Pandemics, Absentee Ballots, and Chatbots: Lessons in Digitally Engaging the Digitally Unengaged
How I used an AI-chatbot to help voters in West Virginia during the pandemic election.
When I started this final semester of the Social Journalism program, I knew that my community practicum had to combine a bunch of different components that I wasn’t sure how to tie together. It was late August 2020, and I had just completed my fellowship with Mountain State Spotlight. In my last week on the job, we had brought together 21 West Virginians on Zoom to talk about the future of local news in the state. One thing that stood out to me was discussion about the upcoming 2020 general election: how as a brand new, investigative newsroom going to cover it?
A not-so insignificant portion of my summer was spent thinking about community partnerships and how we might distribute content to make sure that all West Virginians got the information that they needed. I had an interest in partnering with libraries to distribute print versions of certain stories, and we decided to make all Mountain State Spotlight stories free to republish under Creative Commons guidelines so local daily or weekly papers could run the investigative stories we produced.
The question of whether print media was a viable delivery option for content in West Virginia was a no-brainer; Internet accessibility and reliability has been a longstanding issue across the state. The technology worked, but the economics were strained, which is why a new digital media outlet even began in 2020. Knowing this, when I received a tip from a local resident about how the absentee ballot application process was changing from how it operated in the primary, I realized there was an opportunity for some engagement journalism.
In the June 9th primary, West Virginia’s 55 county clerks sent absentee ballot applications to every registered voter in the state. 50% of the voters that turned out in that election used absentee ballots, a historic high above the average of 2% traditionally. But after assessing the primary and the opinions of county clerks, the Secretary of State’s office decided that they would deviate from that procedure and rely primarily on a new online absentee ballot application portal.
I figured that if I wanted to fill the information gap between the change in rules and voters’ expectations that things would work the same as they had just a few months ago, I needed to work quickly. Across West Virginia, from October 8th to October 14th, local newspapers were going to publish sample ballots. The total circulation of all of those papers together is nearly 300,000 readers. So my idea was to try and convince the West Virginia Press Association to in addition to the sample ballots, print the absentee ballot application. If voters weren’t going to know that they’d need to apply online, print out a paper form, or request a paper form to be mailed to them from their county clerk, I was going to make sure that they could cut out a clipping from the local paper and mail it in to receive their ballot safely.
Unfortunately, the Press Association was not on my side. My low-tech solution was out, and I needed to be a bit more creative. In the Design and Development course at the Newmark J-School, John Keefe had been teaching us about Google’s DialogFlow software, where you could build simple chat bots to talk to people. Working with John, I figured out that I could build a bot that I’d host on the Google Cloud Platform and distribute via SMS text using an integration service called Twilio. If I couldn’t reach West Virginians via print, I’d have to try and reach them via text message.
By texting “BALLOT” to (304) 413–8889, West Virginians could troubleshoot their absentee ballots with the WV BallotBot. Using natural language processing, the bot would recognize what county a user lived in to connect them with the contact information for their county clerk’s office. It could also provide users with info about upcoming deadlines, connect them with legal help at the 866-OUR-VOTE hotline, and connect them with journalists at Electionland to report issues or abuses.
The tricky thing about a bot like the one I’d designed is that it has to be texted first by a user, so how would I get people to text the bot? I had about two weeks until the absentee ballot application deadline on October 28th to reach as many people as I could. So I did what journalists are supposed to do and pitched a story about the whole ballot debacle to Ken Ward Jr., my former boss and the editor-at-large at Mountain State Spotlight.
On October 15th, I published the story, and with the permission of managing editor Erica Peterson, included the phone number for the bot at the bottom of the story. Mountain State Spotlight distributed the phone number through its social media channels as well. It was my hope that if other publications republished the story, they would include the number and that’s how it might find its way into the physical newspapers. On October 26th, the story was republished in 100 Days in Appalachia, but did not include the phone number. I also posted the phone number and answered questions about voting by absentee ballot in West Virginia specific subreddits, like r/WestVirginia.
I set out to contact as many West Virginians as possible, and between October 15th and November 3rd, the bot exchanged 55 messages with 7 different phone numbers, and connected 1 voter with the Electionland coalition. From Twilio’s chat transcripts, I can tell that the majority of bot users were (304) West Virginia area codes, indicating that the bot was successful in reaching parts of its intended audience.
That’s not to say this project was without faults. I had no way of knowing who the people using my bot were, and I had no way of receiving feedback from people. I launched a project without prototyping that only ran for three weeks, and it depended on people who were online distributing information to people who were offline, but entirely without coordination. If I could go back to the drawing board, I would’ve made sure that the bot integrated a listening phase so I could really get an understanding of what voters needed to know to vote safely, I would’ve sought out solid partnerships for distribution, and I definitely would’ve started earlier.
I remain proud in reaching 7 people, because if the bot helped them vote safely, then it made all the difference. We, as journalists, must remember that every vote matters, and hold fast to using our skills toward empowering communities with information to govern their own lives and choose their own governments.
