From summaries to assisting in investigative reporting, how the FT uses AI

The more engaged consumers are, the more loyal they are, and therefore the less likely they are to churn.

Sometimes, though, readers just want the relevant facts of a story quickly and concisely.

For this reason, article summaries have become popular features with news publishers worldwide. Summaries are also an area where AI is being increasingly used by publishers, but not always in the same ways.

The FT is using summaries to both encourage direct engagement and also guarantee that the use of AI is transparent.

However, rather than putting these summaries at the top of an article, readers click a link within selected articles, which produces the summary and informs the user that it is AI generated.

As a bonus, rather than replacing the story, good summaries can actually help draw readers into the article, thereby deepening engagement.

Liz Lohn, the FT's Director of Product, AI and Editorial Tech, joined our Frankfurt AI Forum this week to discuss this and other ways the UK-based publisher has been experimenting with AI.

AI-focused team examines opportunities

The FT has experimented with AI for years, and in early 2024, they set up a small team they called Accelerate AI, Lohn said.

This team was ring-fenced, without having a backlog of requests or old systems to support and maintain, she said. Instead, its goals were to understand the opportunities and threats AI presented for the FT.

The Accelerate AI team looked at what was happening with AI, and examined questions such as: "What are the FT's competitors doing?" and "Who are the new players?"

Then they considered what the FT should do, including employing technologies such as AI and generative AI.

Using AI to draw links in obscure datasets

For example, one area where the FT has decided to deploy AI is to use it to help surface information in investigative reporting.

With the help of AI and machine learning pipelines, FT journalists were able to establish links from a dataset that documents the interests of Members of Parliament in the UK and their financial affiliations, Lohn said.

Such datasets can be complex and difficult for journalists to analyse, she noted, "so, we created a tool for them that ingests that dataset."

This tool understands what entities are there, extracts them, and separates companies from people and distinguishes between different types of financial interests, Lohn said. It then categorises them through metadata even further.

This way, the analysis can be done not only on one person but can also be used to ask more abstract questions, such as "Who gets the most football tickets?" or "Who is the biggest landlord in the UK?"

For those interested in details about how they did this, Lohn wrote an article recently where she explains more about it.

Next up, using AI to encourage comments

Looking ahead, Lohn said the FT is about to launch a feature to help drive consumer comments around their articles. This is also an area where the FT had previously experimented with using AI in various ways. These earlier efforts were abandoned because they either created extra layers of work or were too difficult to scale.

Comments are a powerful tool for engagement, and readers who make them are often among the brand's most loyal users. However, the FT has found that many readers are also unaware of this feature.

Lohn's team has come up with a potential solution to enlighten and encourage these consumers.

"We've created a prompt that basically reads the article and suggests the questions that could drive a good quality discussion," Lohn said. "Then our audience engagement team selects the question from the suggested ones, so it is still human in the loop."

She added that the FT is aiming to measure the quality of the discussion as well as the engagement of readers.

Experiment now to avoid relying on outside help

While there has been a great deal of hype and uncertainty around AI, Lohn encouraged publishers to be proactive.

"It's fine to experiment and have some people building things into the void a little bit. The risk otherwise is to not engage with technology and with the change in the first place," she said.

This also means a publisher risks not being able to build some unique capabilities internally and instead relying on only external capabilities or external expertise when this technology becomes more professional and standardised, Lohn added.

Brian Veseling

brian.veseling@wan-ifra.org