Dispatch 29
Independence is easy to declare and expensive to organise. This week I saw what the organised version looks like, in Zurich, at the annual meeting of Group 39 and during a company visit to Neue Zürcher Zeiting - NZZ hosted by the Swiss news agency Keystone-SDA. The alliance takes its name from 1939, when Belga News Agency and like-minded agencies in Northern and Western Europe built their own network for exchanging facts while news provision elsewhere came under political control.
That institutional instinct has aged well. Editorial independence today depends on three pillars: ownership that cannot be captured, an economic model that does not depend on a single gateway, and technical infrastructure a company actually controls. NZZ offers a working example of the first two, and this week’s other signals show how contested the third has become.
The dispatch of this week is build around this visit to Zürich.
Large platforms are writing the terms on which trusted information is used, priced and distributed, through licensing programmes, crawler policy and assistant interfaces. Publishers, broadcasters and press agencies are answering with standards work, litigation, collective negotiation and internal capability building.
Courts treated the same question from three continents, with the Court of Justice clarifying copyright in social media posts, a German ruling on advertising liability, and new judicial guidance from China’s highest court. Regulators and lobbyists moved in Ottawa, Madrid and Tokyo. Inside the newsrooms, the practical work went on: transparency guidelines in the Netherlands, AI literacy for more than four thousand employees at Mediahuis, and verification tools built in Nigeria and Dubai.
Nine signals this week, in four chapters, running from ownership to protocol to daily editorial practice.
I. On independence as infrastructure
Two signals on what makes a news organisation difficult to capture.
29.1 — NZZ builds its digital future on ownership and reader revenue
NZZ reported growth in its core business and an improved operating result in a statement from the group’s corporate site. I’m mostly interested in the structure behind the numbers. According to the presentation of CEO Felix Graf during our visit, NZZ has roughly 2,700 shareholders, all of them individuals, with no single holder controlling more than one percent of the voting rights, and the title itself dates back to 1780, which places it highly ranked among the top 10 international quality titles editors still study.
Three operational facts explain the strategy. Around 75 percent of traffic arrives directly, which reduces exposure to declining search referrals, social distribution and substitution by AI assistants. Revenue concentrates on paying readers rather than volume, with digital subscriptions as the engine and selective diversification into adjacent media businesses and holdings as the hedge. Technology serves the editorial product through archive retrieval, style consistency and workflow efficiency, under human control, with none of the positioning of an AI laboratory.
This describes a conservative and sophisticated digital strategy: monetise scarce quality, spread the portfolio with discipline, keep technology instrumental. The value of this premium model lies in the demonstration that direct relationships with readers can be built and held, and that dispersed ownership makes editorial autonomy structural rather than aspirational.
Reflections How does NZZ’s resilience coming from its shareholder structure influences the markets willingness to pay for scarce quality? What would a mass-market publisher have to change to reach 75 percent direct traffic?
29.2 — Group 39 marks eighty years of shared independence
Group 39 now brings together eleven independent European news agencies: Belga, ANP, APA, Keystone-SDA, ANSA, dpa, PA Media, Ritzau, STT, TT and NTB. The alliance marked eighty years of cooperation and an expansion of its circle of members, a cooperation founded on the conviction that reliable factual exchange requires a network owned by its participants. The annual meeting in Zurich, hosted by Keystone-SDA, kept that agenda intact while adding the machine layer to it.
Ownership remains the underlying foundation of that model, and it is worth looking at the national picture. Hugues Rey’s survey of Belgian media from 1839 to 2026 describes almost two centuries of continuous consolidation with capital that has remained largely national. Concentration of that kind delivers scale and survival capacity; it also concentrates the decisions about which titles, languages and regions receive investment.
Press agencies occupy a specific position in this structure. They are critical information infrastructure for publishers, broadcasters, institutions and increasingly for machines, which gives their governance a public dimension that their balance sheets alone do not capture. An alliance founded in 1939 to protect factual exchange from political capture now faces a comparable task with commercial gatekeeping, and the instrument is the same: shared standards, shared costs, independent ownership.
Reflections Which shared capability would justify the next decade of an agency alliance, beyond the exchange of content? Where should the line fall between cooperation on infrastructure and competition on products?
II. On who sets the rules and the price
Two signals on licensing programmes, crawler policy and the emerging market structure.
29.3 — Google puts a value on publisher content and keeps the scoring
Google has begun rolling out an AI licensing programme for publishers built on the principle of paying for value, with compensation tied to content that contributes significantly to its systems. Two questions decide what the programme is worth to a publisher: what counts as significant contribution, and who calculates it. Participants at the Digiday Publishing Summit discussed the same dependency under the heading of Google Zero, while OpenAI signed a Canadian agreement with Villager Media covering a limited set of titles.
Individual agreements of that type reward the publishers who arrive early and leave the rest of the market waiting. No technology company can conclude a bilateral partnership with every news organisation in every language, which makes the bilateral approach an opportunistic market entry with a hard ceiling. At ISOJ, the discussion turned to the consequence: publishers now face a new intermediary layer as AI reshapes the business, an interface that sits between the newsroom and its audience and sets the terms of access.
