Funding: Own budget (15), Public broadcaster / state media (7), Tech company (Google/GNI) (4)
AI-generated 'best five Dutch books' list with factual errors — editor used ChatGPT to generate summaries for five Dutch novels without reading any of the books; author Thomas Heerma van Voss identified that the AI-written summary of his novel 'Het Archief' described the father as dead from the start — a factual error since the father is alive when the story begins; Heerma van Voss suspected all five summaries…
Algorithmic accountability investigations done by turning AI on AI — described by the researchers involved as "fighting fire with fire." With academic collaborators, the magazine created sock-puppet accounts and used a vision-language model to simulate user behaviour on TikTok. A second audit used the open ML tool Homepage2vec to classify 2.5 million Dutch-language websites present in large-language-model…
Two AI editorial workflow modules developed with Endava and Google Cloud: (1) "Hot Topic Discoverer" — analyses incoming news articles using weighted algorithms calibrated to each platform's editorial priorities (emotional resonance, geographical proximity, story uniqueness) to surface high-value stories for editors via Google Cloud Vertex AI; (2) ""Consistency Checker"" — validates factual accuracy across…
AI-fabricated-quotes incident: Peter Vandermeersch, a 'Journalism and Society' fellow at Mediahuis and former NRC editor-in-chief, used AI tools to summarise reports and then wrote from those summaries without verifying them, publishing fabricated quotes attributed to real people on his Substack.
'Zorgcowboys' — RAG (Retrieval-Augmented Generation) tool (JournalismAI Skills Lab 2025) making fragmented Dutch healthcare-regulator data (PDFs, open-data files, inspection and financial records) queryable by non-technical investigative journalists; built by reporter Niels van Nimwegen for Pointer's generalist team; version two in development aiming for a ChatGPT-like free-flowing experience with custom tooling…
AI-driven digital twin personas of diverse audience segments — virtual representations built from demographic, behavioural and social media data; deployed during content ideation and scripting to identify missing perspectives, reduce harmful representations and improve inclusivity; addresses cost and logistics barriers of traditional focus groups