When Facebook unveiled a bot platform for Messenger at its F8 developers’ conference, The Wall Street Journal was one of the first news outlets to jump on board.
Launched in April, to coincide with F8, the Journal's Messenger bot sends news stories in response to prompts such as "politics", and are generated by What’s News, a free-to-subscribers WSJ app that offers stories specifically optimised for mobile.
CNN's Messenger bot, launched on the same day, responds to prompts in a similar way, though it is more conversational.
While other outlets are experimenting with chat bots on various platforms – such as The New York Times' election Slack bot – Messenger offers an enormous opportunity for publishers to reach new audiences.
"On Facebook, we have this really robust, loyal community," explained Carla Zanoni, global head of emerging media at the WSJ.
"And I think they want a connection with us that’s more real-time and more intimate."
In this excerpt from WAN-IFRA's Trends in Newsrooms 2016 report, Zanoni and Himesh Patel, creative director at Dow Jones, explain some of the key aspects newsrooms need to consider when developing a Messenger bot.
While bots may offer benefits for publishers, enabling them to reach new audiences and offer new experiences, they also have inherent limitations when it comes to accountability and algorithmic transparency.
A report published at the end of 2015 by the Philip Merrill College of Journalism investigated how automated Twitter bots try to spread news, and found that of the 238 bots analysed, researchers were unable to determine the sources used to generate content for 45 percent of them.
Meanwhile, Microsoft’s AI Twitter bot Tay, which was designed to learn from interactions with other users but which ended up mimicking the language of racists and holocaust-deniers, was a worst-case example of what happens when bots go bad.
However, Zanoni is confident the WSJ Messenger app is as sturdy as it can be, since it is rooted in the "curated experience" behind What’s News, rather than in artificial intelligence.
Having said that, she noted that there may be potential challenges for news bots in general, in terms of what the bot serves up when asked a question about divisive subjects such as politics.
"Are you serving up one side’s viewpoint more often than you’re serving up another viewpoint?" she asked.
As with any news product, the development of a news bot requires close collaboration between different departments.
At the WSJ, editorial and creative teams worked closely together to try to determine what an audience would want from the bot experience.
A basic prototype was deployed two months before the bot was ready to face the public, and the Messenger app was built in just four weeks.
During those four weeks, Dow Jones refined the bot experience with input from David Marcus, Facebook’s VP of messaging products, and from San Francisco-based development company Notify.io.
Legacy news outlets such as the WSJ have a reputation for being more siloed than their newer, digitally native counterparts, so the work that went into the Messenger bot signified "a big shift" in how teams operate together at the WSJ, Zanoni said.
"I love that we’re calling it an experiment, because it is, but it also is a large-scale experiment in how we optimise working together," she added. "And we’re all learning that as we go."
Keep it simple
For news outlets interested in building their own bots, Patel advised keeping things simple – at least to start with.
"In terms of just creating it, I think we all started off as ‘let’s make this as simple as possible, let’s learn, and then we can build on it’," he said.
Patel also underlined the importance of publishers not being afraid to experiment or to fail when developing a new product such as a news bot.
"You’re never going to get it right the first time, because there’s always going to be technical issues or whatever else," he said.
"But that’s just part of the journey of launching something and playing in this field. So I advise: go ahead and try, but keep it simple and learn, then make it bigger and better."
Data and analytics are, of course, essential for determining how audiences are using different platforms, and what kind of content they are looking for.
The WJS’s Messenger bot has an insights dashboard, built by Notify.io, which shows keyword searches, what alerts users have created, and peak times when people are most likely to use the bot.
And while the amount of data available is currently still quite basic, more analytics could be added as the bot became more sophisticated, Patel said.
Voice and authenticity
As with any ventures into new spaces and platforms, Zanoni said it was essential for publishers "to know the core of your identity as an organisation, and to retain that voice and that presence in whatever it is you build."
"There’s a huge pull and draw to mould yourself like a chameleon to whatever the platform needs," she noted.
"But at the end of the day, the audience is coming to The Wall Street Journal to interact with us. They want that essential characteristic or voice or analysis and product."
One way to ensure the longevity of a news bot, continued Zanoni, is to really spend time "refining the voice" of the bot, so audiences feel they are connecting with your outlet "in an authentic way."
"And to have that kind of conversation lead us – and support the journalism that we’re doing – frankly is the most important core of our mission," she said.
- To read the full report visit wan-ifra.org/reports/2016
- If you'd like to find out more about other bots by media organisations, check out our round-up of news bots.
Free daily newsletter
- 3 ways to make data visualisations and interactives work on mobile
- #newsrw: How to follow along with today's newsrewired event
- 'We are going after our future readers' – Snapchat Discover and Facebook Live experiments at WSJ
- Advice from FT and WSJ for getting started with interactive graphics
- How to get started with interactive data journalism