Media executives have started to say this in public, with senior leaders sounding the alarm about AI’s effect on their businesses. The economics are conventional: a scarce input acquires a price when the buyer needs it and the seller can withhold it. Pricing power therefore depends on collective repertoire, auditable measurement and the credible option of saying no, and every licensing programme designed by a buyer should be read as a proposal about who keeps the ledger.
Reflections What auditable evidence would a publisher need before accepting a value-based payment formula designed by the buyer? How does a market reach a workable price when the largest buyers negotiate one publisher at a time?
29.4 — Cloudflare and SPUR turn access into a technical question
Cloudflare published new work on accountability for mixed-use AI crawlers and announced a mechanism intended to end the tradeoff between being indexed for search and being used for AI training. In the past, publishers were tempted to accept training as the price of discoverability. Separating the two purposes at the level of the request gives rights holders a choice they could previously express only by accepting everything or blocking everything.
The model of Cloudfare shows one strategic problem. Websites are built for human readers; machines are better served through an API, where a publisher controls identification, volume, purpose and price by design. Crawling remains a permission problem solved after the fact, while an interface is a commercial arrangement agreed in advance, and the second is easier to audit and to invoice. That is where standards work becomes decisive, and SPUR’s technical lead Alex Springer has described the coalition’s effort to build an economic model for publishers through machine-readable rights, identification, metadata and telemetry.
A shared technical layer of that kind has one condition attached: it must stay separated from pricing and commercial terms, so that competitors can cooperate on protocol while competing on value. Infrastructure built jointly and priced individually is how other regulated markets solved the same coordination problem. Rights that machines can read are worth more than rights that require a lawyer to interpret, because only the first can be enforced at the speed at which content is consumed.
Reflections If machine access moves to APIs by default, why would a publisher still allow crawling? Which functions belong in a shared industry standard, and which must remain each company’s own commercial decision?
III. On the law finding its equilibrium
Two signals on courts, filings and legislative preparation across several jurisdictions.
29.5 — Court filings and executive statements reshape the copyright debate
EANA reported that unsealed court filings indicate the collection of news content for AI training was deliberate and went uncompensated, a finding drawn from documents produced in litigation rather than from press statements. Reuters separately reported that quotes from OpenAI and Microsoft executives may weaken their defence in copyright cases brought by news organisations. Evidence of intent carries weight in any assessment of good faith, and internal communications have become part of the factual record.
In Europe the framework is droit d’auteur and the CDSM Directive, which means the relevant question concerns authorisation and remuneration for specific acts of reproduction and extraction. The Court of Justice added a further element by clarifying the conditions for copyright protection of social media posts, a ruling that matters for every organisation using social content in reporting and for every system trained on it. Originality thresholds decide what may be reused at scale.
Reflections How should a negotiator value machine rights and based on what permitted use? What evidentiary standard should apply and why should publishers accept intent, rather than volume, as the attribution key?
29.6 — Three jurisdictions set different limits on AI and platform liability
China’s highest court issued new judicial opinions on AI-related disputes, drawing explicit red lines, giving lower courts a framework where legislation remains incomplete. In Germany, a court held Meta liable for fraudulent advertisements on Instagram and Facebook, extending platform responsibility for commercial content carried on its systems. Japan, meanwhile, secured budget funding to examine a system modelled on South Korea’s AI-supported response to online copyright infringement.
Canada shows the same movement at the level of policy preparation, where media organisations raised AI and the protection of journalism in meetings with Ottawa. Four jurisdictions, four instruments: judicial guidance, civil liability, enforcement technology and legislative lobbying. Each one changes the compliance surface for companies operating internationally, and each one creates a precedent that counterparts will cite in the next negotiation.
Reflections Which of these instruments will other jurisdictions copy first, and why that one? How should a European rights holder enforce a licence when conditions differ in other markets?
IV. On the newsroom as the product
Three signals on strategy, trust and internal capability.
29.7 — Publishers redefine what a newsroom sells
Axel Springer’s chief executive Mathias Döpfner told dpa that he intends to develop the company into a creator platform, in an interview published by HORIZONT. The Reuters Institute described the same movement across the industry as a major pivot by newspaper groups towards news creators, and the underlying logic is straightforward: individual voices travel through algorithmic distribution more easily than institutional brands, and attention has become the scarce input.
Two other decisions this week show the cost side of that strategy. Reach is cutting 220 editorial jobs, reported by The Next Web and by The Guardian in the context of AI summaries and search referrals, while News Corp has started generating local news stories with AI assistance. Reuters took the opposite route to differentiation by extending its Pure News, Straight from the Source brand campaign with two new videos, and the Reuters Institute gathered five takeaways on how publishers are rethinking audience from its Future of Media Technology conference.
Institutions deliver something a creator cannot: verification capacity, legal protection, correction procedures and continuity when a story becomes dangerous to publish. The strength of a quality newsroom lies in that collective apparatus, and it survives the departure of any individual byline. A company that markets personalities while reducing editorial capacity sells the signal and reduces the newsroom that produces it, which looks more like a budget decision disguised as a distribution strategy.
