This month marks the third anniversary of the release of ChatGPT, sparking significant discussion around artificial intelligence (AI) among the general public. The journalism industry, however, had already been considering where the technology could support content and processes.

At the forefront of this thinking has been JournalismAI, and this year they celebrated the sixth edition of their annual conference with more insightful panels and discussions around the challenges and opportunities that AI presents to the media. Below are some of the key takeaways from the two days:

Changes to search

One of the big challenges for publishers in the last year has been the roll out of Google AI overviews. This has had a huge impact on search with Irene Jay Liu, director of AI, emerging tech & regulation at the International Fund for Public Interest Media, stating that large publishers in Brazil, South Africa, and Indonesia have reported their traffic decreasing by 50 to 60% this year.

Ezra Eeman, strategy & innovation director at NPO, believes this plays into the wider issue of AI and journalism, and that organisations need to be aware that it isn’t just media adding in AI but it’s also the opposite: 

‘The bigger play that’s happening is that media is being added to AI and we’re becoming part of AI systems. And the question is: how do you remain visible in that system?’

It’s something that Uli Köppen, chief AI officer at Bayerischer Rundfunk, has also been thinking about:

‘Search is changing. The old deal, traffic versus content is kind of off the table right now, and we have to see what our USP is as media houses. In my point of view, our USP has always been trusted content and how can we translate trusted content to the AI mediated media environment right now?’

AI literacy in the newsroom

All publishers that are using AI tools and/or processes are scaling these at different speeds within their organisation, but difficulties can come with newsrooms getting left behind. Jane Barrett, head of AI strategy at Reuters, said that ‘you’re seeing this skills gap opening up, of the people who are really getting involved and seeing AI as a constant companion, and those who are moderate, or even zero, users’.

The way to combat this? Both Astrid Maier, editor-in-chief at DPA, and Uli Köppen have had regular in-person training sessions or meetings for the newsroom to ask questions, talk about use cases and much more. Uli said they are ‘slowly trying to raise the AI literacy within the newsroom’. At Newsquest, they are going one step further by giving reporters dedicated time to experiment as Jody Doherty-Cove, head of AI, explained:

‘They can go and build their custom GPTs, find different ways that it works, and essentially, work on their knowledge of how to build these tools. It’s been remarkable. We have built a freedom of information bot which helps us with information requests, and that came from an assistant reporter coming up with the idea of building it themselves’.

Build vs buy

The big question for publishers when looking at AI tools to assist their journalism is, do they build it themselves or do they buy it? 

At Newsquest, they went for the build option with a large language learning model (LLM) drafting out versions of articles based on notes. AI-assisted reporters were appointed to take on this role, which in turn freed up other reporters in the newsroom to get out of the office and make contacts. Jody elaborated:

‘Reporters can’t be turning around press releases anymore, and as a result, you have to change all of your onboarding, you have to change your messaging and you have to change your targets. Those are all things that we couldn’t start implementing if we didn’t build it ourselves’.

Astrid Maier and the DPA have also followed the build option:

‘We have integrated lots of AI functionalities into our self-built custom content management system. We have also been building lots of special GPTs and agents, and you can access them via your DPA ID. Then you can ask things like, please help me find some other angles to the story, or please take this piece and transform it into a FAQ and things like that.’

Building might not be an option for every publisher, but Jane Barrett said that you need to consider your option carefully when it comes to buying:

‘They’re not cheap tools if you’re going to roll them out at enterprise level. So making smart decisions about which tools go where and to whom, what are the problems you’re trying to solve? That’s all very important’.

Further potential uses of AI in journalism

The technology has already had a massive impact on newsrooms but many see further potential. For Dmitry Shishkin, strategic editorial advisor at Ringier Media, it’s about ‘producing less but making the content work much harder’ and that he believes can be done by multimodality. Dmitry created the user needs model and feels that multimodality could fit into that:

‘If a breaking news story happens, you can have an educate me piece created as a first person video, and you can have an update me piece written as a 200 word article, or you can have also give me perspective piece delivered as a Q and A, for example, altogether. But all those things only can happen if your ontology/taxonomy is clean enough’.

Sannuta Raghu, head of AI lab at Scoll Media, had a similar idea, seeing potential for users to get news in the form they are used to or would like it. For example, clicking a button and getting the piece as a mind map. While for Agnes Stenbom, head of IN/LAB at Schibsted, it’s about building more chatbots and ‘trying to make news a practical value that you can with’.

Irene Jay Liu sees the potential having a bigger effect on the industry as a whole:

‘A really big opportunity is in pluralism of media. Imagine that people don’t actually have to be college educated writers anymore, but maybe they know their communities really well and can get the news out there. We could potentially break the traditional elite power barriers that we see in legacy media’.