What can newsrooms learn from Swedish Radio AI strategy?

Sveriges Radio, Sweden’s national publicly funded radio broadcaster started experimenting with AI a few years ago, before the rise of generative models in 2022.

Every implementation is guided by the publisher’s global goal – audience growth. Sticking to that and using AI technologies Sveriges Radio optimised budget, editorial operations, search and content personalisation.

However, AI poses a threat to the public media core value – audience trust. To find out how the news broadcasters, even the public ones, can use AI without losing their credibility, The Fix spoke to Olle Zachrison, Head of Artificial Intelligence & News Strategy at Sveriges Radio.

Transcription unlocking 1.4 million audience potential

Sveriges Radio developed an internal text-to-speech model for transcribing daily 360 news clips and most of hundreds of podcasts. Their technology appears to be cheaper and more accurate than previously used commercial models. Primarily it served for the improvement of internal search, but its mission expands this autumn. 

Swedish public radio is going to display the transcription in its app to make the content more accessible for the Swedes with some hearing difficulties, Zachrison told The Fix. Such difficulties affect nearly 13% of Sweden’s population (1,4 million people), according to the report by the European Federation of Hard Hearing People from 2015. The broadcaster also tests transcriptions of live radio. 

Tailored news offer, credible search and writing assistance

Sveriges Radio has at least two solutions that help curate news playlists and online news offerings, saving time in 33 newsrooms across the country. 

The first one is news rating by the journalists. In the CMS editors assess all pieces according to three criteria: general news value (the magnitude), life-span and Sveriges Radio values. The last one is meant for evaluation of public service values related to the broadcaster’s mission, like whether the report is from the ground and including affected voices. Analysis of 32,600 news clips showed that stories with high general news value – the most important criterion – in combination with high SR-value have 30-40% higher listenership, Zachrison told The Fix.

The second one is PEACH recommendations tool from the European Broadcasting Union (EBU), used not only by Sveriges Radio but other union’s members like Deutchlandradio or LRT (Lithuanian public TV station). It was first implemented when the team identified the need to show the toplists for popular podcasts in radio’s app, according to the comment on the PEACH website from Jörgen Bang, Sveriges Radio Product Owner. 

“It guides the decision process in the sense that it gives our news editors news recommendations. However, we don’t want AI tools to directly influence our editorial decisions but rather to make them more qualitative and – when needed – speedy”, Zachrison told The Fix.

AI also helps Sveriges Radio journalists on the content research stage. In the News Pilot, a news aggregator powered by EBU, journalists can find news from 26 European public news outlets. The service ensures automated translation from 17 languages, context-driven monitoring, and search tools, giving users in the newsroom direct access to information from across the geographical region. Basic interface of the aggregator offers not only text content, but also video with available transcription and translation.

Sveriges Radio actively tests other News Pilot functionality within the joint EBU project “A European Perspective”. It’s a generative chatbot Néo that enables conversational search, and extracts information from a database of more than 2.6 million articles from around 30 EBU members. The users can ask any question, like “Why is Sweden joining NATO?”, the service then picks out ten relevant articles, summarises them and links the user to the original articles, Zachrison said.

Additionally Sveriges Radio tests how genAI can generate suggestions for podcast captions and alt-texts.

Facing challenges

Despite AI’s integration, Sveriges Radio maintains strong human oversight, especially in editorial decisions, as Zachrison names trust issues as the first challenge coming within for public media using the technology. According to a 2023 Eurobarometer survey, public broadcasters enjoy the highest level of trust among EU citizens, with 48% expressing trust in public TV and radio, and the highest levels of trust up to 87% observed in Nordic countries such as Finland, Denmark, and Sweden. 

AI tools assist Sveriges Radio in decision-making but do not replace human judgement. The broadcaster emphasises transparency when using AI-generated content, ensuring audience trust remains intact. The publisher’s genAI guidelines include strict check of all AI generated content which is used in the journalistic process, lack of cloned or synthetic voices in audio content and visible disclaimers if some content piece (e.g. image) was generated by AI. Audiences generally want humans to control the process of the news publication at all times, showed the results of the recent Digital News Report by Reuters Institute. 

Among challenges brought by AI to publishers embracing it Zachrison additionally named prioritisation issues, information security issues relating to external tools usage, and legal complexities. For development in this field Sveriges Radio organised a new team called AI and News Strategy on the commissioning side with three AI project managers under Zachrison’s leadership. They work with the publisher’s cross-functional AI council that has representatives from all units in the company (HR, Technical, HR, financial, info security etc.) and suggest strategic direction to an AI committee in the company’s management. Sveriges Radio is also going to form an acceleration team with both development and editorial representatives. Update: to build concrete editorial AI tools to assist journalist, a new AI team on the developer side is also in place since September 1st.

---