Reflections What does a publisher own in a creator model when the audience follows the person? Which editorial functions must remain institutional for a brand to keep charging a premium?
29.8 — Verification becomes an operational discipline in more markets
A report documented how actors in the Democratic Republic of Congo used Claude to produce fabricated news articles, and The New York Times examined how AI chatbots are shaping election misinformation. Institutions responded with instruments of their own. South Africa’s electoral commission launched a code of conduct to limit disinformation ahead of elections, Dubai released Saraab, an open source model for detecting deepfake videos, and Nigeria’s Centre for Journalism Innovation and Development built Dubawa AI to help journalists and the public verify claims across Africa.
The security dimension belongs to the same picture. The United Kingdom, the United States and the Netherlands issued a joint advisory on Iranian spyware, a reminder that journalists and their sources are surveillance targets and that newsroom security is part of editorial capacity. Meanwhile 404 Media argued that AI agents are already reshaping the internet, which changes the arithmetic of verification at the level of infrastructure.
Capability is spreading outward from the traditional centres of media technology, and Lagos, Dubai and Johannesburg now produce tools and rules that news organisations can use. Verification has acquired the character of a permanent operating function with staffing, tooling and budget, comparable to information security a decade ago. Trust is produced at a cost, by people and systems that have to be paid for, and it stays scarce for exactly that reason.
Reflections Which verification capacity should a news organisation build internally, and which should it source from shared infrastructure? How do editorial teams document verification in a way an audience can actually check?
29.9 — Mediahuis measures AI adoption by the drop in confidence
Mediahuis made AI literacy and adoption mandatory for more than 4,000 employees across Belgium, the Netherlands, Luxembourg, Germany and Ireland, and the group’s head of AI strategy Ana Jakimovska has published what those 4,000 people taught the company about AI adoption. The programme ran in three phases with two full-time staff and five part-time colleagues, reached 78 percent participation, and included the Board in the same training. Self-reported confidence stood at 3.85 out of 5 after the introductory phase, then fell to around 3 once employees applied the tools to their own workflows, and the third phase has produced 177 use cases now moving through prioritisation, build and measurement. It fits a wider repositioning that Jakimovska earlier described as a move from experimentation to adoption.
That decline in confidence carries the strategic lesson. Employees moved from asking whether AI can perform a task to asking which part of the process AI should handle, which part requires human judgement, and how the two connect; Jakimovska concludes that rising confidence on its own measures very little, and that professional hesitation about quality, accountability and expertise often represents judgement at work. For newsroom leaders this reframes the investment case. Licences and pilots are the cheap part of the programme, and the expensive part is redesigning workflows around verification, with the cost falling on the same editorial capacity that produces the value.
The Mediahuis evidence was published in the same week that Dutch media organisations agreed joint guidelines for AI transparency and that comparative research on citizens’ attitudes toward AI in journalism and authenticity across six countries found broad agreement on the human element as the source of authenticity, with more optimism in South Africa and Brazil and more scepticism in Western markets. These three pieces of one governance question can be seen togheter: what the newsroom does, what it discloses, and what the audience accepts. Internal literacy makes disclosure credible, because an organisation can only account for a process it understands, and the number that matters is the one an editor can defend.
Reflections Which measure of AI adoption would survive scrutiny from an editor, a works council and an auditor at the same time? When trained staff conclude that a use case adds little value, how does a management team make room for that verdict?
Conclusion
The weeks dispatch is about who declares independence and who organises it. Declarations are inexpensive and appear in every mission statement. Organisation shows up in a shareholder register, in the share of traffic that arrives directly, in a standard that machines can read, in a training budget for four thousand people, and in the willingness to litigate while negotiating.
NZZ and Group 39 supply the constructive counter-example to the broader anxiety in the sector. Dispersed ownership, revenue from readers who choose to pay, cooperation on infrastructure with competition on products, and technology used as an editorial instrument. None of that requires a company to outspend the largest technology firms; it requires a company to control the parts of its own value chain that determine price.
The pricing question returns in every chapter. A licensing programme designed by the buyer, a crawler policy set by an intermediary, an assistant interface positioned between a newsroom and its audience: each shifts the power to define what a contribution is worth. Courts are establishing the boundaries of permitted use, and negotiations inside those boundaries will establish the price.
Trusted information has the economics of any scarce, costly product. It is produced by trained people working inside institutions that verify, correct and stand behind the result, and it stops being produced when nobody pays for the apparatus. The instructive detail this week came from Mediahuis, where confidence in AI fell as understanding rose. That is what competence looks like in a new domain, and any board that reads falling confidence as failure will fund the wrong programme.
An independent news organisation resembles a building with three load-bearing walls: ownership, revenue and protocol. Remove one and the structure still stands for a while, which is precisely what makes the decision easy to postpone.
Patrick Lacroix writes Cyber Territories in a personal capacity. AI tools are used for research and drafting; the author retains full editorial responsibility.