News fatigue is one of the biggest challenges facing digital media. Audiences are overwhelmed by a constant flow of negative stories, which erode trust, engagement, and subscriptions.

At Iltalehti — Finland’s leading digital news outlet with 2.8 million weekly readers — we are addressing this problem with an innovative newsroom project. By applying AI-powered sentiment analysis and combining it with the user needs model, we are rethinking how journalism can both engage readers and support long-term loyalty.

In Finland, trust in news is exceptionally high, and avoiding news is less common than in most countries. According to research from the Reuters Institute, 27% of Finns avoid the news at least sometimes.

However, readers increasingly showed signs of being tired of the news, driven by an overrepresentation of urgent, negative coverage. For a frontrunner brand, sustaining engagement required major pre-emptive steps.

A sentiment-aware newsroom

Journalists can easily assess whether an individual piece has a positive tone, for example. The challenge becomes more intricate when aiming to evaluate and improve the complete picture in the long run.

We built a sentiment analysis framework powered by AI. Every article is classified as positive, neutral, or negative, and further refined into categories such as constructive, inspiring, entertaining, useful, or provoking.

These insights are presented in real-time dashboards for editors and reporters. They show the sentiment mix of the latest stories, most-read articles, and even each journalist’s own reporting. Crucially, this is integrated with the best parts of the user needs model, allowing editors to see not just how content feels, but also whether it meets key reader needs.

This combination has triggered a cultural shift in the newsroom. Journalists are encouraged to consider both tone and purpose: not only “Is this newsworthy?” but also “Does this help readers in a meaningful way?” Constructive journalism — highlighting solutions and progress — has become an explicit newsroom priority.

The impact is clear: pageviews increased after implementation compared to the previous year, confirming our hypothesis on the importance of our project.

One particularly important step is our GenAI-powered headline tool. It generates multiple headline options using our fine-tuned model, and other AI predicts the hit potential of headline variants. After that, we analyse the sentiments and serve the best mix of headlines with the greatest hit potential for journalists to choose from. At that stage, the journalist can once more consider if the tone is right.

It’s not only negativity that sells

One of the most striking discoveries appears in subscriptions. Data shows that constructive, inspiring, and useful stories convert readers at the highest rate, while sad or negative stories convert less often. Retention data tells an even stronger story: subscriptions gained through inspiring journalism have the lowest churn, with only 49% cancelled within six months. In comparison, subscriptions originating from negative coverage — often leaving readers burned out on bad news — see churn rates above 60% in the same period.

This combination of sentiment and user needs provides the newsroom with a powerful, data-backed rationale for producing journalism that serves society and sustains the business.

since the world is what it is, the total number of negative stories is still highest, and they generate the largest number of subscriptions – . However, we can focus more on more constructive stories with higher conversion.

A scalable model

Our journey shows that AI does not replace journalists; it empowers them. Making both sentiment and user needs visible helps reporters reflect on their work, editors balance content, and audiences feel better served. At Iltalehti, it is already helping us fight news fatigue, improve loyalty, and deepen our relationship with readers.

This project is also a very valuable lesson in change management. Digital innovations are useless if they don’t resonate in the newsroom or if tools are not well integrated into the workflows.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .