This week I begin close to home. Two reports, in De Standaard and L’Echo, bring together four leaders of Belgian news publishing: Christian Van Thillo of DPG Media, Gert Ysebaert of Mediahuis, Bernard Marchant of Rossel and François le Hodey of IPM. Their groups publish most newspapers in Belgium and the Netherlands, a growing portfolio in France, and titles in Ireland, Luxembourg and Germany, giving these Belgian voices a distinctly European perspective. They are also among the principal shareholders of Belga, the agency I lead, whose capital is held entirely by Belgian media companies; Bernard Marchant chairs its board. I write about them with that relationship in mind, and with the conviction that their questions matter well beyond Belgium.
The central tension is easy to state. AI services create real value for their users from journalism that publishers pay to produce, while the terms on which publishers share in that value remain unsettled. Each of the four approaches that tension from his own position. Van Thillo defends the relationship with the reader, Ysebaert wants to build a market for journalism used by agents, le Hodey links the convenience of an AI answer to the cost of the reporting behind it, and Marchant asks for evidence.
The news of the past days puts each of these positions to a concrete test, and interviews from earlier in 2026 clarify what each of the four expects from the next phase.
This edition follows the four voices in turn, then the law and the wider debate. Quotations from Dutch and French sources are my translations.
I. On the relationship with the reader
Christian Van Thillo measures AI by what it does to the bond between a news brand and the people who read it.
34.1 — Christian Van Thillo counts the cost of an answer without a visit
DPG Media publishes HLN and De Morgen in Belgium and, among others, AD and de Volkskrant in the Netherlands, and has owned RTL Nederland since July 2025; for 2025 it reported revenue of €2.05 billion and 6,271 employees. Christian Van Thillo is its executive chairman and chairs the European Publishers Council (EPC). In De Standaard he says: “For now, we can only conclude that big tech has won. They are doing their own thing.” He describes the mechanism in one sentence: “People ask a question, get the answer within the search engine and are no longer directed to our websites.”
The loss goes further than traffic. A visit is the moment a reader recognises a brand and eventually decides to subscribe. NRC’s edition of the report includes Google’s reply that publishers can opt out of AI features in Search and that its AI answers contain links. That reply turns the debate into a practical test: can a publisher make a specific choice, understand its consequences and verify that it is respected?
In his open letter for World News Day, Van Thillo put the economics into two lines: “AI can draw on the record. Journalists create it.” He closed with a sentence that any finance director will recognise: “Journalism has a cost. Europe must ensure it continues to be worth paying for.”
Reflections
How much of the reader relationship can a publisher still build when an answer replaces the visit?
What is visibility in an AI interface worth to a publisher, measured against the subscription it may never sell?
34.2 — Publishers take the question of consent to Brussels
The EPC has filed a competition complaint about Google’s AI Overviews and AI Mode, alleging abuse of dominance and focusing on whether publishers have a realistic choice over the reuse of their work. On 29 October, five French organisations announced their own complaint, which, according to MACCUS, adds French evidence to the investigation the Commission opened in December 2025. The Commission has made no finding.
Van Thillo made the underlying argument in May, after the Court of Justice’s judgment in the Meta case on Italian rules for the remuneration of press publishers. De Tijd quoted him: “This judgment recognises the economic reality that publishers cannot negotiate on an equal footing with dominant online platforms without transparency, access to relevant data and safeguards against coercive behaviour.” The Court accepted, under conditions, national rules on remuneration, disclosure of information and protection against reduced visibility during negotiations. A licence acquires a price once both sides can see the use.
Consent also depends on product design. The New York Times reports that schools switching off AI in Google’s education software will also lose Google Search in school accounts from early 2027. A choice has practical value when refusing one use leaves the other services intact.
Reflections
What would a meaningful choice over AI reuse look like for a publisher that depends on the same platform for a large share of its audience?
How much information about a platform’s use of its work does a publisher need before it can put a price on a licence?
34.3 — Meta pauses its terms after publishers on three continents object
On 5 November, Meta indefinitely paused the update of its commercial and advertising terms, which would have allowed automated tools to collect content from websites linked to business accounts. A spokesperson told Nieman Lab: “We’re pausing the changes as we continue to engage with stakeholders and take our time to get this right.” The Associated Press had reported objections from Norwegian publishers, a court intervention in Norway and competition scrutiny in the United Kingdom, with Meta maintaining that the update concerned its business products.
