Reuters and Rheinische Post (RP) may seem like very different types of news organisations: one is a global news agency serving media outlets around the world, while the other is a regional publisher based in Düsseldorf that produces four regional and local dailies, including Rheinische Post itself.
But when it comes to AI, the companies share an approach that is pragmatic and cautious, focused on exploring concrete use cases – also beyond editorial workflows – while rigorously protecting journalistic integrity.
The two organisations discussed their AI journeys at our Frankfurt AI Forum.
AI supporting call centres and logistics
New AI tools have allowed RP to streamline its customer service. They implemented an AI-powered assistant, available 24/7, that handles all inbound calls from customers – the most common questions being: “Why haven’t I received my newspaper?” and “Can I pause my subscription during the holidays?” said Margret Seeger, Director Digital Publishing and Head of AI at Rheinische Post Mediengruppe.
This has made their call centre operation particularly effective, said Seeger: “Within the media industry, we’re among the two or three top players in that area. And we’ve achieved significant savings with that.”
A similar system is now being introduced in RP’s logistics department, where an AI tool will take calls from newspaper deliverers and reorganise their delivery shifts as needed.
Reuters focuses on speed gains
For Reuters, AI’s core value lies in speed.
“We found that AI can help us be faster. And Reuters does want to be faster – that’s our business model,“ said Sabine Wollrab, Bureau Chief of Reuters for Germany, Austria and Switzerland.
One concrete example of speed gains is Fact Genie, an in-house tool that scans press releases in seconds and suggests key alerts for the newsroom to publish. Crucially, a human editor must still approve any publication.
“That’s a strategy we really stress: AI is a tool, but the journalist is responsible for what is being published,“ Wollrab said.
Read more: From lab to newsroom: How Reuters builds AI tools journalists actually use
The human-in-the-loop approach is is non-negotiable for Reuters, she said: “Trust is one of our selling points. Reuters is a very trusted brand. And we don’t want to sell that for AI.“
Changing the value of the editorial process
As for assessing AI’s impact on journalism, one could simplify the journalistic process down to two parts: the first step is about gathering information, and the second is writing it up, said Christoph Mayer, Partner and Managing Director, AI & Data Science Practice at the Schickler/Highberg management consultancy.
AI changes the value of those two steps significantly, he said.
“The second step is not so valuable anymore, because you can highly automate it. But the first step becomes even more valuable,“ said Mayer.
He estimated that AI will end up handling much of the content drafting and repurposing across platforms, freeing up journalists to focus more on sourcing, researching and storytelling.
From experimentation to structure
With regard to AI experiments, both Reuters and RP have moved from a somewhat unsystematic initial experimentation to a more structured approach involving the integration of AI tools across individual titles and journalists.
Even with a more centralised approach, RP’s local editorial departments still have a lot of freedom to try out new solutions, said Seeger.
“That’s the beauty of having four titles: we have four departments where we can test things. And once we approve something, we can plug-and-play it for everyone,“ she said.
Reuters also began using AI through grassroots experimentation, with journalists testing various tools to solve practical problems. That bottom-up momentum has since evolved into a more structured approach, including the development of custom in-house AI tools (such as Fact Genie).
“Being part of the Thomson Reuters group is an advantage. We can use their Thomson Reuters labs to work with LLMs and other experiments,” said Wollrab.
No AI for video or images
Both publishers are careful to draw a line between AI as a productivity enhancer and AI as a content creator.
Reuters does not use AI to generate journalistic content, video or images, Wollrab said. “With a photo you really want to show what is real, and not some kind of artificial reality.”
Transparency is a core principle, and any AI-generated content on Reuter’s platforms is clearly labeled, including transcriptions and translations.
RP’s policy is similar: “Our editorial guidelines say that we do not publish AI texts or pictures,” said Seeger.
On the other hand, RP has experimented using AI-generated images (labelling them as such) in marketing campaigns – but those campaigns performed worse than those using real images.
Is AI disclosure necessary in every use case? Probably not, said Mayer.
“If a large amount of the text is AI generated, you need to mark it. But if AI somehow helped you in that process, I don’t think you need to mark it,” he said. In his view, minor AI use could be compared to using a tool like a spell checker that no one thinks to disclose.
He also said that audiences may expect less transparency in some contexts, such as automated weather reports.
Enhanced translation tools
Looking ahead, rather than chasing new breakthroughs, both Reuters and RP are focusing on optimising the AI use cases that are already available to them in the near-term.
“There are so many use cases out there already. It’s really about applying them … If we get the basics right, already that would be a huge step ahead,“ said RP’s Seeger.
For Reuters, translation remains a critical bottleneck. “It’s not a word-by-word translation we need. We need to adapt something that’s been written for an international audience in English, in an English news style, into a German news article,” Wollrab said.
Translation tools that can also adjust relevance and structure to different cultural and media contexts would be tremendously helpful for them, she said.
“That’s always what we look for when we get a text from an international colleague. How can we transform it so that our German media clients can use it instantly?“
Hyperlocal AI: The next big leap?
Another future AI opportunity is in creating new processes for gathering and handling information, especially for local and regional publishers, Mayer said.
“There’s a huge opportunity in collecting and processing local information – emails, web crawling, municipality meetings – and developing a system that gives you your own agency feed of hyper-local information. I think this will be the next step, and I see that as a real game-changer,” he said.
RP tried something related to this three years ago: it automatically categorised content so that audiences in specific regions would get news based on their location. However, the tech wasn’t ready at the time.
Now they’re preparing to relaunch the feature using a more advanced, AI-powered solution.
“That’s an insight on how the technology has changed within the last three years,“ Seeger said.
