DPG Media is evaluating whether GPT-NL, a Dutch-language AI model, can assist its newsrooms with headline ideas and routine checks of article copy. The trial covers two publications rather than the whole group, and journalists retain responsibility for deciding whether to use the system’s suggestions.

For headlines, the model is being asked to offer alternatives that convey a story’s central point, fit the publication’s voice and attract readers without resorting to clickbait. Editors assess the options for accuracy and variety, including whether quoted remarks have been represented faithfully. The second application is copy-editing support: GPT-NL can flag lengthy or complicated sentences, suggest ways to divide them and check abbreviations and house style. TNO says editors from two DPG titles are expected to evaluate the results.

The study, which began in summer 2026, is scheduled to continue until the end of January 2027. DPG describes it as a feasibility exercise, not a deployment across all its news brands. TNO says the work will also examine how editorial quality criteria can be made measurable and used to improve the model for particular tasks.

GPT-NL is being developed by TNO, research and education cooperative SURF, and the Netherlands Forensic Institute. Rather than seeking to match the broad capabilities of major international AI systems, it is designed as a more focused model, with Dutch-language context and newsroom requirements among its potential advantages. DPG’s archive is included in training material assembled through NDP Nieuwsmedia, alongside archives from more than 30 national and regional publishers.

That dataset collaboration forms part of a wider effort to build a Dutch model using material made available under agreements with rights holders. TNO says the contributions comprise more than 20 billion tokens and were expected to double the quantity of high-quality Dutch-language data available to the project. The arrangements are intended to protect publishers’ material and provide for appropriate compensation if the model is commercialised.

The project’s ambitions extend beyond newsroom tools. TNO presents GPT-NL as an effort to give organisations greater oversight of how AI is built and used, including the handling of data, intellectual property and development choices. Its stated principles include transparency, trustworthiness and reciprocity, while SURF describes digital autonomy and respect for Dutch and European values as central aims.

Tweakers, another DPG-owned publication, is not participating in this trial. It says it uses AI services behind the scenes, including tools that connect to models from major technology companies, but these are separate from GPT-NL. Tweakers’ published account of its relationship with DPG says editorial decisions remain with its own staff, despite shared operational support in areas such as human resources and some IT services.

Source Reference Map

Inspired by headline at: [1]

Sources by paragraph:

Source: Noah Wire Services