Two days earlier, a São Paulo court granted Brazilian publishers an interim injunction that prohibits Meta from using their content without authorisation, including for AI training, as Valor International also reported. epd medien observes that a postponement leaves the provisions available for later. Digiday explains why leaving is hard: audience teams still rely on Instagram and Facebook for community and subscription sales.
The reporting leaves open whether collective representation, litigation or regulatory attention weighed most in Meta’s decision, and Editor & Publisher urges publishers to use the delay to price machine access. For Van Thillo’s argument, the decisive point is whether the next version of the terms will separate access to a platform from permission to reuse a publisher’s journalism.
Reflections
Which conditions should publishers agree among themselves before Meta returns with a new version of its terms?
How far can a publisher refuse a platform’s terms before the cost to its community and marketing exceeds the value of the content it protects?
II. On journalism for people and for agents
Gert Ysebaert sees AI as a new kind of customer for journalism, on condition that the terms of use are clear and measurable.
34.4 — Gert Ysebaert plans journalism for two kinds of reader
Mediahuis publishes more than thirty news brands, among them De Standaard, Het Nieuwsblad and Gazet van Antwerpen in Belgium and NRC and De Telegraaf in the Netherlands, with further titles in Ireland, Luxembourg and Germany. In 2025 it reported revenue of €1.258 billion and passed one million digital subscribers. In De Standaard, its chief executive says: “Without European regulation, publishers are powerless. The worst thing is that this disruption is happening with our own journalistic material.” He follows with an ambition: “We are going to produce journalism for people, and we are going to produce journalism for agents.” On the licensing agreements already concluded with AI companies, he adds: “That is why I do not necessarily see the deals that have been made as negative.”
The ambition opens a concrete product question. A human reader wants an article, an investigation or a podcast; an authorised agent may need a verified fact, a passage with its source or an update linked to earlier reporting. In the Mediahuis annual report, Ysebaert says that “AI can help us free up time and resources for what matters: distinctive journalism”, and that “soon people will only want to pay us for what other people have made.” At an INMA webinar in January, speaking as president of INMA, he advised publishers to treat platforms as “unreliable and unpredictable infrastructure” and concluded: “It’s not about being found. It’s about being chosen.”
The New York Times shows that both trends can coexist. According to TheWrap, it added 420,000 digital subscribers in the third quarter and raised digital subscription revenue by 14.6 percent, while chief executive Meredith Kopit Levien said that “the Times isn’t immune” to changes in discovery. Few publishers have that scale. Ysebaert’s position brings adoption and rights protection under one investment logic: efficiency creates editorial capacity, and that capacity is what readers and licensees pay for.
Reflections
What does a news product for an agent need to contain that an article for a human reader can leave out?
How can a publisher license its journalism to AI companies and still give readers a reason to subscribe directly?
34.5 — SPUR builds the measuring instruments of a licensing market
In June, Belga reported, in a dispatch published by Nieuwsblad, that SPUR, the Standards for Publisher Usage Rights coalition founded by Mediahuis with the BBC, the Financial Times, Guardian Media Group, Sky News and Telegraph Media, had grown to around thirty members. Ysebaert explained: “Shared technical standards, licensing frameworks and reliable measurement tools are essential to achieving fair and transparent remuneration for our content.” He added: “I believe this is also a turning point for European publishers to join forces and actively shape together how our journalism is used.” On 14 October, DPG Media joined a coalition that by then counted more than 80 members in over 20 countries, including The New York Times, AFP and The Economist, and that had launched a Content Telemetry Standard for recording when and how AI systems use content.
Each element solves a different problem: a standard describes use consistently, a licence states what is permitted and measurement makes it possible to check use and calculate payment. At a session organised by the Argentine publishers’ association ADEPA on 5 November, SPUR advisers David Buttle and Dominic Young distinguished retrieval, use in forming an answer, citation, presentation and audience engagement, since downloading an article, citing it and sending a reader back are different commercial outcomes. The PPA’s licensing briefing of 4 November adds context: large bilateral deals have slowed after 75 were concluded by April 2026, 250 publishers have joined the British collective licence run by PLS, and one Financial Times trial account attempted to retrieve a million articles in eight hours.
