An in-house personalisation engine at Norway’s Aftenposten delivers each reader a front page that feels more relevant, more timely and more engaging.

The story of how Schibsted Media introduced the collaborative filter into its ranking algorithm to reach new and previously underserved audiences is told in an INMA Ideas Blog by senior product manager Erlend Fernandez Stedding, data analyst Hanna Lind Jørgensen and senior product manager Christoph Schmitz.

Over the years, the Curate personalisation engine has driven click-throughs, increasing time spent and helping balance editorial curation with data-driven insights.

“But even good systems can inherit bad habits,” they say. “Like many news sites, our audience isn’t evenly distributed across age and gender. Our average reader is a 55-year-old man. As harmless as that may sound, it introduces a significant bias in content ranking.

“When front pages are optimised purely on aggregate performance, we unintentionally cater to the preferences of the largest demographic. In our case, men above 55, followed by women and men in their 40s.”

The result has been that stories that appeal to these dominant groups consistently surface to the top, while equally strong journalism – whether it’s cultural commentary, sports stories, relationship pieces, or political editorials – crafted for a younger audience, struggle to gain visibility.

Not because the content isn’t there, but because it never gets the traction it needs to perform.

That’s the problem Aftenposten set out to fix. A new signal in our ranking algorithm, the collaborative filter looks at patterns in individual reading behaviour – millions of interactions across our site – and learns what readers with similar habits engage with, rather than relying on topic popularity or impressions.

“It doesn’t know your age. Or gender. It doesn’t care,” they say. “What it does know is what you like to read, and it connects the dots between you and users with similar habits.

“In effect, it allows a 30-year-old woman – one of our most underserved reader groups – to see content that people like her have found engaging, even if it hasn’t bubbled to the top of the general rankings.

“Because let’s face it: If you’re optimising only for what performs ‘well’, you’re optimising for whoever’s in the majority.”

By early 2024, a cross-functional team from Aftenposten and Schibsted had begun testing the collaborative filtering model as a dominant signal in our front-page algorithm.

Aftenposten’s front page consists of teasers, a block that shows pictures and text describing what each article is about. “We started small, initially weighting the signal at just 20 per cent, and gradually increased it to 40 per cent over the year as results came in.

“And the results were telling: Overall click-through rate (CTR) for the automated teasers rose by 23.5 per cent year over year, compared to just 3.6 per cent the year before.

Among logged-in non-subscribers, CTR jumped a staggering 57.9 per cent.

Sales teasers, now personalised instead of one-size-fits-all, saw 32.2 per cent growth in CTR.

Most striking of all, CTR among young target group users grew by 29.2 per cent and their CTR on sales teasers rose 34.3 per cent.

Content categories saw increased article impressions across multiple tests. For example, in an A/B test comparing a 0 per cent baseline with a 35 per cent collaborative filter, impressions rose by 16.7 per cent for health and fitness, 24.3 per cent for history, 21.7 per cent for parenting, and 2.1 per cent for career-related content. This implies that the content that was already produced now also gets its share of the spotlight on the front page for the appropriate readers.

“It’s working. And not just in terms of numbers – it’s working in terms of fairness. Serving all readers

“With Curate’s collaborative filtering now the dominant signal behind 97 of the 100 teasers on our front page (the remaining three are still editorially curated), we’re reaching new and previously underserved audiences without sacrificing our journalistic responsibility.

“Important news still gets featured, but now it sits alongside content that reflects a broader range of interests, especially for those who were previously ignored by a majority-led ranking system.

“We believe personalisation shouldn’t just be about boosting engagement; it should be about ensuring every reader gets a front page that respects their interests – even if they’re in the minority.”