ProSiebenSat.1 and AI: A new matter of fact

When Maren Langbehn enters the editorial office in Unterföhring in the morning, the first thing she does is check the latest figures and reports from the previous day. One click, one glance – and Langbehn knows whether it was a success for her team. During this routine, she doesn't waste a thought on the fact that artificial intelligence is behind the data collection, providing her with valuable insights and knowledge. AI is a natural working tool.

In the digital newsroom, Langbehn and her 13-strong editorial team create news content in the form of text and video for the web presence of the news program "Newstime", its Joyn presence and other digital channels. This makes it part of ProSiebenSat.1's central newsroom, which also includes a TV editorial team with its own TV studio. The newsroom is expected to move to the new mega studio in Unterföhring at the end of 2023. New possibilities await Langbehn's team there, such as green screens and automated cameras. "You just press a button and go live," says Langbehn. The Buzzroom also uses digital channels, but focuses primarily on celebrity and lifestyle topics. "However, Buzzroom and Newsroom have fundamentally similar production processes," says Langbehn.

In the Digital Newsroom and Buzzroom, AI is not an abstract idea, but an integral part of everyday working life. Langbehn and her team not only use AI to analyze data, identify trends and make strategic decisions. AI also plays a central role in content production.

AI at ProSiebenSat. 1: From experiment to guarantor of traffic records

The Buzzroom served as a pioneer. An initial experiment was launched here in 2019: an AI was tasked with translating German videos into English and Dutch to save time. "The results were disappointing,” says Langbehn, "Native speakers had to check and rewrite the AI results.” It turned out that the human professionals were even faster without AI. It was only during the coronavirus pandemic that AI solutions became established in the buzz room. At the time, the Robert Koch Institute (RKI) provided the population with a wide range of data, such as the number of infected people. Instead of manually creating a new video every morning with the latest case numbers for each federal state, the media company developed automated video clips together with Wochit, which provides an online video production service with its WAVE solution.

KI took the data from the RKI and integrated it into an existing video template that editors had created together with the AI Products colleagues from the IT subsidiary ProSiebenSat.1 Tech Solutions. "You came into the editorial office in the morning and the video was already online. That really brought us record traffic,” recalls Maren Langbehn. They now use the same system for automatically generated clips on accidents, police operations, rescue operations and criminal cases. Automated videos for sports results, recipes and stock market data are also in the pipeline. Another AI-based solution is a „neural voice” called Conrad, an artificial voice that ProSiebenSat.1 developed in collaboration with Microsoft. "We didn't have enough digital editors with voiceover experience available at off-peak times such as night shifts and weekends,” says Langbehn. Thanks to the artificial voice, it was possible to compensate for the staff shortage. Conrad is still in use today.

AI can prepare content for new target groups

In the meantime, the use of AI has become established in everyday life. "Of course we experimented with AI early on, but the topic has been exploding in our newsroom for what feels like four months now,” says Langbehn. The team regularly discusses ideas on how AI can be used even more. It is a continuous search for new tools and processes, says Langbehn. The aim is to make video, text and image creation faster and more efficient. "Shortening, summarizing and translating texts with AI is part of daily business,” explains Langbehn. The AI also takes care of the tedious tagging of images and can also create simple graphics. This gives employees a new and fast way to prepare content in multimedia format. In the future, AI will also help with highlight clipping for videos, i.e. it will extract the most important parts from longer videos and create a short spot.

The AI will also prepare TV material for online channels in the future. "We have vast amounts of moving images from ProSiebenSat.1's TV productions, which are currently lying unused in the archive,” says Langbehn. "The aim is for the AI to create a transcript of the individual videos, which editors can later use to write search engine-optimized texts.” This opens up another opportunity to make elaborately produced material available to as broad a target group as possible outside of linear television. Whether text, picture series or video: It always needs a headline and teaser. Here too, artificial intelligence can analyze the content and generate an initial suggestion. For sharing on social media, it creates suitable captions that arouse curiosity and attract users from Facebook, Instagram & Co. to the website.

AI in journalism creates more time for high-quality stories

Despite the wide range of possible applications, AI does not threaten the jobs of editors at ProSiebenSat.1. "We journalists are said to be skeptical people and ask a lot of questions. Of course, we also question artificial intelligence and its use. I think AI has to put up with that too,” explains Langbehn, "but I'm basically a fan of AI.’ She sees AI more as a useful helper for her editors. It can take over repetitive tasks in particular. This would allow journalists to focus more on their actual tasks: creating high-quality stories. Langbehn therefore advocates openness when dealing with AI. "Our editors' jobs will not disappear. In case of doubt, they will change a little,” she explains. Her editors see it the same way, experimenting independently with tools such as ChatGPT and proactively approaching the editorial team with suggestions. According to Langbehn, neither the buzzroom nor the newsroom are completely generated by artificial intelligence. The bottom line is that this would actually cost more time: "If I let the AI write an article, it has to be checked by an editor. It would also have to be ensured that the text meets our quality standards. That would be very time-consuming,” says Langbehn.

Only editors who are familiar with the topic would fulfill their journalistic duty of care. They would check their sources for credibility and reliability before including them in their reporting – unlike AI. The creation or mapping of bias is also a challenge in automated text creation. This means that the AI could distort information in a report or present it in a one-sided way. Large language models are trained with large data sets and draw on a variety of sources from the internet. They therefore also incorporate prejudices and stereotypes into their texts. Editors need to be particularly vigilant when checking this. In Langbehn's team, information is verified by at least two independent sources to prevent a possible bias from being passed on - regardless of whether the content was written by humans or whether AI was also used. "Journalistic due diligence is a top priority for us in our day-to-day editorial work,” says Langbehn, “which is why I have defined clear rules for dealing with AI together with the editorial team.” Transparency is particularly important to them: content edited with AI is labeled accordingly.

Customized AI solutions

In order to respond to the individual requirements of teams within the company, ProSiebenSat.1 has had an AI products team for many years that develops in-house AI solutions. Langbehn and the department are in close contact. "If you are reliant on operating only with ChatGPT & Co. We have special use cases here for which the AI Products team creates tailor-made solutions,” says Langbehn. In the future, she would also be happy to have dedicated AI positions in her team, such as prompters.