Credit: Image courtesy of AJ+

The results of the US elections were mapped, livestreamed and pushed to mobile phone lock screens – they were also presented in a conversational style by news outlets on Facebook Messenger.

The New York Times and The Washington Post both launched their Facebook Messenger bots three weeks before the vote. While the NYT combined automated messages with updates from the outlet's political reporter, Nicholas Confessore, the Post monitored people's feelings about the election.

AJ+, the distributed news arm of Al Jazeera, went live with their bot, Mila, on 5 November. The day of the vote, readers were able to get results in real-time, infographics, and details about the presidential candidates' stands on particular issues.

If they typed in "immigration" or "abortion", Mila would return a fact card, profiles of the candidates, or an AJ+ video that could be watched inside Messenger. They could also send AJ+ photos reflecting their experiences on voting day.

"Our approach was focusing more on the thousands of videos AJ+ has produced since launching in 2014," Tawanda Kanhema, product manager at AJ+ and part of the team who created Mila, told Journalism.co.uk.

"Without a website, there's really no way to package these videos in a way that can be easily accessed, so the bot provided a search interface for people, where they could type any topic and get content on those issues."

In the newsroom, AJ+ producers used the Associated Press elections API to create graphics and visualisations from the results, that were then exported and fed into the bot as people were asking for the latest updates.

At the start of the day, the audience showed more interest in watching the videos Mila sent, but as the day progressed, they were "coming back every minute and typing in 'results' to see if anything had changed", said Adnan Chatriwala, product manager at AJ+.

Mila also encouraged people to share their experiences with voting, such as queueing at the polling stations, and send in selfies with their 'I voted' stickers and other images.

The team built a Slack integration that automatically sent the user-generated content into a Slack channel in the newsroom, where submissions were reviewed by staff and made into 'Voter Voices' cards, which AJ+ also shared on Twitter and Facebook.

"We had about 4,500 users for the bot, 14,000 sessions, and we got upwards of 100 submission from voters, but we ended up publishing between 15 and 20 stories on Twitter and Facebook," Kanhema said.

"It was a good experience for us in how to bring editorial and digital to work together with an idea within a short period of time, as well as introduce a new avenue for users to contribute content to our platform and push that content back to them in real-time.

"It brought a new format that had more engagement than the rest of the content we were doing on election day."

In the past, many Al Jazeera initiatives dealing with user-generated content have been done on a project-by-project basis, said Soud Hyder, platforms lead at AJ+, for example using SMS crowdsourcing in places like Somalia and Uganda before conversations started taking place on social media.

But Hyder wants to figure out a way for a one-off experiment such as Mila to be adapted to other news events and reused outside of the elections.

"We're now going through the data and the audience profiles to see what worked and what didn't, but we hope it will be an avenue to engage with the audience and find out what's happening from their point of view, especially for big news events," Hyder added.

"As we're planning for 2017, we want to have a separate workflow for 'micro-moment' producers at AJ+, so that includes messaging, notifications and bots, which are very lightweight, super-atomised pieces of content that need to be done in the heat of the moment."

The decision to develop a bot that had a character and personality came from wanting to replicate the style of conversations people adopt when they chat with friends on Facebook Messenger, explained Chatriwala, as the team didn't want Mila to be a bot that pushed headlines to people only to get them to come back to the website.

"When I go and open WhatsApp or Messenger I want to get messages from one of my friends and I want to send them a picture about something interesting I'm seeing on the street.

"If we consider the chatbot within our Messenger platform to be just like the way other people are talking to their friends, that's the default model of interaction.

"So we tried to be more careful about the language and referring to ourselves as a person not a media organisation. We also tried to send messages the way a person would, so instead of headlines, we'd send a video or a picture with a description, and people could ask for more if they wanted more information."

In the few days prior to the vote, the bot sent people between two and three updates a day, but the number went up on election day, when Mila sent an alert roughly every 60 minutes in the space of nine hours.

The feedback on the number of notifications sent varied. People in the US were more engaged with the event and wanted more updates, whereas international audiences were more likely to unsubscribe from the bot, so AJ+ is now thinking about how it could develop a way to determine users' preferences for alerts in the future.

The team will also try to figure out how to make initiatives such as Mila part of the everyday newsroom workflow, and work out how and when notifications can be pushed to AJ+ audiences on Facebook Messenger, WhatsApp, Telegram and similar platforms.

"How do we give people updates they actually want so that there is a propensity to consume this information?

"We need to figure out that mix of curation and automation because at the end of the day, we're still storytellers and we can't outsource our storytelling to a bot.

"That human nature of the story still has to come through," said Hyder.

Free daily newsletter

If you like our news and feature articles, you can sign up to receive our free daily (Mon-Fri) email newsletter (mobile friendly).