What a mind-blowing week in GenAI I have had! I caught a glimpse of the future because I had the privilege of hosting INMA’s Master Class on AI-driven personalisation over the past week. There were some radical ideas put out there, which truly expanded our minds based on feedback I have received from our participants.

Plus, AI-generated content is having a moment right now. Many news organisations jumped on GenAI a couple of years ago so it could help their journalists and serve their readers better by actually creating content in formats that readers prefer. We take a look at how some of those efforts are panning out.

Sonali

The news experience of the future

Our master class featured brilliant speakers from different corners of the world who are thinking about where we are headed with GenAI so we can use this technology to effectively tackle our challenges. 

One question they sought to answer: In a GenAI era, where content is malleable and shape-shifting, what does the future news experience look like?

Think about this for a moment. Content is no longer constrained by medium or format or language. You can watch a podcast (the biggest podcasting platform in the world right now happens to be YouTube) or read it. You can listen to Swedish news in Arabic. 

All of these are real examples that exist right now, thanks to GenAI. And in the future?

You might encounter a live blog on a Web page that recognises you have been reading about a particular topic for a while now, so the content it presents to you will adapt to ensure you are getting only new information, as suggested by Jason Sheppard, senior product manager of AI at The Telegraph.

Or when you go to a Web page, both the text and the visuals are optimised for you, “a multi-dimensional approach where the visual component and the article content become part of the model’s decision-making to deliver the best order of content,” as Sheppard said.

You could choose to instantly convert a text article, replete with quotes, on the federal government’s changes to tax rates into a calculator that lets you swiftly calculate how your tax liability will change this year, as suggested by Sannuta Raghu at Scroll.in.

You could use a slider to give you the gist of an article in 20 words or in 2,000 words. Or you can be presented with a summary of a few paragraphs of an article and choose to further expand any paragraph, further and further, to get more and more detail, kind of like a hierarchical tree view, as Google Labs creative technologist Kawandeep Virdee showed us.

Maybe you want to get more details on what an image is about. You could use a slider to do that.

Or when you are visiting a particular Web site, you could type questions into a search box and generate a Web page on the fly that answers your specific questions and lets you save your answers.

In other words, content will be dynamic and modular. You can mix and match it any way you want.

A few thought-provoking questions, then: What will the fundamental unit of news be? And as an editor, what level of personalisation are you willing to allow so AI can augment the content or the experience?

One thing is certain: Personalisation really is no longer optional. Many news brands are struggling to reach younger audiences — and these are people who absolutely expect personalisation, as Martin Schori, deputy publisher of Aftonbladet pointed out. He also made the point that news consumers say they don’t want AI, but if you provide them with a quality experience, they will happily engage with it.

The Telegraph’s Sheppard agreed: “I grew up with the dawn of the Internet and having to wait for a family member to finish a phone call before I could go online. The customers of the future will have had these chat experiences from an early age, and there will be a behaviour shift in consumption as people adapt to using AI tools more in their day to day.”

Where AI-generated content works well

There are plenty of examples of how AI-generated content is raising eyebrows right now. You’ve already read about the Los Angeles Times’ initiative on this. You have probably also come across references to Il Foglio, the Italian publication that created an entirely AI-generated newspaper, complete with opinions and letters to the editor, mainly to be provocative.

Separately, visual data journalism brand The Pudding used Anthropic’s Claude to interrogate data and create a story and concluded: “When we look at what Claude produced, we don’t think it looks anything like a real story we’d make at The Pudding.

“It’s sort of like comparing a woodworking artisan’s table to one from IKEA. The artisans invest immense time and effort into their high-quality pieces, while IKEA produces things quickly and cheaply, and most people probably can’t tell the difference (or don’t care)... . But that doesn’t mean there’s no place for craftsmanship.”

But there are also immensely successful examples of AI-generated content that are actually paying off — mainly because they are transparently the IKEA table rather than pretending to be works of fine craftsmanship and because they are presented to readers who really just need an IKEA table at that moment in time.

For example, AI-generated bullet-pointed summaries seem to be fulfilling a legitimate need for many news brands without undermining their credibility.

The Financial Times created summaries after they realised readers were pasting FT content to ChatGPT and asking for summarisation.

“The most exciting outcome is that no one has needed to change anything factual, and it’s been live for several months now. Only some small stylistic changes have been made to the actual output,” said Liz Lohn, director of product, AI, and editorial tech.

She said audience engagement had not declined. “Not even on the article depth. We have even seen a little bit of a positive impact on overall engagement.”

Sweden’s NTM is seeing strong results with its AI summaries. It has created a tool that grabs the five most-read stories of the moment and summarises them. An editor quickly checks them for accuracy, and then NTM uses GenAI to convert them into audio for broadcast.

This audio product appears to be helping with retention — 40% of logged-in subscribers have used the service at least once — but also with reaching those closer to the top of the funnel better, since half of the total listeners are non-subscribers.

Singapore’s CNA is creating summaries, called FAST, to serve both the time-pressed news consumer and the news avoider. It mimics the TikTok, Instagram and YouTube Shorts swiping experience.

“FAST accounts for 32% of our monthly average pageview growth from the last financial year,” CNA said in a statement. The average pageviews per visit for FAST articles is nearly five times that of regular articles, while the average pageview per unique visitor per month for FAST articles is three times that of regular articles.

Are AI-generated summaries infallible? It looks like Bloomberg is still ironing out the kinks in its summaries, which consist of three bullet points condensing the main points of the article. It has corrected at least three dozen summaries of articles published this year.

Still, “99% of AI summaries meet our editorial standards,” Bloomberg said in a statement. And, in fairness, if we want to put that number into perspective, we would need to compare it to their correction rate on content created by humans.

Worthwhile links

• GenAI and lawsuits: The New York Times’ lawsuit against OpenAI and Microsoft is proceeding apace.
• GenAI for content-generation: Newsquest employs 36 “AI-assisted” reporters who use an AI-powered CMS to rewrite press releases into stories and are tasked with checking facts and quotes are correct.
• GenAI for content-generation II: Authorities used AI to write a report — which ended up containing “fabricated sources and quotes that do not exist,” according to iTromsø’s Rune Ytreberg. “A lot of the researchers quoted in the report said they have never said such things and never wrote books and articles which are quoted in the report.”
• GenAI and models: How Claude “thinks”: It is in a conceptual space that is shared between languages, suggesting it has a kind of universal “language of thought.” It plans what it will say many words ahead and writes to get to that destination. And it will, on occasion, it will give fake reasoning.
• GenAI and algorithms: AI content is evolving to manipulate algorithms rather than engage humans.
• GenAI and marketing: Understanding how AI agents interpret and compare brands is becoming as crucial as traditional brand tracking. 
• GenAI overviews: Research shows no observable impact on news traffic from shifts in Google’s AI Overviews feature, but that doesn’t mean publishers are immune to this.
• GenAI and digital replicas: H&M is working directly with models to create digital replicas of 30 different models to use in AI-generated images for purposes such as social-media posts and marketing campaigns.
• An AI diversion: Is the global economy going to plunge into a recession because Donald Trump used an LLM to come up with a formula for tariffs?

About this newsletter

Today’s newsletter is written by Sonali Verma, based in Toronto, and lead for the INMA Generative AI Initiative. Sonali will share research, case studies, and thought leadership on the topic of generative AI and how it relates to all areas of news media.

This newsletter is a public face of the Generative AI Initiative by INMA, outlined here. E-mail Sonali at sonali.verma@inma.org or connect with her on INMA’s Slack channel with thoughts, suggestions, and questions.

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