A licensing market needs a unit of account. A report carried by Yahoo Finance, citing Adweek and two people familiar with the matter, describes an arrangement under which Claude users received election night polling answers credited to the Associated Press; financial terms and duration were undisclosed, and neither party confirmed the deal on the record. If confirmed, it shows where agencies hold value: timely, attributable information that a model cannot produce by itself. A standard defines how participating systems should report, and adoption, compliance and payment remain the next tests.
Reflections
Which unit of use, a retrieval, a citation or a referred reader, should form the basis for payment?
How can smaller and regional publishers take part in standards that large groups and AI companies are shaping now?
34.6 — A paying subscriber asks whether an assistant may read along
In The News Stack, Bert Kok describes a modest personal test. He bought an eight-week digital subscription to de Volkskrant for €8 and let the AI assistant Codex log in, read and compare three articles about the American midterm elections; one search failed and needed another route. He then had Codex try twenty Dutch news brands: for twelve, article text came back to the apparent end without logging in, for three it did not, and five results were inconclusive. Kok himself warns that these are neither twelve independent publisher decisions nor representative national figures, and that technical access is a separate matter from permission to use the material.
De Standaard records a related episode. During a demonstration, Ana Jakimovska, responsible for AI strategy at Mediahuis, asked an engineer why his AI agent was accessing a site that prohibited automated collection; he replied: “The AI agent can use my computer. So the website sees it as a human visitor rather than a bot.” Access rules written for crawlers now have to account for delegated activity through a browser, where the agent, the user’s entitlement and the purpose of the request all need to be identified.
Two voices in the reportage supply a design brief. Marietje Schaake, a member of the Mediahuis board, encourages experimentation “within clear frameworks and according to principles that do justice to the promise of quality journalism.” Patricia Veldhuis, editor-in-chief of NRC, adds in NRC’s edition: “Our journalists can use AI as a tool in their research, provided we can then verify the results ourselves.” The subscriber with an assistant is a paying customer, and a reading aid, republication and training each deserve their own answer.
Reflections
Which forms of assistance should a subscription include, and where does use begin that requires a separate licence?
Which access conditions suit a legitimate subscriber who reads with the help of an assistant?
III. On the cost of reporting
François le Hodey connects the convenience of AI to the income that keeps a newsroom running.
34.7 — Le Hodey calls automated summaries industrialised plagiarism
IPM publishes La Libre Belgique, La DH, L’Avenir and Paris Match Belgique, and in May le Hodey was re-elected president of the European Newspaper Publishers’ Association (ENPA). In L’Echo, he says: “For consumers, it is wonderful to have these extremely effective summaries, but this is industrialised plagiarism taken to its maximum. There is no longer any way to monetise what we publish when it is automatically plagiarised.” He also points to “the enormous imbalance in negotiating power between press publishers and these major international players.”
His recognition of the user experience makes the commercial question more precise: how can a useful new service contribute to the production of the material on which it depends? In late October, ENPA and the EPC joined other European and Canadian associations in a transatlantic coalition asking for tools to authorise, monitor and enforce the use of their content.
Pierre Petillault, director of the French Alliance de la presse d’information générale, draws on the collective negotiations with Google, as quoted in L’Echo: “Without our unity, smaller publishers would have received almost nothing.” He adds: “A law alone is not enough. Without the political will to apply it, nothing moves forward.” Collective representation makes smaller titles visible in a negotiation while each keeps its editorial identity, and enforcement turns a statutory right into a credible negotiating position.
Reflections
How can a service that users find genuinely useful be brought to contribute to the cost of the reporting it summarises?
How can collective negotiation strengthen smaller titles while preserving each publisher’s commercial choices?
34.8 — Rossel and IPM combine to finance distinct titles
In Trends-Tendances in May, le Hodey explained the planned combination of Rossel and IPM: “The existence of professional, high-quality and independent newsrooms is directly linked to the ability to generate revenue, in other words to monetise the news offering.” He stated the intended outcome: “Our merger project will strengthen the economic capacity of the press, and therefore the possibility of maintaining quality journalism.” On 3 July, the Belgian Competition Authority authorised Rossel’s acquisition of IPM’s press activities, subject to commitments to preserve the diversity of the French-language daily press; according to MIND Media, the le Hodey family enters Rossel’s capital.
The same economics apply to the physical newspaper. After the end of the bpost concession, le Hodey warned in La Meuse, published by Sudinfo: “Without the tax credit, our model risks a cataclysmic crisis.” Marchant added: “A subscription in a rural area will tomorrow cost more than one in an urban area.” He also explained why the groups keep several titles: “We cannot reach every community with a single newspaper. We want to reach as many people as possible in Wallonia and Brussels. And that requires diversity.”
