MINNEAPOLIS — For years, political campaigns have tried to learn everything they can about voters: What they drive. Where they shop. Whether they own a home, a boat or have a particular interest in equestrian sports.

The data helps campaigns make educated guesses about people’s politics, then decide which voters are worth pursuing and what might persuade them. It is a practice that predates social media, let alone ChatGPT.

But a new generation of artificial intelligence tools is shifting the art of political messaging away from broad-based appeals toward more carefully tailored micro-targeting.

Campaigns can now use AI to sift through research, study audiences and develop political materials in a fraction of the time those tasks once required. Generative technology can produce dozens of variations of an advertisement or appeal, while increasingly sophisticated algorithms help determine which messages reach which audiences.

The advances are arriving as Minnesota campaigns prepare for November’s election, raising questions about how extensively candidates are using the technology and what its growing presence means for voters. Much of the attention has gone to fabricated images and videos. Less visible is the work happening behind the scenes: the research, analysis and decisions about who should see what.

Nora Raube, who directed digital advertising for Sen. Amy Klobuchar’s 2024 reelection campaign, said the fixation on AI-generated content obscures the more consequential ways the technology is entering political work.

Campaigns can use specialized tools to research opponents, identify where potential audiences spend time online and analyze information that would otherwise require hours of manual work. The technology can help political operatives make decisions faster, often without revealing much about how those decisions were reached.

But how effective are the machine-assisted messages campaigns then produce? The research, so far, is mixed.

An old political ambition

The machinery behind modern political targeting has been built over decades.

Minnesota’s voter registration records provide campaigns with names, addresses and voting participation histories, though not which candidates people supported. Private companies supplement those records with commercial information, demographic characteristics and predictions about political preferences.

L2, a national voter-data company, maintains roughly 850 underlying data fields, according to its president, Paul Westcott. That includes details about homeownership, vehicles, consumer habits and political contributions.

Some information proves more useful than others. Westcott said pet ownership generally offers little insight into political behavior, but an interest in equestrian sports might help analysts find their ideal audience.

L2 combines that information with surveys and statistical models to estimate things it cannot know for certain, including whether someone is likely to support a particular candidate.

Westcott calls the process “sophisticated scientific guessing.”

Alvin Zhou, a University of Minnesota researcher who studies AI and strategic communication, described the difference in practical terms.

An advertiser once might have written five versions of a message, then tested which performed best with different audiences. Newer generative tools can produce far more variations from an initial idea, reducing the time and labor required to create them.

The change is in the capacity to generate and refine material, Zhou explained. Systems that once selected among a handful of advertisements can now be paired with tools capable of producing many alternatives. From there, campaigns can flood the airwaves and social media feeds with tailored messaging.

The campaign behind the message

The Minnesota Star Tribune contacted every major statewide campaign to ask how they collect and use voter information. Kevin Poindexter, campaign manager for Republican gubernatorial candidate Lisa Demuth, agreed to an interview.

“Data drives everything that our campaign does, whether that be where we’re going, who we’re talking to, and what we’re talking to them about,” Poindexter said.

He described dividing voters into groups, including those who might be persuaded to support Demuth and those already inclined to support her but who need encouragement to cast a ballot.

Westcott has seen the effects of targeted outreach in his own Pennsylvania community, where campaign door hangers appear on some houses but not others.

“Your household is being picked for a reason,” he said.

For people working in digital politics, AI can make the research behind such decisions more efficient.

Raube said campaigns can use audience-analysis tools to determine whether particular groups are more likely to spend time on YouTube, Reddit or streaming services. Other tools can help research opponents, follow online trends and organize large amounts of information.

But the technology does not always make its methods clear to the people using it.

“To people who don’t work in that data collection, it is very mysterious,” Raube said.

The appeal of AI extends beyond targeting voters. Ron Schutz, the Republican candidate for Minnesota attorney general, said he uses the technology almost every day for research and data analysis.

“I frequently use it for research on a whole range of topics both of a general nature and legal research,” Schutz said. “And I always ask that the answer include source citations so I can double check the answer.”

He also uses AI to combine information from different sources into spreadsheets, a task he said the technology handles particularly well.

Still, the precision of political targeting has limits. Advertising platforms impose restrictions, voter lists do not always match online accounts and campaigns may never know exactly which individuals saw an advertisement.The persuasion question

Timothy Lynch, a political scientist who studies campaigns and elections at the University of St. Thomas, cautioned against assuming that more sophisticated advertising makes voters easier to persuade.

Many Americans have settled political preferences. Others never encounter the advertisements intended for them. Campaigns can improve the efficiency of their outreach without necessarily changing the outcome of a single vote.

Research into AI-generated political persuasion has produced mixed findings, with some experiments showing that AI-written messages can influence attitudes and others suggesting their effectiveness is comparable to ordinary human-written appeals.

The distinction matters because campaigns may find AI valuable even without evidence that it makes their messages more persuasive.

For smaller political operations, the appeal could be especially practical. Candidates for the Legislature, city councils or school boards often have limited staffs and advertising budgets. Research that once required outside help, or simply went undone, may become more accessible through AI-assisted tools.

Lynch noted that local candidates also have smaller constituencies, making direct contact with voters more feasible than in statewide races. The value of new technology may therefore vary considerably depending on the campaign.

There are broader questions about the information reaching voters.

William McGeveran, dean of the University of Minnesota Law School and a privacy scholar, said political targeting carries consequences beyond ads for potato chips or toothpaste.

“The stakes are high when it’s democracy decisions,” he said.

Campaigns have always tried to identify receptive audiences, choose persuasive arguments and avoid wasting money on people unlikely to listen. The new tools do not eliminate the uncertainty involved in those judgments.

A campaign may have hundreds of details about a voter and still be wrong about what that person believes.

But artificial intelligence is making it easier to act on those assumptions, turning research into political material with less time and labor.

For the voter, the result may be indistinguishable from an ordinary campaign advertisement: a familiar argument, a candidate’s photograph, a request for money.

The difference is how much of the work behind it may now be done by a machine.

©2026 The Minnesota Star Tribune. Visit startribune.com. Distributed by Tribune Content Agency, LLC