This is the last of our Generative AI Initiative blogs. We are wrapping up this initiative now because AI has become ubiquitous, to the point where it no longer makes sense to focus separately on it. 

Instead, INMA will feature AI across the span of its other initiatives: The Newsroom Innovation Initiative, the Readers First Initiative, the Advertising Initiative, the Product and Tech Initiative, and the brand new Young Audiences Initiative. 

As for me, I’m going to be acknowledging my roots as a newsroom manager who loves a smart AI product and the culture change that comes with it. I’ll be running the Newsroom Innovation Initiative in 2026, which will focus on all aspects of editorial-related innovation, with a heavy emphasis on AI use cases and best practices. 

I will also look at innovation in other newsroom-related functions, such as the creation of new AI-focused roles and workflows, as well as the tricky question of change management in an era of AI tools. 

If you’d like to keep hearing about these topics, please sign up for the newsletter or join our Slack channel.

And now, a look at what we have learned over the last two years.

I took a look back across our discussions, Webinars, and shared learnings from the INMA community worldwide since we launched the GenAI Initiative in January 2024. Here’s a look at some key trends that have emerged:

The pivot from AI experimentation to widespread adoption

Two years ago, we were trying to get a sense of what this technology is capable of, running dozens of disconnected experiments. Now, I see new organisations focusing on end-to-end content solutions and hear conversations about strategic transformation. 

Many companies have tested the waters and some are now building systems that create capabilities for future solutions rather than just seeking quick wins. The goal has evolved from simple workflow efficiencies to genuinely driving scale and ROI.

We are grappling with the culture change required for AI to move from a side project to a central capability and finding out transformation calls for an ongoing commitment to change, not just technology deployment.

How’s the change going? 

We are often running into significant challenges in adoption. Keeping employees excited about using new tools has not proved to be easy. 

Different companies are managing it differently, some with remarkable success: Some are internally tracking employee AI use (like this leaderboard at Axel Springer’s Business Insider) and encouraging experimentation that genuinely relieves staff pressure (as at Thomson Reuters). 

The New York Times engaged over 100 reporters and editors early in the process, underscoring the need to “listen more than you talk” and ensure the journalist remains part of the process from the very start. Amedia managed to get half its newsroom on board by understanding how journalists work,  meeting them where they are, and focusing on practical utility over technological wizardry. 

AI-driven personalisation and agentic systems

It’s time to start waving goodbye to the 800-word inverted pyramid article. Multimodal appears to be the future. 

Content has turned out to be incredibly malleable in the GenAI era, with publishers acknowledging the consumer expects personalisation and the era of offering a “one-size-fits-all” product is over. 

We’re talking about evolution beyond simple recommendation engines. It’s about handing control to the user to select how they choose to consume content. 

This focus on liquid content that can be easily transformed, repackaged, and delivered across any medium is driving the next wave of products. We saw news products that let readers select their preferred format, be it a quick summary, an “easy language” version, an audio version, or one in a different language. 

Most news brands have also tried conversational interfaces, like Aftonbladet’s Valkompisen election chatbot. The bot answered over 150,000 questions and demonstrated GenAI can engage audiences on complex topics provided the bot is trained on reliable content and is instructed not to make things up. 

We saw AI-driven personalised audio playlists, such as at Schibsted’s Svenska Dagbladet and The Washington Post’s new personalised, customisable podcast, where the listener picks the topics and the voices they would like to hear about, at a length of their choosing, and will soon be able to ask questions as of the podcast as well. (This appears to be off to a bumpy start, though.)

There is also Time’s agentic system that lets the user ask for information to be synthesised and presented in different forms and languages as publishers attempt to create intuitive, fluid, multimodal experiences to prevent audiences from simply migrating to answer engines.

With the rise of agentic AI, news publishers are also considering a future where the user experience and monetisation are radically different. For example, if an agent summarises my news for me and a different agent buys my shoes for me, then it no longer makes sense for my favourite news site to place an ad for shoes above an article. 

What will this mean for our business models?

The AI-driven revenue opportunity — and concrete cost savings

GenAI has also opened the door to the possibility of new revenue-generating products, and a few news publishers have started walking through it. 

Nikkei’s Minutes by Nikkei, which uses GenAI to condense multiple articles into a digestible, lower-cost subscription product, demonstrates the power of repackaging and monetising content for a new market segment. 

AI has also been mobilised in advertising, with tools like The New York Times’ BrandMatch, which builds personalised ad-targeting segments by matching an advertiser’s brief with relevant articles and engaged audiences, and People Inc.’s D/Cipher, which uses consumer behaviour to define user intent and target advertising more effectively at scale.

And amid a lot of hand-waving about how much money AI could save us, we also saw AI create some real savings in costs, particularly in Germany. OVB Media saved half a million euros by building an AI editing tool. Nordwest-Zeitung and the Weser-Kurier implemented a customer service voice bot that paid for itself in six months, versus projections of a 12-month payback period. Note that part of the reason is these are specific, narrowly defined use cases rather than generic enterprise-wide rollouts.

The scramble for quality content 

AI models need vast amounts of high-quality data to train and infer, and the supply of publicly available content on the Internet is going to fall short. This has led to a “bustling data market,” with tech companies quietly paying for or stealing content, including transcribing YouTube videos. 

We are sitting on a goldmine, a continuously refreshed source of authoritative, verified content.

And yet, a new threat has emerged: the quiet dismantling of paywalls. ChatGPT, Perplexity, and other GenAI assistants have shown they can provide accurate summaries of paywalled content by synthesising publicly available fragments from social media snippets and other sources. This is an existential threat to the subscription models many publishers have spent years building. 

News publishers are exploring new commercial agreements while also doubling down on brand differentiation and counting on the “flight to quality.” As traffic from search becomes less reliable, the connection between a trusted brand and a loyal audience becomes our most defensible asset. 

The rising premium on human judgment 

Even as newsrooms across the world use AI for efficiency, one thing is clear: Human judgment remains vital and non-negotiable. 

We have seen several missteps that show how GenAI assistants often make things up or how news organisations rush to publish information without verifying it. Your brand matters more than ever. You need to keep a human in the loop to safeguard that. 

And so, the role of the journalist is being redefined as AI turns out to be a powerful assistant for tasks like transcription, translation, and archival research.

We see the emergence of the “full-stack journalist,” particularly in regions like India, where GenAI is being used to bridge language gaps and create efficiencies. In another example, the Minnesota Star Tribune used AI for quick translation during a breaking news event but ensured human experts reviewed the output, catching crucial discrepancies before publication. 

The industry is also automating tedious, time-consuming tasks like comment moderation and print laydown, which means making hard choices about workflow and product. But this also frees up journalists to focus on reporting original stories, writing with sensitivity, and developing relationships with sources — tasks the machine cannot do well and which our employees excel at. 

Indeed, it is the humans behind our brand that our audiences will truly connect with, as the emergence of creators shows. We need to highlight their voices, roles, and personalities, transparently and authentically bringing the news consumer behind the curtain where we create our news products — because people follow people.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

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