The Telegraph set out in March to launch 12 significant uses of AI in its newsroom over 12 months.
The tools include newsroom workflow aides, consumer-facing AI services and internal data discovery tools. They are packaged together in an internal tool called Pulse AI so staff have a one-stop shop for the new services.
The Telegraph‘s director of technology Dylan Jacques told Press Gazette’s Future of Media Explained podcast he had found it more valuable to “learn through doing” instead of getting bogged down in policy questions over AI.
“What we decided to do at the beginning of this year was really make a concerted step change and a bit of commitment to the organisation that we’re going to see through a practical set of tools in practice. We didn’t want it to be ‘this is coming, this is coming’ and it drag on too long.
Related
“We essentially pulled together a bit of a mission statement, if you like, that said we need to understand this in practice quite soon. We didn’t want to be spending six months or 12 months on a concept that then turned out to not be that valuable or hard to achieve.”
‘If you can improve engagement, all boats rise with that tide’
Jacques said the new products, which are using a “whole range of tools from across a multitude of vendors”, are already providing “real business value” and “definite investable opportunity”.
He explained: “So far it’s been tied into improved customer engagement. If you can improve engagement, then all of the boats rise with that tide really, and subscriptions will flow from that.”
The Telegraph has been running a subscriptions-first strategy since 2018 and it ended last year with 1,035,710 subscriptions, including more than 200,000 added with the company’s acquisition of the Chelsea Magazine Company.
Jacques said AI-generated summaries at the top of email newsletters have led to a 20% increase in click-through rates. These summaries are not used on every email – some are still human-written – but Jacques noted that as personalised content grows and people are sent different articles depending on their interests, journalists will not be able to write a version for each reader, whereas gen AI can.
Telegraph readers may also have noticed AI-generated summaries appearing at the top of some articles with a disclaimer.
Jacques said: “There was a lot of internal debate as to whether or not this is something that actually really adds value to the consumer, and points can be made either way. Obviously, our journalism is our product. It’s what we’re all here to do. And using AI products to summarise that, it might help certain people, whether they’re time poor on their commute, but if it detracts from people really spending the time to fully examine the journalism, it’s not necessarily the best thing for the organisation and journalism in the long term.”
Swedish daily Aftonbladet said last year that people were spending longer on articles with AI-generated summaries at the top. Jacques said this has not proved the case at The Telegraph but that page views per sessions tend to go up when people are given these summaries.
“We quite quickly understood that the dwell time did dip, but actually that’s not the end of the world.
“We’ve A/B tested showing summary points on articles at a fair amount of scale and we’re reasonably confident that page views per session goes up by a fairly clear, statistically significant number.”
Modern gen AI technology has meant ‘big leap forward’
The Telegraph is also looking at using generative AI to translate its flagship Ukraine: The Latest podcast which has an “enormous international audience” including Ukrainian and Russian speakers.
“One of the things that gen AI has enabled us to explore is translating our podcasts in a way that is usable and digestible in those native language,” Jacques said, adding that previously the tech meant it could be done but “in quite a robotic way”.
“I think they rely on sort of translating it to text and then reverse translating it back,” he said. “And actually what some of the more modern gen AI services have enabled us to do is create translations that are quite true to the voices of the presenters… and it does it in quite a seamless and impressive way. And actually it’s maintaining the cadence of voice and the fluidity of the discussion much better than the previous, much more robotic services could do. So that’s been a big leap forward.
“And I think that’s going to be a really interesting one because that’s another area where while we would love to have journalists and audio presenters in all sorts of languages, it’s not realistic to expect that we would ever be able to extend our model to ever achieve that. So this is us being able to achieve something, broaden our audience and really get to a much bigger market than we previously were able to purely by the gen AI tech.”
Other products rolled out so far include an SEO headline generator and a localisation tool suggesting changes to language and context for audiences in the US, for example by telling journalists that they might want to add a quick explanation of Remembrance Sunday.
Jacques said that “giving some context is very helpful as clearly you want there to be the minimum amount of obstacles in the consumer’s reading of something, and if it immediately feels like they’re reading something that they clearly just don’t understand or don’t have context from, we’re not going to see great engagement through it”.
Staff can also use an editorial research tool based on all of The Telegraph’s previous content to help them more easily add context or find past quotes. And they have a new AI data assistant helping them quickly get the answers to analytics questions like “how many subscriptions did we get yesterday?” or “what was engagement like in the money section of the website in the US last week?”
Jacques said that previously “you would have to get through quite a bit of comparative analysis and dashboard comparisons. This is now available at people’s fingertips quite instantly.”
Telegraph AI tips: ‘Learn through doing, be agnostic, invest in the capability’
Also potentially in the pipeline is a customer-facing search chatbot for the site based on its own archive, similar to functions rolled out this year by the Financial Times and The Washington Post.
“We have internal services that do that already and we’re just contemplating whether or not we want to extend that into the consumer-facing realm,” Jacques said.
“We’ve got a very ambitious multi-strand expansion plan for more of this next year. We’ve seen some real successes so lots of reasons to invest, and we really want to establish ourselves as a market leader in the gen AI space.”
Asked for his tips for other news publishers that may not be as far through their journey rolling out generative AI products, Jacques said: “I would say if you feel you’re being bogged down with policy type stuff, I would try and move to a model where you’re learning through doing rather than doing too much academic evaluation of whether you should.”
He added that it is best to be “quite agnostic” with the technology: “There are lots of tools out there, they change all the time, different models get released all the time which suddenly makes a use case that was unfeasible last month this month totally feasible. So you need to be quite across the technology and trying different things all the time. Hitching your flag to one vendor or another is probably a mistake in this area.”
Jacques added: “You want to invest in the capability within your organisation as much as you want to invest in the use cases.” The Telegraph has around six people dedicated to working on generative AI. The questions to ask when deciding on staffing, he noted, include “what would they be doing otherwise?” and “what are the development priorities within the organisation?”
Jacques continued: “You might find that, you know, all of a sudden you want to turn something off and spin something else up that’s new and you can do that easily… providing that you’ve got the people and the capability – the use case specifically maybe is a bit more transient.”
Email pged@pressgazette.co.uk to point out mistakes, provide story tips or send in a letter for publication on our "Letters Page" blog
