Across more than 70 hours of programming over five days at the INMA World Congress of News Media in Berlin last week, what emerged most clearly wasn’t only what we learned — it was what we got wrong. 

Quietly, repeatedly, and from speakers who had no coordinated reason to say the same thing, the data and the operators kept invalidating strategies that significant portions of our industry are still executing. 

This is a catalogue of those reversals: the conventional wisdom we built our last decade on, and the evidence from INMA Berlin that it no longer holds.

We got the trade-off between reach and loyalty wrong

For years the strategic frame has been “we are in the business of loyalty, not reach” — a comforting line that justified shrinking ambition. 

Greg Piechota, drawing on data from hundreds of news Web sites and Reuters Institute survey work, showed an almost perfect correlation between brand size and user loyalty. 

Larger brands have more loyal users, not just more users. Smaller publishers are doubly penalised: fewer readers overall, and a less-loyal subset of those readers. 

Byron Sharp’s marketing science — mental availability and physical availability as the engines of growth — applies to news as it does to consumer goods.

The implication is that cutting reach to focus on loyalty is not a strategic choice. It is, on the data, a path to a smaller and less loyal audience. 

We got the relationship between subscriptions and advertising wrong

The “pay us and we'll turn off the horrible ads” pitch has become a default in subscription marketing.  

The data gathered across the week says it is actively destructive. The New York Times found its most-engaged users are most receptive to advertising. A publisher in Latin America reported 1% of users who are subscribers generating 72% of combined subscription and ad revenue.

Another publisher cited by Piechota in Europe: 1% generating 81%. Bonnier’s data shows subscribers produce four times the ad revenue and four times the pageviews of non-subscribers. Logged-in users command a 100% price uplift on advertising.

Super-users are simultaneously our most valuable subscribers and our most valuable ad audience. We have been turning off the highest-yielding inventory for our highest-paying customers, and calling it a benefit.

We got the young-audience question wrong

The dominant news industry narrative — that younger audiences only want short-form, won’t pay, and are lost to platforms — is empirically wrong on every count. 

FT Strategies’ study of 5,000 respondents across five countries found 68% of younger audiences will commit to in-depth stories or even two-hour documentaries — after a topic has captured them.

They want a different entry point, not a different depth. 

They will pay: Podme told INMA recently they retained 200,000 subscribers for its podcast platform, 50% of them under 35. Meanwhile, 64% of under-25s consume news daily. 

What they reject is the legacy product, not the substance. Treating them as a niche to be eventually converted to the old model is the inverse of what the data says they are: a preview of how everyone is starting to behave. 

We got pricing wrong

Die Zeit charged digital subscribers €25-€26 (matching print) when competitors offered €10-€15 — and won. They reinvested into editorial expansion from 100 to 550 editors. NZZ doubled prices in Germany and grew faster, with Pro subscribers churning at half the rate of standard subscribers.

The Pioneer is selling basic subscriptions at €28 per month, lifetime subscriptions at €5,000, study trips at €25,000 — and a circle of 40 supporters pays €10,000 per year for close access to the editorial team. When The Pioneer voluntarily raised prices from €25 to €28, half of users opted to pay more or upgrade.

Piechota's provocation cut to the bone: If our business is super-users, can we imagine a product worth €10,000? The ceiling on subscription pricing turns out to be self-imposed. The willingness-to-pay for distinctive, mission-led journalism is meaningfully above what we have been charging for it. 

We got the production cycle wrong

The assumption that ad creative takes weeks, that product development takes quarters, that newsroom workflows take half a day per piece — all of this collapsed in Berlin. 

Russmedia built and shipped a live real-time energy-consumption advertisement in 14 hours, with sales staff doing the work, not engineers. Hindustan Times generates 97.4% of its code by agents and ships roughly 1.7 billion lines a month, having compressed its software development lifecycle from three or four months to weeks. Funke rebuilt its news app in roughly four months.

Bauer’s agentic newsletter pipeline reduced production from two or three hours to about five minutes. Holtzbrinck took its San Francisco AI Hub from idea to launch in six months and through 30 explorer projects in nine months. Tamedia’s hackathon-built infographic tool: four hours for three people to prototype, six developer hours to harden.

The metabolic rate of the news industry has shifted, and the gap between operators and observers is now measured in months, not years.

We got the article wrong

YLE made it explicit that we are still operating “article factories” while consumption has gone liquid. Bauer renamed itself a “liquid publisher.” Holtzbrinck hosted a Liquid Content Summit. Die Welt’s hourly audio briefing decomposes the news day into five-minute machine-assembled segments. Ippen routes content adaptively across more than 30 channels.

