Why Aftonbladet's Readers Reject AI Articles - But Embrace AI Chatbots

When Aftonbladet experimented with AI-generated articles, readers pushed back with an argument the 195-year-old tabloid hadn't anticipated: "We can read AI-summarized news on Google AI overviews. We come to you because we don't want that.”

Aftonbladet has spent more than two years building AI products - and learning where the boundaries are. Martin Schori, Aftonbladet's Deputy Publisher and Director of Editorial AI & Innovation, shared this and other refreshingly honest feedback with me on a Zoom call in early February. 

Aftonbladet, founded in Stockholm in 1830, is Scandinavia's largest newspaper and Sweden's most-read online news brand, reaching more than 3.5 million daily readers and viewers on mobile devices and desktop, with online video, narrated stories, podcasts and print newspapers. The tabloid brand is majority-owned by Schibsted Media, a Norwegian company headquartered in Oslo.

Aftonbladet's chatbot Hej Aftonbladet, handles around 50,000 questions daily. The chatbot draws on three years of Aftonbladet journalism via RAG architecture, based on GPT-4o. After nine months and 7 million questions, Schori reports that the chatbot has had virtually no hallucinations. Aftonbladet’s election chatbots are specialized version of the general chatbot and performed even better - 600,000 questions on US election day, with login conversion hitting 40% versus the typical 5% for Scandinavian media campaigns, Schori wrote in his INMA column.

Text-to-speech, summaries and personalized catch-up features are all accepted. "People just see that as a part of our regular offering nowadays," Schori told me.

But AI-touched articles trigger resistance, even with transparency labels. "Whenever we create it to increase the user experience or our distribution or new ways of consuming our journalism, that's fine. But as soon as it appears in articles, if there's a footnote saying this article has been at least partly generated by AI, then we get some feedback."

Schori acknowledged that younger audiences may hold legacy media to higher standards precisely because they encounter AI content everywhere else. ”They see so much AI all over the place, so they don't really find it valuable.” 

Aftonbladet's dedicated eight-person AI Hub, launched in October 2023, has been reintegrated into the regular organization. Six to seven editorial staff now work on AI, with technical development centralized at Schibsted.

Aftonbladet is adopting a feature from Norwegian sister publication VG - an AI-generated module showing what has happened since a user's last visit. The feature targets readers who have a relationship with the brand but aren't daily visitors.

"The time of really fast experimentation is probably over," Schori said. "Whenever we are thinking about doing something now, it has to align with either revenue or efficiency. We do still have a small chunk of innovation, but the vast majority of new products will have to align with that."

Interestingly, Hej Aftonbladet launched without clear key performance indicators (KPIs) - something Schori now calls "a key learning." A subsequent Schibsted reorganization left the product without dedicated development resources for months. Asked about conversion from chatbot logins to paying subscribers, Schori said they don't have that data. The team has also dabbled in commercial applications, including twice testing a “shopping buddy” during Black Week (Sweden’s version of Black Friday) that surfaced affiliate content. “We had some revenue coming in from that”, Schori said. But he was unable to provide metrics, making return on investment (ROI) difficult to assess.

Aftonbladet has tools that produce "perfect rewrites" from press releases. They're not using them to a great extent. "I'm not sure that there is a demand from our audience to read tons of articles created by AI," Schori said. "Maybe it's time for more ‘human’  - trying to build deeper relations and having our journalists taking more place. There's a reaction to all the AI slop on the internet. Maybe our value should be the opposite."

Editorial AI tools have underperformed expectations overall. In a September 2024 interview on the Newsroom Robots podcast, Schori called the promise that AI would automate boring tasks and free journalists for investigative work "a bit exaggerated." That assessment hasn't changed. "If you were convinced that would be the reality, then I would say it's still disappointing," he told me. The exceptions: transcription and processing large datasets, such as newsrooms using AI to surface stories from the Epstein files.

The larger question Schori is wrestling with: In a world filling with AI-generated content, is human-created journalism the differentiation strategy?

The instinct to answer "obviously yes" deserves scrutiny. Plenty of journalism -commodity news, earnings reports, event recaps - was never differentiated by human craft. Readers tolerated it but they didn't seek it out. AI can handle that layer and audiences may not care.

But Aftonbladet's experience suggests something specific: Readers who choose a legacy news brand are actively seeking an alternative to the AI-saturated information environment. They're not paying for content that exists elsewhere. They're paying for judgment, for curation, for a point of view - things that currently require humans.

Schori frames it as “a reaction to AI slop." That creates a chance for market positioning. If AI-generated content becomes the baseline - free, abundant, undifferentiated - then human-created journalism becomes the premium tier. The business case isn't about resisting technology as a principle but about understanding what you're actually selling.

Aftonbladet is still investing heavily in AI. The difference is where - delivery, distribution, and infrastructure but not authorship. That distinction may be the strategy worth copying.

Schibsted is one of the most sophisticated media companies in Europe. If they're still figuring out KPIs and discovering that audiences reject AI content while embracing AI delivery, most newsrooms will face the same learning curve.