Canada offers a current comparison. News Media Canada welcomed the renewal of the 35 percent journalism labour tax credit and a permanent Local Journalism Initiative in the federal budget. Tax support, public funding and payment for AI reuse each have their own place in the financing of a newsroom.
Reflections
How can consolidation strengthen the economic base of the press while keeping the editorial diversity readers value?
Which part of the cost of distribution, on paper or on a screen, should public policy help carry, and on what conditions?
IV. On evidence and investment
Bernard Marchant holds that a right becomes commercially useful the moment its holder can prove how the work was used.
34.9 — Marchant asks whether the common training signal will hold
Rossel publishes Le Soir and Sudinfo in Belgium and, through Rossel France, La Voix du Nord, Courrier picard, L’Union and Paris-Normandie; the group reported consolidated revenue of €597.4 million for 2025 and, according to Le Monde, is negotiating to acquire 80 percent of Groupe Sud Ouest. In L’Echo, Marchant says: “It is up to us to prove things, which clearly shows that regulation needs to go further. But it is not very difficult to prove, and we have already been able to demonstrate it.” On a proposed common mechanism against unwanted scraping, he adds: “We will have to see whether it is watertight. Or whether it is once again something they do to ease their conscience.”
On 30 October, Anthropic, OpenAI, Google and Microsoft said they would support a common robots.txt signal, defined in an Internet-Draft by Anthropic’s Kevin Kelley, so that publishers can state one preference for all participating providers. The signal concerns the use of content for model training: OpenAI describes it as a way to express preferences about “foundational generative AI training”, and it leaves search answers, retrieval and browsing agents to other arrangements. Microsoft adds a separate preference for grounding answers, effective from 1 February 2027. Anthropic has begun implementation, OpenAI expects to follow “in the coming months” and Google next year, and Le Figaro describes the caution of French publishers towards a voluntary mechanism of this kind.
The EPC’s executive director, Angela Mills Wade, puts the problem precisely in her response to the Advocate General’s opinion: “The absence of evidence of use is not necessarily evidence that no use has occurred. Effective copyright protection requires meaningful transparency, traceability and the ability to verify how protected content is exploited by AI services.” A right to refuse is a first step; a publisher also needs to see its choice respected and to establish when it has been breached. Marchant’s test is operational, and a mechanism earns credibility through demonstrable compliance and a reliable remedy.
Reflections
What evidence should a publisher be able to obtain before it can consider a preference signal effective?
Who should carry the cost of proving use: the publisher, the AI provider or a public authority?
34.10 — American newspapers sue the supplier behind AI answers
On 2 November, The New York Times and, in a separate action, three other American dailies sued Brave Software in federal court in Manhattan, The Next Web reports. They allege that Brave collected their journalism, including content behind paywalls, and sold access to AI developers. TechCrunch adds detail; these are allegations, and Brave will answer them in court.
In the existing New York litigation against OpenAI and Microsoft, Reporters Without Borders and the actors’ and writers’ unions filed briefs supporting the publishers, while OpenAI points to submissions that support its defence.
Marchant’s emphasis on proof applies to the entire chain of supply. Permissions and evidence have to remain attached to content at every intermediary as well as at the final chatbot, and a workable licence makes that chain of access and onward use intelligible.
Reflections
How can a licence describe the permitted chain of access when content reaches an AI service through several intermediaries?
What should publishers expect from the courts, and what can only a negotiation deliver?
34.11 — Marchant and Philippot link enforcement to the citizen
In a joint interview in L’Echo on 31 October, Marchant meets Jean-Paul Philippot, outgoing head of RTBF. Marchant says: “In fact, regulation exists. The problem is enforcing it. Local publishers should not be left alone to police global players.” Philippot adds: “The AI giants are spending incredible amounts on infrastructure on the assumption that content is free, thereby denying intellectual property.”
On a possible joint video service, Marchant says: “An alliance with RTBF makes the most sense, given the quality and complementarity of the content. But it requires transparent governance.” Philippot sets a condition: “A public-service media organisation must absolutely be able to retain a direct link with the citizen.” The two disagree openly about RTBF’s free written news online, and agree on cooperation and enforcement.
