What role does journalism play in a world dominated by AI?

The Dutch news sector is at a tipping point, and everyone in the room knows it. “The way people get their information is changing rapidly,” said Herman Wolswinkel, director of NDP Nieuwsmedia, at the opening of the debate ‘Media Policy of the Future’ yesterday in Amsterdam. “The informative function of news media is being taken over by AI models that have been trained on the hard work of editorial teams.” According to him, this is not a technical detail but a democratic alarm bell: if news no longer comes through journalism but through generative summaries from large platforms, what will remain of checks on power, pluralistic debate, and independent information provision?

The debate, organized by NDP Nieuwsmedia, the trade association of private news publishers, was not coincidentally held on the eve of the national elections. Two panels explored the situation that has arisen. Read more about panel 2 here.

Herman Wolswinkel © Butino

The first discussion brought together science, politics, Big Tech, and publishers: Alyt Damstra (Scientific Council for Government Policy, WRR), Volt candidate Bjorn Beijnon, CDA candidate Hemin Hawezy, Google Benelux spokesperson Rachid Finge, and Mediahuis Nederland CEO Rien van Beemen. The moderator was Monique van Dusseldorp.

The central question: who will safeguard the democratic function of journalism now that AI, search engines, and platforms are redrawing the gateway to information faster than policy can keep up? There was a strong consensus on the possible solutions.

A world in which Big Tech is the gatekeeper

Van Dusseldorp immediately set the tone: we live in a world in which there is war on the continent, digital sovereignty is faltering, and the democratic constitutional state is demonstrably vulnerable. Yet the election campaign focused mainly on asylum quotas, purchasing power figures, and nitrogen variants – not on the infrastructure of information. Yet that is precisely what is changing so radically, she argued.

New AI browsers, such as the Atlas browser launched this week, provide answers without clearly citing their sources. Large language models from Google and OpenAI, among others, present ready-made conclusions, often without context or source links. “The entire ecosystem of the web as we know it is changing,” said Van Dusseldorp. The user gets an answer, no longer a route to journalism. The traffic, the point of contact, the reader: they all remain within the platform. This makes the role of large tech companies in distribution, the advertising market, and – via AI systems – content presentation even more dominant, she said.

Journalism is a democratic public service – and policy is lagging behind

Alyt Damstra, who co-authored the WRR advisory report “Aandacht voor media” (Focus on Media), reminded the audience of the core message of that report: media have a democratic function (informing, monitoring, facilitating debate), and it is becoming increasingly difficult to perform that function properly in a “platformized” environment.

Her analysis is scathing: news media are being displaced in the online domain by platforms that rank, summarize, and market other people's content. At the same time, public debate is becoming more toxic and less safe. The WRR's conclusion: ambitious media policy is not a luxury but a prerequisite for a healthy democracy.

However, Damstra said, politicians have not really taken up this urgency. In fact, “A few weeks after our report was published, there was a threat of a VAT increase on news.” This measure is at odds with the idea that independent journalism must remain accessible to all. The WRR advocates a new type of media policy that no longer thinks in terms of the classic divide between “public broadcasting” and “commercial press,” or ‘newspaper’ and “TV.” “That world no longer exists. Everything is digital, everything competes for the same attention, and the threat to pluralism now comes mainly ”from outside": the global platforms. Policy must recognize that," said Damstra.

From press publishers' rights to AI training

If Bjorn Beijnon (fifth on the Volt list) could name one initial measure for a new cabinet, it would be the extension of press publishers' rights to AI training data. In plain language: news publishers already have (theoretical) rights to compensation if their work is reused or aggregated. But that right mainly concerns the distribution of news articles as we know them. It does not cover the large-scale scraping and archiving of those same articles to train AI models that then generate their own news-like summaries—without source, without context, without income for the original creators.

According to Beijnon, this gap must be closed. Not only nationally, but also at the European level. Because as long as models from parties such as OpenAI, Google, and other major players are fed journalistic work on a massive scale, the value of that work will disappear within the sector itself. This not only affects the revenue stream of editorial offices, but also the origin of our information: who ultimately determines what sounds “true”?

He linked this directly to digital autonomy. Europe must not only draw up rules, but also build its own infrastructure: its own storage, its own computing capacity, its own large language models (LLMs), and a better distribution of control over data. This will prevent dependence on a few American (and soon perhaps also Chinese or Gulf State-funded) players.

“Not democratization but concentration of power”

CDA candidate Hemin Hawezy added a political dimension. “The great promise of Big Tech was democratization: access to information for everyone, anytime, anywhere. The opposite has happened,” he said. Instead of an open, pluralistic, resilient media landscape, he sees a concentration of power, dependence, and eroding public values.

