You know how we’ve talked in the past about how revolutionary it is that GenAI gives the news reader control over the user experience? Today, we feature a look at one such transformative idea, where the reader decides how much detail they want from a written text.

Bonus: It comes from a small company with few resources, a reminder you don’t need vast amounts of money or technical talent to make GenAI work for you.

Plus: A look at some of the big-picture thoughts and concerns of leaders at various news media brands across the world as they buckle in for another year of riding the AI transformation wave. 

Sonali

Personalisation of news: one news brand’s innovative idea

In need of a bit of inspiration? Let me tell you about a small newsroom doing extremely innovative audience-facing work with generative AI. 

I’m talking about Scroll.in, an Indian digital news publication founded in 2014 and currently counts about 35 employees. (To everyone who hates the well-resourced New York Times repeatedly being held up as an example to emulate — this one is for you.)

What they are doing is remarkable: building a generative user interface, which provides a high degree of personalisation in real time, customised to fit the needs of its audience. In fact, they are working on seven different ways in which text can be converted to another interface. 

Take a look at one of them: a slider the reader controls so they can receive information in as much or as little detail as they like. You can see the different options in these four screenshots demonstrating Scroll’s ability to provide different levels of complexity in text:

As with any newsroom, implementing personalisation has been a process.

“The biggest complaint about content personalisation, especially from an editorial point of view, is that the editor says, you know, the user may want cat videos all day; are we meant to give them cat videos all day?” said Sannuta Raghu, who leads Scroll’s AI Lab. 

This solution conforms to the editors’ intuition for what is relevant and important as a first draft of history while also making news accessible through different formats, she said. “How do we make it more accessible to users in terms of the knowledge and the value they're taking away from it?”

Another format Scroll is working on is converting text content into a calculator.

For example, coverage of the recent Indian federal budget included 2,000-word articles, 500-word articles, and quotes from the finance minister on what the new tax rate slabs are, a topic Scroll users care about.

“Is the article the best way to put this forward? Or are we able to do this as a calculator, for example?” Raghu asked.

Scroll has two employees working on this project full time and plans to launch a product by September that treats content as raw material for the user interaction.

“The very, very broad plan is to build a system where if you put in an article, it decides what the best format could be and spits it out at the other side, depending on understanding context and understanding the type of story. I don’t think we are there yet, but ideally what we want to do is to take a particular story and turn it into any of these formats at the push of a button.”

As with many innovations, the story behind this one started somewhere else. Raghu’s small team needed to produce social and short videos to cater to strong demand in the Indian market, where 600 million people have access to smartphones and cheap Internet data. But they did not have the bandwidth to create short shelf-life videos, which typically took five or six hours each to produce.

“We started working on what was essentially text to video,” she said, and that project, supported by a grant from Google, led them to these interfaces. “We are thinking of disaggregating the whole workflow and sort of separating text to text, text to social cards, and text to video, and then with the added text to interface element as well.”

By the way, the team now produces about 20 such videos a day. It can produce many more, but the constraint is now that editors like to look over the output before publishing, and they cannot handle more than 20 a day.

Want to hear Raghu speak and ask her your questions? She will be presenting at our GenAI Master Class in April — sign up here.

GenAI experimentation enters the scaling phase

As we move through 2025, one trend is emerging in my conversations with INMA members: The initial flurry of experimentation with GenAI is giving way to slower, more controlled scaling of projects.

After several months of working feverishly on interesting projects, of rolling around in digital sandboxes, many members are now letting the dust settle and proceeding with caution as they thoughtfully roll out solutions. 

“We’ve done a lot of experimentation with some pretty good pilots for the newsroom and for engineers. We’re now in the space of selecting two or three of those and investing in them properly,” the chief technology officer of one Asia-Pacific publisher told INMA. 

“What’s been interesting for us is that pilots have been successful. But if we just took those capabilities and didn’t incorporate them into workflows, it would almost add a time tax. The work now is actually getting those capabilities into established workflows.

“This time last year, it was, ‘Let’s farm the ideas,’ whereas now it's more like, ‘Let’s focus on two or three.’”

Another executive in Asia echoed this sentiment. When asked what they were working on, he said, “There isn’t a showstopper. We’ve been working on smaller projects and we’ve been focusing on enabling AI on the backend of operations, editorial, the CMS, just generally streamlining.”

The head of an AI lab in North America said something similar: “The point we’re at right now is a mix of getting pilots into a more stabilised place and getting a new round of pilots for our newsroom workflow going.

“We have enough adoption around AI that it is coming up in everyone’s road maps,” with the AI lab simply acting as a consultant, he added.

Which is not to suggest the path to this point has been smooth and free of friction. Indeed, some publishers are honest about what didn’t work as well. For example, the head of product at a North American publisher told us about a chat product they spent months working on: “We started testing it. At the end of the day, the result it gave you was not as good as ChatGPT or Google Gemini. So, it got shelved — by me.”

Often, part of the challenge is getting the rest of the news organisation on board with AI projects. “My goal is to better explain and communicate what we do,” a European head of data science said. 

Or, as a German executive told INMA: “The ultimate challenge that we are facing is potentially cultural willingness to adopt AI solutions. Neither the business nor the journalists are really up for AI-driven innovation. We always emphasise the gains in productivity and how far along testing we are.”

An executive at an Indian brand said using AI was now table stakes. “Initially, I saw it as greenfield, but now I am being more considered. It opens up a lot of opportunities everywhere. Any competitor can use it for the same purposes that you can use it for. 

“At the end of the day, what makes one news outlet different from another is its specific point of view and that is not something that an AI can give you.”

Worthwhile links 

• GenAI and translation: Sinclair is using GenAI for live simultaneous translation of broadcasts.
• GenAI and training: Who is training models to write better and factually? Looks like it is journalists.
• GenAI and readers: Consumers expect news organisations to put in place guidelines on AI use and be transparent with them.
• GenAI and referral traffic: How much traffic are news publishers receiving from AI-generated summaries in search? Looks like the volume is up sharply but still not significant.
• GenAI tools: A useful list of free, open-source tools for journalists.
• GenAI and fact-checking: Tools used by over 40 fact checking organisations working in three languages across 30 countries.
• GenAI and agents: A handy course from HuggingFace.
• A non-AI diversion: Cyborg cockroaches could help with search and rescue operations in disasters. 

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.

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