I came across some striking examples of AI being used by newsrooms across the world and thought I’d share them with you because each one is genuinely helpful, solves a real problem, and delivers real results.

The use cases range from breaking news to archival research, from video to text, and from algorithmic distribution of stories to automated comment moderation. 

It’s not just about the technology, though. In each case, please note the key role that journalists play — human judgment and ingenuity is still vital for any of these to succeed.

1. The Minnesota Star Tribune 

The Star Tribune used AI to do some heavy lifting after news of a mass shooting broke. As its journalists started digging for information on the shooter, they discovered a series of videos that had been published the morning of the attack.

The videos showed hundreds of pages of text from the shooter’s journal in what appeared to be a foreign language. An AI Lab member quickly uploaded the data to ChatGPT and asked it to translate the pages.

But the newspaper also went one step further: It asked experts to review key passages in the translated documents. They found crucial discrepancies in the translation, which were caught before any work was published.

Through the translations, the newsroom was able to swiftly piece together an article on the assailant’s background within a few hours.

“The AI and human translations allowed us to understand the violent, dark, and racist thoughts of a person who would fire more than 100 bullets into a church full of children and elderly people,” said Tom Schek, The Star Tribune’s investigative editor.

Its journalists also learned to coax information out of the LLM by introducing workarounds. For example, when the AI rejected queries on hateful rhetoric, they prompted it by saying that the input was fiction written by a friend who had asked for feedback on her work.

2. The Austrian Press Agency 

The agency is using GenAI to meet new accessibility standards, providing alt text for infographics so screen readers can easily interpret a graph or a chart.

APA had not been providing alt text because it took too long to create it manually and then to check it. “The amount of workload for our editors was actually not feasible. It would have been too expensive,” said APA Chief AI Strategist Katharina Schell.

The APA team developed a prompt that included an instruction to refer to the title and the subtitle, and also reflect the theme of the graphic. The AI was also required to read all the data and refer to the most important, while identifying the source and citing it at the end.

Before the tool was deployed, infographics editors were asked to rate 150 sample narratives the machine produced to assess how good it was. Now, the editors simply need to press a button in a tool they are already using to generate the alt text.

An added benefit? 

“Whenever the alt text generator was not able to explain correctly what happens in the infographics, it was mostly not the problem of the alt text generator but the problem of the infographic. So we found that the infographics were often not easy to understand and that we had to improve the infographic in order to create correct alt text,” said Christian Haslacher, APA’s head of data journalism.

3. Times Internet 

Times Internet team has put a lot of thought into building a system that combines traditional machine learning with LLMs to algorithmically distribute content through push alerts, complementing its human editors, who can override the machine.

The editorial judgment agent considers how important or unimportant a story is. It also examines the news peg as well as who is speaking and the tone of the emotions in the writing, encoding elements of editorial experience into the algorithm. It also looks at geographical relevance to the reader and at how fast-moving or stale a story is.

“It allows us to algorithmically distribute without compromising trust. This also lays the foundation for editorially sophisticated tasks like headline writing,” wrote Ritvvij Parrikh, senior director of product management, and Arghya Roychowdhury, who oversees personalisation for Times Internet.

“While this release addressed what the user should and should not see, in a future release, we intend to tackle the same aspect from the personalisation perspective — what the user wants to see based on their interests.”

4. Funke Mediengruppe 

Funke has created a tool that turns written stories into short video reels to help them reach younger audiences who spend time on social media.

“How can we accelerate the process of getting our content, which is already published, on to social media without creating a lot of effort for the people who are working on this? Because creating reels takes a lot of time and we needed a solution that allows us to do this more efficiently,” said Paul Elvers, head of AI at Funke. 

A journalist can pick a template in the tool, paste their text in, and upload photos and videos. They arrange the material in the order they want and can modify the tone of voice. The AI then generates text sequences for the images, which can be manually edited. Users pick an AI voice for narration as well as pronunciation styles and background music, animation effects, orientation, and the possibility of subtitles. 

“The Reel Machine perfectly embodies our approach: AI as an amplifier, not a substitute for journalistic work,” Elvers said. “While our editors continue to do the important work of researching, interviewing, and observing, our tool allows them to turn these precious written stories into engaging reels in minutes. In this way, we reach more people with the information they need about their region.

“This is exactly what AI in journalism should do: to bring the painstakingly developed, unique stories of our journalists to different target groups more efficiently.”  

5. The Toronto Star 

The Star uses AI in its comments section to automate content moderation and spot story angles. Commenting is open to all readers, while articles are paywalled. This feeds the Star’s efforts to gather first-party data: Commenters constitute a quarter of all registrations.

Editors receive a daily e-mail with half a dozen story pitches based on the previous day’s comments, conversations, and likes. Reporters can also query the comments to find summaries of what readers are saying about topics on their beat. 

“The AI effectively transforms months of community conversations into a searchable database of potential sources, story angles, and reader concerns — turning audience engagement into a competitive reporting advantage,” as journalist Ulrike Langer wrote. “The AI specifically hunts for personal anecdotes and human experiences in comments, which creates more compelling content for many topics.”

6. The Philadelphia Inquirer 

The Inquirer designed an AI-powered assistant to retrieve and summarise archival content, while linking directly to the source and providing transparent citations.

The Inquirer’s archives are housed in different systems, depending on how old they are. Retrieving information is difficult because reporters need to guess date ranges and hope they have their keywords right.

The new assistant is a conversational interface that understands natural language and synthesises the information reporters are looking for. 

“For longer-term tasks, like ‘Summarise a decade of coverage on XXX’, reporters said it could save several days of research,” wrote David Chivers of the Lenfest Institute. 

“The goal was not to replace reporters’ judgment but to free them from hours of tedious searching, give back creative time, and model a solution that other newsrooms with deep archives could adopt.”

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

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