The article-as-canonical-unit assumption — that a story is something that gets written once and distributed everywhere — was nowhere defended in Berlin and was actively dismantled across at least eight sessions.

What flows is not the article. What flows are the facts, sources, and judgments inside it, rendered into the form a particular audience needs in a particular moment.

We got bilateral deal-making wrong

For three years the news industry's response to AI scraping has been to negotiate individually. Berlin closed that chapter.  

Sky News, BBC, the Guardian, the Telegraph, and the Financial Times co-founded SPUR. RSL and the RSL Collective engaged more than 1,500 publishers with CDN-enforced licensing standards from Cloudflare, Akamai, and Fastly.

COPE in Austria built a publisher first-party data alliance against an 86% platform share. DRIVE assembled 30 DACH brands sharing 150 metrics. Schibsted’s collective Nordic posture produced an OpenAI deal that pays journalists monthly.

The math is now obvious. Individual deals weaken collective leverage. Platforms use the existence of “a few deals” to forestall regulation. We have, for years, been negotiating against ourselves while the value was being set bilaterally at a discount. The closing-ceremony commitment to elevate INMA’s policy voice reads as the industry’s own admission of this error.

We got brand safety wrong by accepting it

The US$2.8 billion annual loss from keyword-based brand-safety tools that classify legitimate news as "unsafe" while classifying Meta as "safe" is not a new problem. 

It has been documented for years. A story about tennis blocked for “smash.” A vineyard story blocked for “alcohol.” Vendors paid each time they block, creating a perverse incentive to over-block. Better, context-aware tools exist — Mobius, Illuma, others.  

The shock from the Q&A after Gabe Dorosz's session was that the fix is collective and overnight: a critical mass of publishers demanding their agencies adopt context-aware tools could reclaim the US$2.8 billion almost immediately.

We have been losing this money not because the technology was unavailable but because we hadn’t acted in concert. That is a strategic error, not a market condition.

We got the comments section wrong

The Toronto Star told INMA recently they saw a 13-fold engagement uplift simply by letting editors answer comments below the line. The figure surfaced twice in Berlin almost in passing.

This is the cheapest community-building intervention available to any newsroom, and most newsrooms have walked away from it — outsourced moderation, disabled comments entirely, or treated the section as a liability to be managed rather than a relationship to be built. 

Seznam.cz scaled to 70,000-100,000 comments per day with AI moderation handling 85%, and then built a social network from it that grew, in 10 months, to twice the size of X in the Czech Republic.

Community wasn't the consolation prize for losing platforms. It was the asset we abandoned to platforms.

We got the ad unit wrong

Nobody in Berlin defended the legacy display model. Nobody defended programmatic open-exchange as a growth strategy. Nobody defended chasing social referral traffic for ad inventory. News Corp Australia explicitly named that pursuit “empty reach.” 

The static display rectangle, priced per thousand impressions and measured by clicks, was treated across the entire Advertising Seminar as a depreciating asset to be replaced — by the New York Times’ FlexSuite, by Bonnier’s logged-in data products, by COPE’s first-party alliance, by Russmedia’s outcome-priced utility apps, by Bonnier’s in-store retail media partnership with Visual Art, by News Corp Australia’s vertical-video pivot, by IAB’s CDN-edge stack. 

We spent 15 years optimising a unit that was structurally broken. The fix is not optimisation. It is replacement. 

We got the ownership question wrong

The Atlantic has 1.5 million subscribers and is profitable. The Guardian has reader contributions from every country on earth, including the Vatican, on a no-paywall model that just turned 10 years old.  

Both speakers — Anne Applebaum and Katharine Viner — argued that ownership structure is now a competitive moat. 

Laurene Powell Jobs’ Emerson Collective doesn't depend on government contracts, which immunizes The Atlantic from regulatory leverage. The Scott Trust exists in perpetuity for the express purpose of editorial independence. Applebaum’s account of Hungary, Poland, Russia, and the United States made the same point operationally: the modern autocrat’s playbook works through ownership leverage, advertising leverage, and regulatory leverage.  

Ownership that is structurally insulated from those vectors is a defensive asset. 

For years we have treated ownership structure as an accident of history. It is, in 2026, a strategic variable. It can be designed for.

We got our place in the AI economy wrong

Jonathan Roberts’ framing was uncomfortable for the room because it inverted the victim narrative. AI needs three pillars — chips, power, and information — and information is the under-funded leg.  

