Across Europe, the Middle East, and Africa, newsrooms are experimenting with conversational AI

A new report from FT Strategies, “Taking AI from the back-end to the front page,” digs into AI adoption case studies at 16 major newsrooms across Europe, the Middle East, and Africa. All of the publications participated in AI Launchpad, a program run by the Financial Times’ media consultancy and Google News Initiative (GNI) from June to December of last year.

Many newsrooms in the AI Launchpad cohort joined in a period when generative AI tools like headline generators and automated story taggers were being held up in the industry as a way to streamline editorial workflows and, perhaps, provide a bump in productivity. By the end of the program though, several newsrooms ended up finding value in these technologies by building conversational AI tools for readers — in other words, chatbots.

Several of the case studies look at news outlets’ attempts to find new ways for readers to interact with their archives and boost audience engagement. In a couple of cases, though, the chatbots were built to try to handle customer service questions, lessening the burden on staffers.

Al-Masry Al-Youm, a daily newspaper in Egypt, built a chatbot to help its readers navigate its archive of over 3 million stories. Key to the pilot was partnering with Miso.ai, a third-party vendor that offers “AI search engine” products. Al-Masry Al-Youm wanted to move beyond basic keyword search on its site, and the Miso team was able to integrate a combination of semantic search algorithms and generative AI into its experimental chatbot.

Readers could ask the chatbot simple questions and get responses that linked to Al-Masry Al-Youm’s past articles. First, the team created a “sandbox” to test the chatbot on 100,000 articles, before widening it to cover the publication’s entire archive. Importantly, Miso was able to help the publication fine-tune its models for the Arabic language and customize its UI to read from right to left. FT Strategies calls the chatbot “the first tools of its kind in the region,” largely because of its fluency in Arabic.

Daily Maverick, a South African digital outlet and weekly print newspaper, linked up with the vendor Bridged Media, which offers “no code, AI toolkits.” One of the publication’s goals was to create an agent that could handle basic customer service requests and lessen the load on human staffers.

In its tests, the tool responded accurately to 78% of questions. But after launch, the Daily Maverick didn’t see a drop in customer service questions. The publication’s running theory is that the “agent was deployed in the wrong channel,” since most customer service questions arrive by email. Looking ahead, the team is exploring how to embed these technologies into its customer service inbox, rather than on its site.

Ruhr Nachrichten, a regional daily newspaper out of Dortmund, Germany, also explored a partnership with Bridged Media. Rather than building a chatbot that could comb through the publication’s entire archive, the team prioritized recency by surfacing articles published in the last 30 days. The chatbot could answer questions in German about events in the area, local politicians, and trending news. In tests, the publication measured a 91% “success rate in responding to queries.”

In all, during the test phase, the tool clocked 1,400 reader interactions, with an average of 50 seconds per session, suggesting engagement. The publication also measured performance via  clickthroughs to articles cited by the tool. In its trial run, Ruhr Nachrichten measured a 31% clickthrough rate to stories.

Even with the high engagement numbers and apparent story discovery benefits, some readers wrote in with complaints about hallucinations and other factual errors in the tool’s outputs. The complaints prompted the publication to rework its disclaimers and “usage guidelines” to better set reader expectations.

The report highlights that many countries are currently experiencing declining trust in news and that AI experiments, especially when used irresponsibly, could further erode that trust. “News organizations must preserve trust while embracing experimentation with technologies that are still new and unfamiliar,” write the authors.

You can read the full FT Strategies report here, including all 16 case studies from the AI Launchpad cohort.