When voters ask a chatbot about a candidate, there's a good chance the answer comes from Ballotpedia. Research from the Center for Campaign Innovation found it cited in about 30% of chatbot answers about candidates, and in 42% of Google Gemini's. In this episode of the Campaign Trend Podcast, Eric Wilson talks with Leslie Graves, founder and CEO of Ballotpedia, about why.
Graves says Ballotpedia never set out to win over AI. She learned about the citations from a reporter only a couple of months ago. Her explanation comes down to three things: every article follows the same structure, so models can read a ballot measure in Arkansas the same way they read one in California; the Candidate Connection survey provides original, consistently ordered content found nowhere else; and the site is text, which is easier for a model to consume than video.
That survey is also where candidates have a voice on a staff-written site. About 35 to 40% of candidates from the state legislature through the U.S. Senate fill it out this cycle, and Graves says she can tell when consultants have told a whole slate to do it. Eric notes that counterparts on the left have been enthusiastic about the survey, while he struggles to get Republican candidates to publish even a basic issues page.
Graves explains why that's a risk. Voters asking an AI to compare candidates want to know where they stand on three or four top issues. If a campaign's website is silent, the model may build its answer from something like an old Reddit comment from a supporter. Graves also shares that Ballotpedia's traffic is up 20% over 2024 despite industry fears about zero-click search, and that page views on a race typically run 20 to 30% of the votes eventually cast in it.
The conversation also covers who actually reads Ballotpedia, why the site is adding coverage of roughly 44,000 local candidates this year, and why a universal AI for election information across 3,200 county websites is still a long way off. On writing for machines, Graves's advice is to skip trying to game the algorithm and write for the human reader, since the algorithms are trying to anticipate what people want.