For Hawezy, “media policy” is not just about journalism, but about digital sovereignty: who owns the infrastructure, who manages the algorithms, who determines the rules of visibility, and who captures the advertising revenue? He wants substantial investments – “a growth fund of more than €12 billion” – in key technologies, European cloud solutions, and chip production on home soil. Not as industrial policy in the abstract sense, but as a prerequisite for a free press that is not trapped in a few American dashboards.

But, he admitted, this is a long-term project. The short term is more painful. Professions such as journalism and interpreting are under pressure from automation and AI assistance. According to him, political leadership means doing two things at once: protecting what is crucial to democracy – independent news gathering – and being honest about the fact that jobs are going to change.

All participants together © Jim Verhoef

Implementation of the AI Act

Where Beijnon and Hawezy mainly outlined the system, Rien van Beemen (CEO Mediahuis Nederland) came up with a concrete list for the prospective formateur: implement the European AI Act in the Netherlands, quickly and strictly. This is urgently needed because news publishers pay journalists hundreds of millions of euros a year to check facts, monitor power, and expose abuses. That journalistic work is the raw material for reliable information. According to Van Beemen, that raw material is now being included in datasets behind enormous language models without permission or compensation – including content behind paywalls. This undermines both the revenue model and the independence of journalism.

He wants two guarantees in the Dutch implementation of the AI Act and related regulations: First, transparency: publishers must be able to see in black and white which of their content has been used in AI training or AI summaries. Second, agreements must be made regarding compensation and control: according to Van Beemen, there must be an enforceable licensing model that allows publishers to enter into agreements with AI companies and platforms. Not only about money, but also about the terms of use.

According to Van Beemen, this is not a nostalgic resistance to technology. On the contrary, he emphasized that publishers embrace AI as a tool. But without clear “guardrails,” journalism will simply be drained of its raw material. That, in his words, is “more existential than a cookie battle.”

What does the user want?

Rachid Finge of Google was in a difficult position: at the negotiating table and in the dock at the same time. He called it “a pity that we are the only tech company that has come” and tried to dispel some of the mistrust.

Yes, he acknowledged, users increasingly want direct answers, summaries, and explanations. People are asking less “show me 10 sources” and more “what is the answer?” And yes, Google is experimenting with AI overviews, AI mode, and summary blocks at the top of the page. But according to Finge, that doesn't mean Google wants to exclude publishers. “We still send 9,000 clicks per second to publishers worldwide. We want to keep it that way,” he said. That's why Google is now testing “source attribution in the summary itself,” for example, “according to NRC,” as a clickable reference. This should ensure that traffic continues to flow, even in an AI interface.

Finge also defended the role of algorithms. Without algorithmic recommendations, he said, many new creators, ranging from NOS op 3 analyses to local creators, would never be found. He positioned YouTube and search results as places where pluralism is actually growing: more voices, more perspectives, more niche expertise than in traditional linear formats.

But beneath the polite tone, a strategic reality shone through that no one could ignore: Big Tech is redefining the interface between citizens and information faster than legislation can be formed. Publishers are demanding enforceable guarantees, politicians are demanding sovereignty, Brussels is building frameworks – and in the meantime, the models continue to run.

Our own LLM

But publishers are not sitting still either, Van Beemen emphasized. He explained how virtually all major Dutch news publishers are working with TNO on their own Dutch-language large language model. Not as a defensive reflex, but as a demonstration of capability: "High-quality AI in Dutch is not possible without journalistic content from here. So if you want AI in our language to be accurate, factual, and remain contextual, then the sector should be at the table."

That project is scheduled to go live in 2026. The goal is twofold: one, to show that innovation is not exclusive to Silicon Valley; two, to lay the foundation for a licensing model that allows the government and tech companies to access high-quality Dutch-language data while upholding journalism.

Beijnon (Volt) took this further: why continue to think in terms of “one model that determines everything”? Why not have an interoperable ecosystem of European, open, connected models in which publishers, governments, and citizens themselves control which data is used, shared, and viewed? In such a scenario, “Eurostack”, a European digital stack of infrastructure, cloud, data, and AI, becomes not a political buzzword but a public utility.

400 million

However promising, it does not change the tension of the moment. Van Beemen: “Local reporters, regional editorial offices, investigative journalists: they cost real money. Together, we spend 400 million euros a year on journalists. That money has to come from somewhere.” And that will surely work out, he concluded. “Journalism is tougher than people think. The Haarlemsche Courant already existed in 1656. Every generation has had its crisis. We'll sort out our own business model. But then politicians must enforce the European rules.”

Whether the democratic function of journalism will soon be determined in The Hague, Brussels, or Mountain View is no longer an academic question. It is election material. It just wasn't on the election agenda. That is precisely why NDP Nieuwsmedia organized this debate.

Read part 2 of this debate here.