Hello, everyone. It is spring in North America, a time typically associated with reawakening and renewal. It’s the perfect time to take stock of some new ideas in GenAI and to consider how they could yield fruit in the months ahead.

Today’s newsletter first takes a look at an experiment that could have been a good idea, but it went off the rails because (spoiler alert) the news brand did not keep a human in the loop. 

And then, we explore a way to mitigate the threat of GenAI news brands face from search engines, which were traditionally a rich source of traffic but increasingly are a source of “zero-click” AI-generated answers.

Sonali

An AI cautionary tale from the Los Angeles Times

If you’ve been reading this newsletter over the past year, you will know I am constantly looking for innovative ways in which news organisations are using GenAI.

Often, the use cases simply present more efficient ways to do what we are already doing rather than anything brand new, so I am particularly intrigued when I see use cases for work that has never really been done before.

It is in that spirit, then, that I would like to tell you about the Los Angeles Times. 

As you probably already know, this is a troubled U.S. newspaper that has seen a great deal of upheaval and turnover in recent years. A few weeks ago, it introduced a new AI feature, Insights, which provided AI-generated “perspective” on content and appeared alongside its Opinion columns, editorials, commentary, and similar content that provides a point of view on a particular topic.

“The purpose of Insights is to offer readers an instantly accessible way to see a wide range of different AI-enabled perspectives alongside the positions presented in the article. I believe providing more varied viewpoints supports our journalistic mission and will help readers navigate the issues facing this nation,” Executive Chairman Patrick Soon-Shion wrote in a letter to readers.

You could argue there are many good reasons for doing this. News brands are trying hard to rebuild trust with readers — what better way than by transparently showing them that they are aware of different points of view on a topic? Most news organisations are also trying to reach new audiences — surely showing alternative perspectives could help them get beyond their core subscribers?

These AI-generated Insights were not reviewed by the newsroom. You can guess what happened next.

A columnist published an article about the white supremacist group the Ku Klux Klan. The AI-generated Insights came up with “different views on the topic.”

Surely playing down the KKK’s role as a violent group that targets Black Americans is abhorrent (and especially meaningful after a renowned racist has just been elected president). The backlash was swift. The LA Times ended up pulling the feature from the column but leaving it intact on others, leading observers to speculate that they really did not understand the potential magnitude of the problem.

This AI application is problematic for other reasons as well.

The AI seems pro-AI and misrepresents articles critical of AI, and the sourcing (via AI engine Perplexity) is also dubious. Also, isn’t the whole point of an opinion column to provide you with a cogent argument supporting a particular point of view rather than to spell out all the possible perspectives on it?

It comes just months after the LA Times introduced a different AI feature (via news app Particle) that classifies where a piece of content falls on the political spectrum and flags that for readers as Left, Center Left, Center, Center Right, or Right.

This “bias meter” has been panned for lacking transparency on how it concluded which label ought to be applied — which is ironic because transparency is surely the point of the label.

In that vein, then, who is doing transparency and trust-building well?

• You have already read here about Every, which uses GenAI to let readers interrogate the article to find out what was not included in an article. 

• Scandinavia’s Schibsted has introduced “ethics boxes,” titled “This is how we think,” to help readers understand why they made certain editorial decisions, such as naming a suspect in one news story and not in another. These boxes piggy-back off AI-generated fact boxes and summaries across the site.

• Brazil’s Aos Fatos has built a GenAI fact-checking chat product that can be accessed on platforms such as WhatsApp and Telegram.

Date for the calendar: April 3-10: Generative AI for Media Master Class

Next week, we kick off our latest master class on using AI for personalisation. The line-up of speakers and topics is terrific. Please join us for these sessions on April 3, April 8, and April 10.

Which content will weather the onslaught of AI-generated overviews? 

You may have seen a few studies in recent weeks suggesting news brands face more trouble ahead because of AI-generated overviews in search and Google’s AI Mode.

For example, the Tow Center for Digital Journalism found chatbots were generally bad at declining to answer questions they couldn’t answer accurately, offering incorrect or speculative answers instead, and that many were accessing content on sites that had blocked them.

