I recently had the good fortune of chatting with an innovative INMA member that has been quiet for a long time about their experiments with GenAI, despite their history of successful data and machine learning projects. Today’s newsletter outlines how they envision GenAI as a core component in some of the use cases for a radically different news experience they are working on.

We also take a look at what INMA members consider to be the toughest nut to crack when it comes to GenAI: the question of user adoption. 

Thanks for reading!

Sonali

Telegraph envisions UX transformation via GenAI

Systematic experimentation with generative AI has the potential to be transformative and move publishers away from the article being the atomic unit of content, according Dylan Jacques, director of technology at Britain’s Telegraph Media Group.

His team has the goal of creating 12 use cases for AI in 12 months. 

“I didn’t for a minute think that we’d have 12 barnstorming, revenue-driving cases,” he said. Instead, the point is to treat each case as a building block for future capabilities and to gather useful data so the Telegraph can figure out what to place “big bets” on. 

“We need to be delivering a major feature every single month in front of customers, and it needs to be a live thing,” Jacques said “The team has very, very clear rules of the road.”

The Telegraph produces machine learning-driven newsletters, which draw from its recommendation engine, with a GenAI summary on top. 

“Instead of having five images and headlines that feel very aggregated, it will have a one-paragraph summary that says, ‘Dylan, you may have seen in the news today that Keir Starmer has announced…’ — so what we end up with is the equivalent of the editor introducing it in a human way.

“They provide quite a significant hook into the content itself. It’s reasonably simple but it’s quite effective. We’ve had a 20% uplift in click throughs from those articles. That was a really important proof point for us, and that paved the way for us to start inserting generated summaries into digital articles at the top.”

The team is now working on extending this concept further. For example, consumers may appreciate a quick audio briefing of the day’s key news stories with the option to click through to any of the articles to dive deeper.  

The Telegraph is also working on two other projects because Jacques believes the news consumer’s experience will be totally different in three or four years.

The first is a working prototype of what the future of news consumption may be.

“What I’d really like to do is have a significantly changed customer experience,” he said. “We’re working with our design team on a prototype that is really advanced and what the future of news consumption is going to be. It’s radically different from the traditional experience of an article and how one might consume five stories on a commute, and it’s all through the lens of GenAI.”

The team plans to run pilots of the five or six components of this new user experience with their audience.

The Telegraph is also building a chat interface based on its extensive travel content, which caters to a different audience from its core politics news audience: “We have amassed a huge catalogue of travel content. It’s a little bit hard to navigate because it’s a cacophony of lots of different things.”

A chat product would ask the consumer what they would like to do and use that as a springboard for providing relevant content. The reader would also be able to ask about specific needs — such as travelling with children or stopping somewhere in particular on the way — and receive an itinerary or a guide for the trip.

“Publishers have not historically been very good at monetising their archive and the reason for that is that the atomic unit has been the article. The leap that we’re now seeing with GenAI is that you can stop thinking about an article because you’re not going to be returning just a list of articles, some of which might not be at all relevant. 

“But there might be a paragraph in one of them which, when combined with a paragraph from another one, becomes better and more than the sum of its parts, richer. And I think travel’s a really good example of that,” Jacques said, citing obscure facts in various articles that could be recombined to create a useful guide for a visitor to a city as either text or audio.

ICYMI: Did you hear the brilliant speakers at our Generative AI Town Hall? You can still watch the recording here.

GenAI adoption: Why is it so hard?

For all the experiments the news media industry is running with GenAI, what is the biggest challenge they face? 

Leaders are concerned about a slew of risks, ranging from unpredictable costs to reputational damage due to unreliable output. But the one that pops up most often in conversations with those building GenAI applications is: dealing with humans.

“Employees don’t resist AI because they fail to understand its benefits — they resist because they’re not convinced those benefits apply to them. And why should they fully support a change that might disrupt their jobs or overturn familiar processes? AI anxiety is real,” data scientist Tobias Zwingmann wrote recently.  

“To be clear: The human factor is what makes or breaks your AI road map. Without engaged employees, AI will never reach its full potential.”

Indeed, an informal poll during the end-of-year INMA GenAI Town Hall, for which almost 500 people registered, showed user acceptance and adoption of GenAI was the biggest challenge for the largest number of respondents.

“AI is a people business,” as Robert Zilz, head of data at Germany’s Kölner Stadt-Anzeiger, recently said. 

“Building products and applications is the easiest part of the process. Then, putting in place related processes is a bit more difficult. But organisation and culture are the most challenging to change.”

It turns out running a GenAI pilot can be easy — but scaling GenAI adoption across an enterprise often proves to be difficult. People are afraid of being replaced by the machine and, in some cases, will even use the tool but pretend that they did not. 

A recent survey by Slack of more than 17,000 workers across the world found about half (48%) of all desk workers felt uncomfortable admitting to their manager they used AI for common workplace tasks because they feared it made them seem less competent or appear lazy.

In other cases, the novelty and excitement of having a cool new toy wears off, and users shrug off the new technology.

What works best? When editors build tools themselves, as they do at Omni in Sweden. Similarly, Newsquest and The Financial Times in the UK give journalists a GenAI toolbox to experiment with and build their own solutions. This is now possible because GenAI is so easy to work with, unlike older AI, which required a more technical background. 

It also helps ensure GenAI efforts are focused on solving real problems. 

“The advantage of being led by editorial is that you don't end up with AI solutions looking for journalism problems. This also means that journalists are having to think about things like costs and data more than before, meaning a generally less siloed workflow,” wrote Enders analyst Niamh Burns in a recent report on AI usage in the British media industry.

Other important steps are to reduce fear by specifying clearly where AI can be used and to have leaders model that by using it themselves, according to Wharton professor Ethan Mollick. 

“Figure out how to reward people for revealing AI use. If productivity gains happen, workers need to benefit as well,” he said, citing not only monetary benefits but also incentives such as the ability to work from home. “Give others the opportunity to show their uses as well.”

Worthwhile links 

• GenAI and advertising: This can go wrong and be insensitive and in poor taste.
• GenAI and investment: DMG Media invests in AI content marketplace Prorata.AI.
• GenAI and investigations: One man + a US$20/month ChatGPT subscription = a bot that interrogates the federal budget document in the Philippines.
• GenAI assistants: A handy guide with best practices and templates.
• GenAI and misinformation: This tool uses AI to monitor and detect misinformation.
• GenAI and licensing: Authors can opt out of an agreement between an unnamed AI platform and HarperCollins.
• GenAI and traffic: Still worrying about Google search referrals dropping off? Many consumers have already moved on to answer engines.
• GenAI and trust: We need more than just labels to build trust with the news consumer.
• GenAI and lawsuits: India’s ANI sues OpenAI for stealing content and damaging its reputation.
• GenAI and content moderation: TikTok eliminates hundreds of positions as it shifts content moderation work to AI.

An AI diversion

I love this idea: A bot that wastes phone scammers’ time.

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.

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