Personalizing the News Media Experience with Generative AI

Discover how major German news platform Zeit Online is using LLM technology to improve content discovery

The advent of large language models (LLMs) holds great promise for industries that rely on textual content. Adopted by forward-thinking organizations, generative AI technology opens up a wealth of opportunities for innovative applications. This case study explores how Zeit Online, a leading news site in Germany, leveraged the latest generative LLMs to create a groundbreaking product for personalized news consumption.

Driving reader engagement with customized content

News sites want to provide their readers with articles that they will find relevant, useful, and sometimes even entertaining. But the user experience on these platforms has traditionally suffered from a lack of personalization, making it difficult for readers to explore topics of specific interest to them or to engage more deeply with content. The question for many news organizations has been how to increase interactivity and discoverability while maintaining journalistic standards.

Zeit Online's proactive AI strategy

Zeit Online was quick to recognize the potential of generative AI to transform news delivery. Soon after the launch of ChatGPT, the Zeit Online team began to explore how the technology could be used to improve and extend the platform’s services. They started experimenting with LLMs to build a product that would make news discovery and engagement more intuitive and personalized.

"Generative AI has this foundational quality – you can build so many things on top of it. We're just seeing the beginning of how that's going to play out." – Sebastian Horn, Deputy Editor-in-Chief at Zeit Online and Director AI at Zeitverlag.

A product-centric approach

The team was small but agile, focusing on identifying meaningful applications of LLM technology within their digital strategy. They adopted a product mindset, using frequent testing, feedback, and iteration to guide their development process. This approach helped them quickly arrive at a product that they recently released for beta testing to all of their digital subscribers.

The tool called “Ask Zeit Online” (or “Fragen Sie Zeit Online” in German) allows readers to ask questions about current events and get answers based on Zeit Online's news archive, providing brief summaries with citations and links to the original sources. It is especially useful for readers who want to get up to speed on a topic quickly, such as when they lack the context to understand a particular news story.

Choosing the right tools

To make the most efficient use of LLMs, the team implemented a modified retrieval augmented generation (RAG) setup in their product. With RAG, an organization can seamlessly integrate its proprietary database – in Zeit Online's case, their large collection of high-quality articles – with the latest LLM technology to deliver a customized and fact-based media consumption experience.

deepset Cloud allowed the team to move quickly from concept to product, without having to first implement their own AI infrastructure from scratch. "With deepset Cloud, we were able to focus almost immediately on rapid iteration and user feedback to refine our solution."

Iterative development for impact

Rapid iteration is a best practice in product development, and one of the core tenets of deepset Cloud, because it allows AI teams to build the best product possible. Instead of developing their product in a vacuum and only showing it to real users upon release, teams can test every aspect of their evolving prototype, including the user experience, the underlying data, evaluation metrics, and more.

The Zeit Online team conducted more than 150 rounds of rapid development to refine their pilot into a robust and scalable product capable of delivering personalized news insights while mitigating the risk of misinformation.

The result, "Ask Zeit Online," is a modern LLM-powered experience that provides users with tailored answers from Zeit Online's extensive archive. A round of beta testing with one thousand users proved the value of the product and led to improvements in the user interface and a deeper understanding of user needs.

The future of AI-powered news media

Since its launch in January, the tool has become a staple on Zeit Online's homepage, praised for its ability to demystify complex issues and provide transparent attribution to the original source material. User feedback has been very positive, underscoring the product's impact on news consumption.

Beyond the existing platform, the Zeit Online team is working on other versions of the product. They are looking to build a number of internal and external AI-powered features that will help readers and journalists understand the world and consume news in a more personalized and interactive way.

Embracing AI for a competitive advantage

Zeit Online's journey illustrates the transformative role of generative AI in news media and sets a precedent for the industry. The success of "Ask Zeit Online" is a clear indication that AI will play a critical role in future media developments, requiring a proactive approach to AI adoption across the board.

deepset Cloud: The platform where AI teams realize their LLM products

deepset Cloud is assisting an ever-increasing number of AI teams in the development of highly scalable, reliable, and industry-leading LLM products.

Schedule a demo to see how deepset Cloud can benefit your business.