Summary
This Reuters Institute case study is a useful legacy benchmark because it describes a newsroom that did not just buy a chatbot and improvise. JP/Politikens Media Group worked with university researchers for almost four years to build recommendation systems and a multitasking tool called Magna, while explicitly constraining those systems around editorial values.
Why It Matters
This matters for journalists because it documents what mature newsroom AI adoption can look like when the goal is not speed alone.
- the project joined newsroom staff and outside researchers over a multi-year build, test, and rollout cycle
- the team used a `values compass` to balance reader, journalistic, business, and technical goals before deciding how to optimize systems
- they rejected some optimization paths, including direct sentiment-based recommendation, because they thought those would conflict with editorial values
- they describe both product-side uses, such as recommender systems, and workflow-side uses, such as transcription, text analysis, story enrichment, summaries, and tone adaptation
What the Source Says
Reuters Institute reports that JP/Politikens Media Group and university researchers launched the Platform Intelligence in News project in October 2020 and ran it for three years and nine months. The team included about 17 people from the publisher and three universities, with support from Innovation Fund Denmark. The article says the project produced a multitasking tool called Magna and recommender systems now used by the company's titles. It quotes the company's Head of AI explaining that the team banned direct optimization on sentiment signals, limited recommendation of some topic types such as violent crime, and built systems meant to stay close to what a human editor would ordinarily recommend. The article also says Magna was being rolled out across the publisher's brands for writing assistance, story enrichment, summaries, quizzes, and channel-specific versions of finished stories.