Deciding where to start, both with identifying use cases for AI in publishing and actual implementation, can feel overwhelming. From securing data sources to getting colleagues on board, here is some advice from those who have already taken the leap, in this extract from our Practical AI for Local Media report.
‘Just do it’
With apologies to a certain well known sports equipment manufacturer, the best way to get started with practical AI is to ‘just do it’. Start with a smaller project, convince your newsroom that it will bring them and the publication real benefits, and get going.
But before you dive in, educate yourself on the basics of practical AI so that you can have informed conversations. Joseph Hook, Editor at PA Media’s RADAR said: “I think there’s a general misunderstanding over what AI means and I don’t think that the sort of rhetoric you hear about the likes of ChatGPT is necessarily helpful.”
The JournalismAI starter pack is an excellent starting point for beginners wanting to understand the huge potential of AI. The Journalism AI project developed the free online resource to help people at the start of their AI journey to begin exploring these technologies. Professor Charlie Beckett, Director, The Journalism AI Project said, “It’s a question of making sure you know a bit about AI first. Do a little bit of self education.”
Find your people
It’s also a good idea to involve people that are at least curious about what AI can bring to your organisation, and that maybe shouldn’t be your IT department “The best use of these technologies has been where you’ve got editorial engaged,” said Beckett. “Otherwise, you’re just going to waste a load of time creating a very shiny tool that nobody wants to use.”
Stavanger Aftenblad’s Online Editor Elin Stueland agreed that it is important to bring the newsroom with you, to help them see that automated content will not replace their journalism. That is about highlighting the benefits AI can bring; surfacing stories in data, providing alerts on unusual activity, publishing updates at speed.
These all free up time in the newsroom for what Cynthia DuBose, Vice President, Audience Growth & Content Monetization at McClatchy called ‘real journalism’.
Practical AI should be founded on realistic expectations, not ChatGPT-level hype and the general advice is to start small. Hook said to look for the small steps that can be taken to help improve processes. “Automate something you’re doing already,” he suggested.
Beckett, who has experience of dozens of AI projects since the Journalism AI project began in 2018, said it’s best to start relatively small before you leap into something more strategic. “Get used to doing one or two projects with one or two technologies,” he said, “But people do scale up very quickly. I’m really surprised how some news organisations, relatively small or medium sized, have said, ‘Now let’s go for it’.”
Cecilia Campbell, Chief Marketing Officer at United Robots said many clients start with data around a small area covering a few postal codes. That helps them understand what to measure, what’s working and what’s not working. “That’s normally the model,” she said. “They’ll do a small and inexpensive pilot first.”
“If you’re hearing that AI is something you should be using, it’s about getting an understanding of what AI is to start with, what is encompassed within that, and what small steps can be done to improve processes in a workplace.”
Joseph Hook, Editor, RADAR AI
Problems, problems, problems
None of the people we spoke to for this report talked much about the technologies they were using. All spoke at length about the problems they were trying to solve.
Jens Pettersson, Head of Editorial Development at NTM told us it is key to know your audience, their information needs, and the gaps you need to fill. He said, “What are your pain points? What do you need to change? Then you can see if these kinds of production approaches map into that. Can you solve some of this with robots?”
With Stavanger Aftenblad’s junior football reporting project successfully reducing churn, Stueland frames things a little more positively: “I would start with seeing all the possibilities. As long as you have data, it’s a never ending path of possibilities.”
Luuk Willekens, Data and Innovation Manager at NRC said AI has to help journalists or readers, ideally, both. “With our automated, personalized, newsletters we wanted to find ways to engage our readers more with our journalism and create more time for our editors to do other things.”
Secure your data sources
Data is the fuel that drives practical AI in local media. Whether the data sets used come from local and national government, for sports associations or from commercial data suppliers, securing a reliable source of accurate data is key. Remember Stueland said if the data is chaotic, you will never succeed.
Hook said the majority of the data RADAR uses is public data from official sources like the government and the health services. But they also use data from a handful of charities who are large enough to have localised data of their own and, following lockdown, used top-level data from payment card providers to show how spending habits were changing. “We could basically see people returning to pubs in every town,” he said.
Equally for projects like NRC’s personalised newsletter project, integrating internal audience data with automation systems is crucial to delivering meaningful products and to support ongoing communication with users.
Plan your outputs and distribution
The end point for your automated content is as important as the starting point. NTM’s Pettersson said it helps to think through how and where automated content should be published.
NTM uses an algorithm to help editors control the front pages of its websites. As well as start and stop times, editors assign stories news values from one to six, depending how important the news is. Pettersson’s team put threshold values on property sales and those over the threshold were given a high news value so that the algorithm will put them in a better position on the front pages.
“Try to be smart on how to visualise this kind of content in your editorial mix,” he said. “So it doesn’t just get put up on the bottom of the page. Use it as a true value for your customers.”
Promotion is as important as presentation; publishers need to let people know that this new content is available. Campbell said, “You have to think the whole way through, ‘How can we maximise what we get out of this?’. Somehow people think that success is built into this, the traffic or whatever. You can’t publish and leave it, you have to treat it like any content, you have to promote it.”
“You have to think the whole way through: how can we maximise what we get out of this? You can’t just switch it on and leave it, you have to treat it like any content. You have to set KPIs, you have to measure, you have to promote.”
Cecilia Campbell, Chief Marketing Officer, United Robots
Iterate against clear goals
Treating automated content like any other content means applying clear KPIs. Whatever your practical AI implementation is designed to achieve you need to measure its impact, from reducing churn to increasing engagement.
Everyone we spoke to explained the iteration processes they went through to get their automated content output to where they needed it to be. Stueland at Stavanger Aftenblad described changes made around player safeguarding, finding ways to report games in an interesting way without naming specific players.
Willekens explained how it was important to change newsletter story selection because too many opinion pieces were being included, sometimes eight articles out of the 12 in the newsletter.
Dubose said the promise to editorial teams was that automated content would never be published until they said it was ready. Training the bot to deliver text that was acceptable involved a lot of back and forth. “They would edit it, it would go off, it would come back, they would edit it again,” she explained.
“They would say, ‘Hey, we would never describe it like this. This is how we would write it.’ They were the decision makers.”
Pettersson advised a ‘just launch’ approach but not without measuring impact. “Get it out there and see the response from the audience. Follow up on a regular basis. What kind of effect does this give us? If we change this, do we get more people reading? Change one piece at a time and see how the audience responds. If they respond badly, change it back and try something else.”
This article is an extract from our new report, Practical AI for Publishers, sponsored by United Robots. Download it for free below, and listen to our special podcast documentary episode featuring the voices and experiences of the report’s interviewees here.
Your details will be used to send you the Practical AI report, as well as information about future Media Voices reports and United Robots communications. Please note United Robots and Media Voices are joint data controllers for this report.