Presidential campaigns in 2024 were building voter profiles with 500 to 1,000 individual data points per donor, using machine learning to predict who would give, when, and how much, according to the Brennan Center for Justice. That level of targeting didn't exist a decade ago. Today it's operational standard.
AI is embedded in four core campaign functions: voter targeting, fundraising, ad optimization, and voter contact. Both parties use it. The tools are available to whoever can afford them, and the price is falling.
On the fundraising side, the results are measurable. Tech for Campaigns, tracked by the Harvard Ash Center, found that AI cut the time needed to draft fundraising emails by one-third. For a campaign sending hundreds of messages a week, that's not a marginal efficiency gain. It frees staff for work that still requires human judgment, and it lets smaller campaigns compete with larger ones on volume.
Voter contact is where the numbers get more complicated. A joint study from Cornell and MIT found that AI-powered chatbots shifted voter opinions by an average of 3.9 percentage points, making them roughly four times more effective than traditional political advertising. That sounds like a breakthrough. The problem is what the same study also found: the chatbot models most effective at persuasion were also the least accurate. The most convincing AI wasn't the most truthful.
That tension sits at the center of a larger concern: synthetic content.
The 2024 cycle produced documented, high-profile examples from both parties. A deepfake robocall used President Biden's voice to discourage New Hampshire Democrats from voting in the primary. The FCC ultimately imposed a $6 million fine, according to NPR. On the other side of the aisle, former President Trump posted AI-generated images depicting Taylor Swift as endorsing him, and Elon Musk shared an AI-manipulated clip of Kamala Harris's voice, as reported by NPR and Rappler.
These incidents drew attention and condemnation. But the feared wave of sophisticated AI deepfakes many predicted for 2024 didn't materialize at scale. The Harvard Ash Center concluded that synthetic media remained a secondary concern relative to pre-cycle expectations. Research from the Knight Institute found that "cheap fakes," low-tech video edits and out-of-context clips, were used seven times more often than genuine AI deepfakes during the cycle.
The more persistent infrastructure problem is at the local news level. There are now an estimated 1,200 "pink slime" sites, outlets designed to look like local news but built to push political content with little accountability or transparency, according to Intel 471. These sites now outnumber legitimate local papers in many markets. AI lowers the cost of running one to near zero.
None of this means AI is making campaigns worse in every dimension. It is making them faster, more targeted, and more personalized. The question is whether accuracy and transparency can keep pace with efficiency. When they don't, the voter is left sorting through the difference.
Part 2 of this series covers the regulatory fight over AI in politics, why federal and state governments are pulling in opposite directions, and what voters from both parties say they actually want done about it.