Ekstra Bladet used 3 core goals to create an AI-powered news experience

By Kasper Lindskow

Ekstra Bladet (part of JP/Politikens Hus)

Copenhagen, Capitol Region, Denmark


Since ChatGPT’s launch in November 2022, AI’s potential for news publishing has gained mainstream attention. However, at Ekstra Bladet we began our journey toward becoming an AI-powered news brand in 2020, and we now have multiple AI applications live on our platform.

In 2020, as we became aware of the transformative potential of AI, we asked ourselves a few strategic questions:

  • How will different AI technologies affect the news publishing value chain?
  • What will be the time frame for industrial maturity?
  • How do we integrate AI in a way that is aligned with our values and avoid the adverse effects seen on social media?

Enter the PIN 

Our answer was the Platform Intelligence in News (PIN) project. The 3.5-year programme involved collaboration with three universities and multiple sub-projects aimed at addressing three core pain points in the existing news experience on our own news platform:

  1. Readers would miss relevant news stories if they did not visit us hourly because of the rapid pace of our news flow.
  2. Readers would miss the depth of our coverage of news events, as they were led back to the “ala carte” menu on the front page after reading a news story.
  3. Readers would not find all information relevant to news stories and instead need to go off our site to Google for context on persons, organisations, trends, or even related news stories.

Like Netflix — but not 

Incidentally, the three types of pain had been successfully addressed with AI by tech-driven media firms. Netflix, for instance, had already used recommender systems to broaden the discovery of entertainment video. Spotify had used another type of recommender system to produce a deeper music experience via auto-generated playlists. Google had used natural language processing (NLP) to provide a richer experience by providing fact boxes or text-based summaries answering readers’ questions.

We drew inspiration from these tech-driven media firms but, at the same time, knew we did not want to become like them. For that reason, we set out to address the identified pains while focusing on adapting AI technologies to the unique characteristics of the news domain, including the rapid pace of news and our editorial values.

More specifically, we formulated three core goals:

  1. We wanted to create a new iteration of the news experience that was broader, deeper, and richer, and thus addressed the core reader pains we had identified.
  2. We wanted to do so via AI systems aligned with our editorial and maintain control of the news flow rather than delegating it to tech firms.
  3. We wanted to contribute to a healthy norm setting for AI in the news industry in the process.

Accessing AI competencies

To achieve our ambitions, we formed partnerships with universities, leveraging generous funding and collaboration frameworks from Innovation Fund Denmark (IFD). We collaborated with four departments at three universities (University of Copenhagen, Copenhagen Business School, and Denmark Technical University) to develop AI systems and analyse news flow effects and ethical implications.

An innovation team at Ekstra Bladet — consisting of an ML specialist, a content automation manager, and two industrial PhDs — was crucial in bridging the gap between university partners and our data science team at Ekstra Bladet.

A look at the Platform in News project's organisation.
A look at the Platform in News project's organisation.

What we did

We focused on developing and aligning technologies in the AI domains that we expected would have the greatest impact on text-driven news experiences and would reach industrial maturity in the 2020-2023 period, including:

  • Recommender systems (RS) for personalised content matching.
  • Natural language processing (NLP) that connects related articles, enables interests-based reader segmentation, and links articles to relevant facts in external databases.
  • Natural language generation (NLG) that automatically generates relevant text descriptions and, in some cases, fully fleshed-out news articles.

We aimed to adapt these technologies to the news domain, addressing technical challenges and aligning with editorial values. Recent AI advancements, including content-based recommender systems and large language models, made us confident in achieving our goals within our three-year timeframe.

What happened

In 2022, we began implementing our AI systems at ekstrabladet.dk to broaden, deepen, and enrich our news experience, and we were happy with the results.

Recommender systems and NLP systems performed impressively in A/B tests (see figure below) and when we implemented them at scale, we achieved significant global effects:

  • We increased traffic (+80%), subscription sales (38%), and paid content usage (35%) across all horizontal positions on ekstrabladet.dk.
  • Further, our NLP systems for related article discovery achieved 10%-11% CTR rates, surpassing the most popular articles with a CTR of 4%-5%.

This shows the results from A/B tests of recommender systems vs non-AI based methods for news selection in 2022.
This shows the results from A/B tests of recommender systems vs non-AI based methods for news selection in 2022.

In February 2023, we also launched a beta version of our NLG pipeline for local news, combining human-written articles (identified with named entity recognition) with autogenerated content via a mix of rule-based systems and GPT-based generative summaries. While this test has not yet been evaluated, initial results are encouraging, and we expect that generative AI will play an increasing role for ekstrabladet.dk in the future.

About Kasper Lindskow

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.