For a while, it looked like AI was going to blow up campaign politics in 2024. Powerful new tools, new persuasion techniques, less policing of social-media platforms, all were leading up to a landscape transformed, maybe dangerously so. With less than three months before the election, despite a handful of controversies and deepfake scares, it hasn’t quite panned out that way. That’s not to say AI has been irrelevant to politics. In fact, it’s becoming more integrated into the ground-level practice of campaigning and persuasion. But its effects have been more subtle, less of a revolution and more of a nudge in the direction things were already heading. Here’s how it looks in practice: On a marketing call towards the end of July, Alexander Jones, a senior survey analyst at the AI-driven marketing firm Resonate, enthusiastically broke down the voter landscape as the company’s AI models saw it. On a slide listing 10 voter segments, Jones elaborated on one that seemed particularly persuadable, which his firm’s algorithm had sifted out. “Cyber Crusaders… socially conservative, but fiscally liberal,” he said. “This is a male dominated group, often in white collar jobs. They are majority-minority and majority-millennial and very religious.” “This is the persuasion target going into Election Day. They are high turnout and deeply divided. They get their name Cyber Crusaders due to their religiosity, as well as being the most trusting of online ads both on Facebook and YouTube.” Resonate is one of several firms offering AI-driven insights about voters to political campaigns in a bonkers election year. It promises to use advancements in machine learning techniques like neural networks and clustering to make useful sense of millions of voter profiles. But if the slideshow results sound familiar to anyone who remembers “soccer moms” or the myriad other sliced-and-diced bits of the electorate that fascinate pundits — well, you have a point. Experts see these newer uses of AI as very much part of a long-evolving spectrum. I called my colleague Sasha Issenberg, a POLITICO editor in California who wrote the definitive book on the early growth of data-driven politics. “There’s nothing conceptually new about this,” he told me. “About 20 years ago, the availability of consumer data, changes in database architecture and advances in statistical modeling made it possible for campaigns for the first time to have predictive insights about individual voters, as opposed to treating them as parts of large demographic categories or geographic zones.” Still, the evolution of AI — not powerful new large language models like ChatGPT, but good old-fashioned machine learning — as a tool for political microtargeting means the field is slowly getting more sophisticated in what it can do. The minute breakdown of the voter landscape Resonate presented in July was built on a vast tranche of data pooled from offline sources, consumer behavioral profiles and the company’s own identity graphs. Resonate’s chief marketing officer Ericka McCoy said primary data set consists of roughly 250 million privacy-scrubbed profiles of real people that have tens of thousands of data points attached to them. The company queries those personas in real-time to gauge voter responses to the news. Though it all might seem a little dystopian, or at least claustrophobic, there are actually potential upsides to the public sphere. For instance, the dropping cost of more insightful, user-friendly AI tools could help bridge the gap between the haves and have-nots in political campaigning. “A type of profiling that might have been available to a Senate campaign could become available to a city council candidate,” Issenberg said. “If [AI] allows non-technologically skilled campaign staffers to fully use microtargeted data, that's a big deal on campaigns.” In a frenzied election year, with a new presidential candidate suddenly on the ballot and PACs raising and spending money in the billions, any edge to target undecided voters could be decisive. “In politics today, most everybody is already reflexively one belonging to one party or the other — and so those people are not going to be necessarily persuaded to change sides,” said Prof. Stephen Farnsworth, a professor of political science at the University of Mary Washington. “But for people who are at that sliver of voters in the middle, any technology that is more effective at connecting with them can be helpful.” It’s not just the big-ticket presidential elections. Resonate's McCoy said smaller political campaigns fighting down-ballot races had also contracted Resonate’s voter targeting services. “We’re at every level,” she said. She also said the company was agnostic about which party or candidate bought its services — Democratic, Republican, independent. “We're like any other business — we're in the business to generate revenue for our shareholders,” McCoy told POLITICO. The company did not disclose which political campaigns were already using its insights. One catch: When it comes to actually persuading voters, the efficacy of microtargeting is difficult to determine. Some ad targeting is definitely better no ad targeting, according to a 2023 MIT study — particularly for swaying people’s opinions on a single policy issue. But the same study suggests that tailoring ads to people based on multiple characteristics instead of sticking to just one wouldn’t necessarily bring additional voters to the fold. “Engagement metrics aren't always the same thing as persuasion value, and so that's where the challenge is for anyone who's using these optimizations and AI,’ said Larry Huynh, president of the American Association of Political Consultants. Just because lots of people are responding to a political ad doesn’t mean the ad seller is changing hearts and minds. They could just be resonating with their core audience — doing what politicians have done for decades, whatever the technology.
| 
