Democrats insist their campaigns need to take AI chatbots seriously. And now some think they have early answers for how to actually make that happen.
A recent study, commissioned by Democratic-aligned public relations firm Orchestra and shared first with NOTUS, analyzed how voters use large language models to extract information about candidates and issues — a new digital frontier where the parties will battle for primacy. The research, conducted by the AI-marketing platform Evertune, estimated that chatbots are asked 650,000 prompts about politics weekly, primarily through ChatGPT and Google AI.
Chatbots, and how voters use them, are injecting a complicating dimension into campaign digital strategy this year. It’s another way to target voters, strategists from both parties acknowledge, and they’re racing to figure out how to influence the information those AI-generated bots spit out. In 2024, early iterations of these chatbots pumped out false and misleading data on voting, raising more warnings for both parties.
“The best campaigns are usually the ones that have the best sense of how the electorate is living their lives and receiving and sharing content and information at that moment and do their best to mirror it in their own resourcing and approach,” said David Plouffe, who advised the project. “LLM now joins the stage with Tiktok, Instagram and YouTube as a core challenge and opportunity for campaigns to navigate and master.”
Trending
Chatbots won’t forecast a winner or endorse a candidate, the study found when it tested those questions. But when asked about the candidates generally, their platforms and campaign promises, that’s when the quality of a candidates’ digital footprint became more clear.
In one test case, the researchers compared the AI-generated answers for the policy positions of Democratic Georgia Sen. Jon Ossoff and GOP Rep. Mike Collins for their hotly contested Senate race, where it found Ossoff’s easy-to-read and well-annotated campaign issues pages gave him the upper hand in generating citations. “Answer engines repeatedly return to Ossoff pages,” the research found, because they were “detailed” and “clearly labeled.”
But Republicans hold an advantage on their economic storytelling through chatbots, the study found — a key issue for the midterms, which have been dominated by cost-of-living concerns. Through repeated engagement with the chatbots, the Republican economic pitch, centered on producing more energy coupled with reducing the cost of government, is more coherent than Democrats’, the study said. Democratic arguments are “fragmented,” as the models “do not consistently assemble them into the same overarching account,” the study said.
“When a party frames an issue consistently across campaign content, official records and media coverage, models are more likely to reproduce it coherently,” the study said. “If the response sounds scattered, the model may be reflecting deeper communications problems: the lack of a consistent message, the lack of a sufficient paper trail supporting that message — or both.”
Out of the AI-generated answers it tested, the study found chatbots predominately cite news coverage, but the “strongest URL” references were explainers, both from news outlets and partisan sources, like party platforms and campaign websites. The study urged campaigns to clarify their own candidates’ digital presence on issue-specific topics as a way to influence those answers.
It also found that YouTube clips of previous interviews or coverage — once seen as a one-off news cycle hit that doesn’t live beyond a single day — can take on new life through AI-generated answers through chatbots, which frequently cited those videos.
The study laid out other advice to campaigns: Treat campaign websites as resources, not just fundraising hubs and create content that “starts with a problem,” the same way a voter interacts with a chatbot.
Author
