A Vogue editor told Marcel Semmler, Bauer Media’s global product chief, that a typical feature runs to roughly 20,000 words of reporting — interviews, notes, background — to produce a 4,000-word published article. What happens to the other 16,000 words?

As AI is increasingly intermediating search results, it has already scraped those 4,000 words, summarised, and served back to readers who rarely clicked through. 

The remaining 16,000 words — the journalist’s raw materials, verification evidence, her domain knowledge, and judgment calls — are still standing.

“That knowledge,” Semmler told the European Publishing Congress in Vienna in late June, “AI has not taken from us yet.”

His observation matters for news marketers. If AI can reconstruct your published content for free, whatever subscription acquisition funnel was built on it evaporates. 

In Semmler’s opinion, there is a treasure trove in the knowledge hidden behind the published articles: “We have to treat this knowledge as our actual asset … . The product is not the article. It is the knowledge, and the system this knowledge lives in.”

In his vision, AI technology allows recombining all that knowledge into different containers — articles, podcasts, or video stories — whatever format the audience or individual consumers need and the publisher can make money with. 

At INMA conferences, this ambition has been billed “liquid content.”

Content economics in the AI era

Back in 2024, Harvard Business School’s Thales Teixeira presented his take on the future of content at the CEO Roundtable at Vail as a tree with different fruits.

He advised CEOs: “Match the cost of production to the market value of content but keep supplying all thanks to AI.”

• Cherries: Exclusive, relevant, high-demand journalism will continue to get made by humans. 

• Bananas: Commodity, low-relevance, low-demand content needs to be made by machines. 

The latter is still necessary to round out the news experience but no longer where you capture value.

This year in May at the Nordic AI in Media Summit in Copenhagen, Harvard Kennedy School of Government’s Shuwei Fang added some nuance to the idea.

In her opinion, whether humans or AI produces content is not the differentiator. AI collapses the production cost and the article stops holding value on its own, so businesses have to find value somewhere else. 

This may lead to the market resembling the barbell — split between the extremes:

• One end of the market will be driven by luxury economics, where the moat sits not in the content but in the intangibles: trusted brand, human taste, customer captivity. 

• The other is commodity economics, where the moat sits not in the water but in the pipes: infrastructure so integrated it’s hard to rip out. 

She gave an example of Bloomberg, one of the most profitable news businesses on earth, as a commodity feed of data and news wrapped in an irreplaceable terminal.

The case for signature journalism

Major news publishers in the INMA network seem to be placing multiple bets, often on both ends of the Fang’s barbell, including on the luxury end.

Last year, at an INMA Newsroom Transformation Initiative Webinar, GerBen van ‘t Hek and Cormac Bourke laid out their pan-European company Mediahuis’ focus on “signature journalism.”

It is defined as authentic, brand-matched, high-value, made by humans for humans, even if all journalists get access to the latest AI tools assisting them at work. 

“We have to become the Taylor Swift of news,” van ‘t Hek said, explaining the ambition is not becoming the Spotify platform but the irreplaceable artist on it. 

Hence the metric is “subscriber attention time,” not anonymous pageviews. For example, across Mediahuis’ Irish titles, the strongest-performing category in the signature mix is human interest. “People love stories about people,” Bourke said. 

Having just crossed 100,000 digital subscribers on the island and record aggregated attention time, the signature content was credited as part of this success.

Speaking in Copenhagen, Gard Steiro of Norway’s Schibsted described how his VG newsroom tracks the share of “super content” that is “impossible to copy by AI.”

That content is defined by three tests: 

• Did a reporter physically leave the office to get the story?

• Is the report written from the location where the story is actually happening? 

• And does it contain exclusive information that isn’t publicly available elsewhere?

To double down on shoe-leather reporting, Steiro said he needs to free up capacity by automating other workflows, for example, “liquifying” news articles into podcasts and videos with AI.

The barriers for AI transformation

Executives of all three companies see culture and organisation as the main barriers and levers for faster adoption of their strategies — not the technology. 

“The product design always represents the structure of the organisation,” Bauer’s Marcel Semmler said, referring to Conway’s law. Computer scientist Melvin Conway observed in 1968 that fragmented, siloed teams create clunky, disconnected products because of lack of communication and therefore insufficient coordination.

Semmler’s fix at Bauer is through clear definition of roles and ownership. Editorial holds the coverage agenda, the standards, and journalistic quality. Product holds the user experience and the journeys people take. Commercial holds pricing and monetisation. None of the three owns the brand solely.

The most radical and riskiest ideas at Schibsted are not tested by the VG’s main newsroom but as parallel cross-functional projects — in Gard Steiro’s words: “speedboats” ahead of “the tanker.” 

An example he gave was VGX, which is a new mobile app aimed at young audiences. AI agents cluster content from all Schibsted sources or wires, reassemble into snackable formats, and publish. 

Human oversight is limited to one editor working alongside the engineering team, tweaking the product’s master prompts but not editing content itself.

It’s only the beginning

None of these bets are proven, nor finished. 

Harvard’s Fang cautioned Copenhagen: “We’re at the beginning of a transformation arc, and we’re really in for a marathon, not a sprint … . We should be suspicious of the solutions that we come up with that require the least amount of change.”

Which brings us back to that Vogue editor’s math. Perhaps the 4,000 published words were never really the product, but neither, entirely, was the 20,000 behind them.

If knowledge is the product in the era of AI content, where is the moat? In the stock of what we already know or the flow of what we keep finding out? In the judgment what 4,000 words to publish? Or in who owns the relationship with the person asking?

Interested in signature journalism? Tune in to my interview with Dennik N’s co-founder Tomáš Bella in the series Subscription Masters next Wednesday, July 15.

Greg’s Readers First newsletter is a public face of a revenue and media subscriptions initiative by INMA, outlined here. INMA members can subscribe here.

Banner art: Adobe Stock By by EmmaStock.

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