Generative AI can help news companies build new tools that allow readers to access content efficiently based on their personal interests and needs, from content-driven chatbots to article summarisation tools.

During the recent Generative AI Master Class, part of INMA’s Generative AI Initiative, media leaders from Axel Springer, The Globe and Mail, Frankfurter Allgemeine Zeitung, Süddeutsche Zeitung Digitale Medien, Medien Hub Bremen-Nordwest, Bergens Tidende, and India Today shared how their companies are using AI to better serve their customers.

Axel Springer’s chatbot engages readers

As a company with a reputation for innovation, Axel Springer in Germany has taken bold steps to stay on top of trends. The company’s journey into AI began in November 2022 when OpenAI released ChatGPT.

“We started experimenting and testing and organised ourselves into groups that we called a task force,” Alexandra Irina Nicolae, senior technical product owner, generative AI, said. Within that group, it formed the Knowledge Network Group, which is designed to share information about GenAI topics for journalism: “We can also come together and learn from each other.”

Listing several ways Axel Springer brands are using AI, Nicolae said German brand Bild offers users an easy way to use AI thanks to a chat feature that’s built around specific topics or stories: “So you can start talking to ChatGPT about whether you should have a cat or a dog as a pet. And that’s very easy on, so you are not starting to really chat.”

Using buttons on the site, users can go deeper and start interacting with ChatGPT, leading to a conversation. It is easy to use, so it has been popular even with non-technical people.

“They do have 2.5 million unique users per month, and more than 40 million answered questions,” Nicolae said. “So people are using those buttons and they’re spending twice as long on the platform as before. And 80% of the people are actually using [it], which proves that there is a use case around combining an assistant with content.”

The Globe and Mail’s RobBot helps answer reader questions

The Globe and Mail is using ChatGPT-4 to build a chatbot to answer personal finance questions. The widely-read Canadian newspaper is known for its business journalism including its Report On Business (ROB) content. The RobBot is named after this as well as popular columnist Rob Carrick, Mike Pletch, managing director of product and UX, said.

The new RobBot will be The Globe and Mail’s second chatbot iteration. The first used Facebook’s Similarity Search to find articles that were the closest match to a question posed. Pletch said it worked really well at finding relevant information and extracting it from articles, but it led to an odd experience where the user wasn’t given any context or wording around it.

“This time, we’re trying to improve upon that using Generative AI,” Pletch said.

The new RobBot will still have a user submit a personal finance question, for instance, “How much will I need to retire?” Then the RobBot will find that answer within The Globe and Mail’s article database then generate an answer and cite sources, which will be hyperlinked headlines that users can click into to read the original article.

RobBot is being built entirely in-house with just two students who are acting as consultants, Petch said, so it’s been inexpensive for the company. They plan to go live this summer and figure out how to monetize it later. Pletch said his main goal now is to develop something accurate and useful for readers that will be an added benefit for subscribers which should help with retention.

FAZ creates an article summarisation tool

When Frankfurter Allgemeine Zeitung (FAZ) in Germany surveyed its digital subscribers, they said they wanted a better overview of the news and faster access to content. And those who canceled said they did so because of lack of time.

“They felt like they were paying for something they didn’t even have time to read,” Nico Wilfer, chief product officer, said.

So his team used ChatGPT-3.5 from Open AI to build a summarisation tool that is now part of the FAZ mobile app and is available solely to subscribers. It launched in April 2023, and FAZ has seen improved retention and engagement since then.

The tool is popular enough that FAZ successfully combined it with several in-app subscriber campaigns to trigger subscriptions. Readers were prompted to subscribe to try the publications summarisation tool.

So far, the tool has performed really well for a relatively small audience. It is still offered exclusively for digital subscribers. So far 13,000 users have tried it and 81% of those surveyed say it is helpful.

Those who try it are also twice as engaged as other subscribers in terms of visits, though Wilfer acknowledged it is hard to tell if that is because it attracts more engaged readers in the first place.

Moving forward, FAZ hopes improved retention due to the summarising tool will help the publication increase digital subscribers to 300,000 by the end of 2025.

Süddeutsche Zeitung Digitale Medien launches advanced question-answering bot

Carmen Heger, head of data science at Süddeutsche Zeitung Digitale Medien, said the company’s advanced question-answering system was designed to deepen subscribers' engagement.

This system leverages AI to fetch and generate responses from SZ’s vast archive, allowing users to ask questions about news and events from 2023 and directing them to relevant articles.

