individual will find interesting. Journalism, it says, must retain room for context, unexpected insights and issues audiences may not know to seek out.
VRT Innovation is also exploring conversational journalism through the Flemish Narrate research project. That work examines how AI could make news more personal and accessible while retaining a journalistic frame, and was presented to media and innovation professionals at the World Imec Technology Forum in Leuven.
VRT has tested dialogue-based formats on subjects including one year of Trump and the war in Ukraine. In those trials, users put questions to VRT NWS journalists and received answers instead of reading a single overview article. The format is intended to help people work through complex stories and obtain further explanation when needed.
VRT is developing its AI applications around its own journalistic sources, editorial choices and storytelling. It says this matters because generative AI systems can search large quantities of information but often do not clearly identify their sources, making reliable material harder to distinguish from misinformation. Established news organisations are also often underrepresented in such systems.
More than a quarter of Flemish audiences already use generative AI to search for news topics. VRT does not expect to launch a full chatbot in the short term, however, because dependable delivery needs more development and its archive is largely made up of video and audio, which remain difficult for current AI systems to handle.
Its stated principle is to use AI to present information from VRT’s own news archive clearly and accessibly, while keeping the future of news human-centred and interactive rather than fully automated.