Le Monde’s profile of Rossel adds the investment logic. “The effects of scale give us the capacity to invest, so we have to grow,” Marchant says, and: “For us, the press is not an instrument of power.” Le Soir’s editorial director Christophe Berti adds: “The first independence of the press is economic.”
Reflections
How can public and private media cooperate on distribution while each keeps a direct relationship with its own audience?
Which institution should carry the burden of enforcement when local publishers face global companies?
V. On the law in search of equilibrium
A European legal opinion and a wider international debate show how courts and negotiations each settle part of the terms.
34.12 — The Advocate General separates facts from expression
On 3 November, First Advocate General Maciej Szpunar delivered his opinion in Like Company v Google Ireland (C-250/25), the first generative AI case before the Court of Justice. A Hungarian publisher argues that Gemini reproduced its articles; Szpunar advises that retrieving a press publication, reusing it to generate an answer and providing that answer “may, in certain circumstances, infringe the rights of press publishers.” His test turns on two factors: whether the chatbot accessed the publication unlawfully, for instance by circumventing a paywall, and whether the answer reuses its form of expression so recognisably that the user no longer needs the original. Under his proposed approach, the provider could be liable where such use is inherent to its business model or where it has failed to take credible and effective countermeasures.
The press publishers’ right protects the form of expression and the investment behind it, while the facts reported remain free. Szpunar considers the training questions inadmissible for lack of evidence of training on Like Company’s articles. L’Echo reads the opinion as an opening for publishers, while De Tijd calls it half-hearted. A technology lawyer quoted by L’Echo, Geoffroy Blondiau, notes that the 2019 copyright directive “did not anticipate AI use.”
The EPC welcomed the reasoning that “making journalism freely accessible does not mean consenting to its use by AI chatbots”. The opinion is advice and binds neither the Court nor the Hungarian judge, who will decide the dispute once the Court has ruled.
Reflections
What responsibility should an AI provider bear when its answers remove the reader’s reason to consult the original reporting?
How can publishers’ rights remain effective when an AI answer draws on their reporting without visibly reproducing their words?
34.13 — The wider debate gives each side its own words
A DW report of 7 November widens the circle. Funke publisher Julia Becker recalls that search engines once sent traffic in return for indexing: “But now that exchange has broken down.” DW director-general Barbara Massing adds: “AI systems shouldn’t just use quality journalism. They should also pay for it.” Legal scholar Martin Senftleben proposes a collective payment for each answer that draws on journalism, and the report also records concerns about the effect of regulation on smaller AI developers.
The De Standaard reportage quotes a remark from litigation documents attributed to Microsoft research director Brent Hecht, “the greatest theft of human labour in history”, which Microsoft calls a personal opinion. Commerce Secretary Howard Lutnick argues: “We must not put up barriers to the next generation of innovators.” Kara Swisher observes: “The big-tech guys love the word frictionless; everything is seamless.” Google describes its cooperation with publishers through “commercial licences, product innovation and joint AI pilots.”
Each of these statements represents a real interest, and cooperation can be judged by measurable results: authorised use, meaningful discovery of the source, sustainable payment and continued editorial control.
Reflections
Which friction in reading the news should technology remove, and which belongs to the work of verification?
How can innovation policy encourage new AI services and secure the supply of reporting on which they depend?
Conclusion
Four Belgian publishers, four emphases and one shared conclusion: journalism needs a commercial relationship with the systems that use it. Van Thillo defends the reader relationship, Ysebaert builds the product and the standards, le Hodey ties viable newsrooms to revenue and Marchant asks for proof.
The central question is how value is shared between those who produce original reporting and those who build new services from it. A workable relationship between the two needs permission that can be exercised, use that can be measured, sources that remain identifiable and payment that supports further reporting.
These conditions affect news agencies as well, since our work often reaches the public under another brand. Belga’s media ownership anchors a shared commitment to the reporting infrastructure on which its clients depend. For news agencies the question is how AI services that rely indirectly on agency reporting will contribute to the ecosystem that finances this reporting.
Philippot describes an economy that pays for infrastructure and treats content as free, rather like a printing house that pays for paper, ink and presses and expects the manuscripts for free. Whoever wants to deliver reliable answers today has an interest in the reporting of tomorrow. Who pays for the next story?
Patrick Lacroix writes in a personal capacity. AI tools support research and drafting; he retains full editorial responsibility.