They also saw generative search tools make up links and content licensing deals with news sources providing no guarantee of accurate citation in chatbot responses.

Trying to understand where your news brand is being featured in AI overviews? 

You can reasonably expect 70% of the pages ranking in overviews to change over two or three months, suggesting the results are volatile, another study showed. What’s more, 80% of consumers now rely on AI-written results for at least 40% of their searches, reducing organic Web traffic by 15% to 25%, according to research by Bain and Company. About 68% of LLM users rely on these platforms for researching, gathering, and summarising information, and some 48% use them to understand the latest news and weather.

And, at a time when many news brands are struggling to build sustainable digital subscriptions businesses, it turns out more than one-fifth of young Americans are willing to pay for an AI subscription.

Yet, there seems to be another, more positive trend many media industry observers are independently picking up on and which news publishers can easily capitalise on: the resilience of video.

“Video remains notably resistant to AI disruption, primarily because audiences deeply value authentic human experiences and genuine storytelling,” wrote analyst Scott Purcell. “Despite rapid advances in technologies like OpenAI’s Sora, people continue to gravitate toward content where real individuals share relatable experiences, emotions, and insights — qualities AI struggles to convincingly replicate.”

Similarly, Future Today Strategy Group’s Sam Guzik envisions a future where the Internet is overrun with low-quality AI-generated media. “Even as search engines and digital assistants try to sift through the vast volumes of content generated every day, consumers spend less time browsing because it is unpleasant — if not impossible — to find what they’re looking for.”

In this era, broadcast content is perceived as more reliable for two reasons, he says: 

“First, the cost of broadcasting establishes a significant barrier to entry, making it economically infeasible for low-value publishers to distribute content that way. Second, on-device AI models constantly curate broadcast data, creating a personalised stream of news and information that is easier for consumers to manage.”

As the Reuters Institute for the Study of Journalism pointed out in its annual digital news report, video is becoming a more important source of online news, especially with younger audiences: “Short news videos are accessed by two-thirds of our sample each week, with longer formats attracting around half.”

And yet: “The main locus of news video consumption is online platforms (72%) rather than publisher Web sites (22%), increasing the challenges around monetisation and connection,” the report added.

How are different news brands mitigating the threat from GenAI with video? Here are some examples:

• News Corp Australia is experimenting with vertical, mobile-first video.

• Mediacorp in Singapore has built an automated clipping tool and is working on multimodal retrieval augmented generation.

• Germany’s Der Spiegel has launched an audio and video podcast for young people, particularly news avoiders.

• The Washington Post uses humour to draw younger viewers to news on TikTok.

• Britain’s Future Plc uses GenAI to edit video for different distribution channels.

• The Hindu in India feeds a video into an AI application, which chunks it into chapters and creates time stamps for social media.

• This blog post outlines a GenAI workflow for converting long-form content to compelling short-form videos.

• Singapore’s Straits Times created new video features to reach and engage its audience more effectively.

• Sweden’s Expressen uses data to ensure its TikTok offering feels relevant to young viewers.

Worthwhile links 

• GenAI newspaper throughout: An Italian newspaper used AI for everything — writing, headlines, quotes, “even the irony,” and ran AI-generated letters to the editor.
• GenAI and summaries: Britain’s Independent will launch a news service which will summarise its journalism for “time-poor audiences.”
• GenAI and training: AI bots’ scraping activity more than doubled from the third quarter to the fourth quarter of 2024; scrapes that bypassed robots.txt grew by over 40%.
• GenAI and agents: Are they catching on?
• GenAI and agents II: Er, what exactly is an agent? Everyone has their own definition.
• GenAI and LLMs: A handy guide to which GenAI model is good at which tasks.
• GenAI and regulation: Spain will impose hefty fines of up to €35 million for not labelling AI-generated content.
• GenAI and jobs: AI-related jobs are the fastest-growing roles right now, LinkedIn says.
• GenAI and independent thinking: Reasoning models cheat at chess.
• An AI diversion: Turns out therapy chatbots get stressed as well — and mindfulness is the solution.

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.

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