The implementation was user-centric, ensuring subscribers could easily navigate and use the system. Questions could be posed either by selecting from a set of predefined queries or by entering custom questions, thereby enhancing interaction and engagement.

To help users understand the AI-generated nature of the responses, the interface was carefully designed to label outputs clearly as AI-generated and to provide links to source articles for transparency and verification.

After launching the AI system, more than 6,000 subscribers interacted with it. This number is particularly significant given that it was a new feature, and the launch was done “very silently” to a subscriber base of 270,000 digital subscribers.

“And of course, we had a frequently asked question section that explained how it actually works and what can go wrong and that it can go wrong,” Heger said. This open communication has helped set realistic expectations and encouraged users to contribute constructive feedback, establishing a continuous improvement loop that directly impacts their content consumption experience.

Medien Hub Bremen-Nordwest improves service for print customers

At Medien Hub Bremen-Nordwest in Germany, AI plays a surprising role for print customers. Fabian Rosekeit, the company’s head of CRM and growth, said despite the company’s efforts to convert customers to digital, many still cling to print — and when the newspaper isn’t delivered on time, they’re quick to pick up the phone.

With around 200,000 subscribers, the company received an average of 30,000 calls per month — mostly regarding delayed or missing newspapers. AI seemed like an obvious solution to help handle the staggering number of phone calls.

Thanks to a good existing database, Rosekeit said they knew many of the customers’ phone numbers. That allowed them to create a bot that could identify the customers from their phone number and process simple print complaints. In addition to letting the customer know there is a delivery delay, the bot can register a complaint case in the system.

The MVP goal was to have the bot handle about 20% of all calls; that number has now surpassed 30% and is closing in on 40%. As a testament to how effective the bot is at handling complaints, Rosekeit said only about 5% of callers want to talk to the customer service team.

“We are really happy with these results,” he said. “And our agents are happy, too. I think this is an important part of such a project because when the topic of AI comes up, it also creates fears: ‘Will I lose my job?’ So it was really important to involve the service team in this project and I think we were able to show them that it is helpful for them too — and that they have more time for the second-level support.”

Bergens Tidende simplifies customer feedback processing

Users have all seen them: rate the performance of this site with little smiley or frowny face emojis. That’s how Bergens Tidende in Norway conducts customer satisfaction surveys on its Web site. Since the collecting and organising and analysing of the data proved to be time consuming and costly, Bergens Tidende decided to bring AI in for this part, Magnus Helgesen, the company’s UX lead, said.

Using a product called BERT, which taps into information from the entire digital collection at the National Library of Norway — 200 years worth of literature — Bergens Tidende can group the comments by likeness instead of sentiment. There’s also a summary of all the comments from the last 30 days and those comments are inserted into a prompt using the AI model.

With some human oversight, they take all the comments and verify the themes. They can highlight good ideas from users or pain points for them. They also can share out the themes they’re seeing to the proper departments.

“So a lot of them go to the newsroom, but some also go to the product team or the consumer business,” Helgesen said. “It’s nice to see this qualitative data together with the quantitative data. It’s much easier to make actionable points from week to week.”

India Today creates AI news anchor

India Today is now a year down the road after launching its AI news anchor, Sana. Sana speaks multiple languages with no learning curve, never takes a day off or gets sick or tired. In April of 2023, Sana made her primetime television debut on India Today’s nine o’clock newscast. She delivers news headlines alongside their main TV anchor.

During those beginning stages, there were a lot of legal and ethical implications to consider, Vibhor Gandotra, head of strategy and special projects at India Today, said. They talked about copyright and personality rights and how far they could go in emulating a real news anchor. Once they had a proof of concept, they were able to quickly adopt more uses for Sana besides just a television and digital news anchor.

“She’s also a social media influencer,” Gandotra said. “She’s a virtual identity. That’s a new trend, which is picking up within the media space and even outside media space where AI influencers are actually commanding similar recognition and premium from advertisers as compared to real anchors.

“One of the surprise areas where we got interest from within our business was actually from radio business. This is something which we had not anticipated when we were building Sana. She became India’s first AI radio jockey.”

Sana is also being used as a recommendation engine since her brand and name are now resonating with people and using her is more visually appealing then a text-based recommendation engine, Gandotra said: “Over the course of the past year from a line of code, she’s now turning into an award-winning anchor.”

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