Publishers are not adversaries to AI. We are essential drivers of the AI economy. 

The current paradigm is unsustainable on its own terms. AI companies have assumed the search-traffic economy would persist while they free-ride on it, scraping content as input without pass-through economics.

But the incentive to write for search has already collapsed. Without an investment thesis for new knowledge, sources disappear. AI systems are only as good as their inputs, and inputs are degrading. Yet share prices presume AI will get better, not worse, even as the information feeding it gets worse. That contradiction is the market opening.

Publishers are best positioned to build the licensed, structured, rights-cleared content marketplace that enterprises and AI companies actually need. There are now roughly 30 venture-backed companies in the "content as a service" market, and capital is flowing.

We've been arguing for protection. We should have been building the market. The capital is there. The demand is there. The competence is ours. The error is that we framed ourselves as the harmed party rather than the indispensable one.

We got the timing of the agentic threat wrong

Jimmy Wales reported an 8% year-over-year decline in human traffic to Wikipedia, with a corresponding drop in new contributors. 

If that’s happening to the most-trafficked information source on earth — one that AI models depend on for training — it’s the canary that says agentic AI isn’t a future threat to publisher funnels. It’s an active one, measurable now. 

The reader-to-participant pipeline that built Wikipedia is hollowing out, and the mechanism that's hollowing it out is the same one hollowing out our subscription funnels.

We have been treating agentic AI as a 2027 problem. The data Wales presented says we missed it as a 2025 problem.

We got platform-immunity geopolitics wrong

Section 230 is a 1996 American law. Sovereign states elsewhere have no obligation to honour it. The brand-safety Q&A surfaced this almost as an aside. 

The entire global platform-immunity regime — under which Meta, Google, and Amazon evade liability for the scams, deepfakes, and disinformation distributed on their networks — is built on voluntary global deference to American law. 

Sweden has already removed Meta from its IAB board over scam ads and triggered a one-hour takedown legislative rule. The EU’s DSA compels platform transparency. Australia is strengthening copyright. The United States did not grant an AI training carve-out.

For most of the platform era we accepted, as a pre-condition, a liability regime we had no obligation to accept. The lever exists. It has barely been pulled.

We got culture wrong

Transformation is not only technology.

We kept treating transformation as a technology problem. 

Stephanie von Unruh’s point was simpler and harder. The advantage is culture — short decision cycles, empowered teams, and the ability to execute before the market moves again. 

In INMA’s pre-Congress survey of media companies, change management was the No. 2 concern of CEOs after financial performance.

Publishers have to fundamentally rethink how their organisations are wired. Tagesspiegel was the most popular stop on one of the study tours because Editor-in-Chief Christian Tretbar is radically reorganising, and he’s doing it with conviction and clarity. He’s folding product into the newsroom – which makes so much sense.

Turf wars between editorial and product (both claiming they make the product) are endemic in our industry and need to go. Secondly, Christian is organising around reader needs, not product categories – another feature of cutting-edge digital organisations. 

We got brand wrong

Brand is not marketing. It is the operating system.

For years, we treated brand as marketing: awareness, campaigns, logos, slogans.

Berlin argued the opposite. Brand is now the operating system of the news company. In a world where AI decouples discovery from destination, brand is the signal that survives when the funnel breaks. 

The pattern beneath the pattern

What unites these errors is a single structural mistake. We treated decisions as fixed market conditions:

• The trade-off between reach and loyalty was never real; it was an assumption. 

• The price ceiling on subscriptions was never real; it was a habit. 

• The platform-immunity regime was never inevitable; it was a choice. 

• The ad unit was never the natural form of advertising; it was a default we stopped questioning. 

• Bilateral deal-making was never the only way; it was the easier way. 

• The article was never the canonical form; it was the legacy form. 

• Brand safety as it stood was never imposed on us; it was a vendor relationship we kept renewing.

INMA Berlin’s deepest finding wasn’t that AI is changing the industry. It was that significant portions of what we have called “industry conditions” for the last decade were assumptions we made and then forgot we made. 

Every reversal in this catalogue is the same shape: an operator somewhere stopped accepting the default, ran the experiment, and found the default was wrong.

The opportunity is the same shape, too. Whatever else we are still treating as a fixed condition — about pricing, about format, about ownership, about negotiating posture, about what a young person will or won’t pay for, about what a community will or won't build, about what an advertiser is actually willing to pay for — is probably the next reversal.

The question to leave INMA Berlin with is not what we should do next. It is which of our remaining assumptions is the next one to die.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .