AI Election Laws Cover the Ad. The Doorstep Is a Blind Spot.

Alex Bond is 29 and lives in Pennsylvania’s 10th Congressional District. When two canvassers from Swing Left came to his door, he told them he thought AI was “terrible.” What Bond did not know, as The New York Times reported, is that his comments went into an app that synthesizes hundreds of conversations like his and sends what it finds back to the people running campaigns.

“Everything a person is saying is a data point,” Violet Kopp, Swing Left’s organizing program manager for the East Coast, told the Times.

Now count the rules written to govern AI in this election. Public Citizen tracks election deepfake legislation state by state. Axios, citing the National Conference of State Legislatures, reports 29 laws in effect, with California’s and Hawaii’s permanently enjoined on First Amendment grounds. These statutes govern synthetic political messages that get published or sent to voters. A fabricated video of an opponent, an unlabeled synthetic ad, a cloned voice on a robocall. I could not find one that requires a canvasser to tell you that the conversation you are having is going into an AI system.

Ground Truth sits inside a broader shift. In a small, consultant-heavy May survey of 68 campaign professionals split roughly evenly between Republicans and Democrats, 87% said they used AI daily or several times a week.

Swing Left announced the program in April. Ground Truth pairs volunteer conversations at the door with an app that captures open-ended notes and AI-assisted analysis that looks for patterns across them. The app was developed with a startup called Everybody Techworks and lets volunteers “record open-ended notes in voters’ own words.” NOTUS reported that volunteers take handwritten notes during a 10- to 15-minute conversation and enter their observations into the app once they leave.

The instinct behind all this is one I think is right. Most campaign canvassing goes to people who already vote, the ones organizers call triple primes. Ground Truth sends volunteers to knock every door in a district, including the ones traditional target lists leave off. Swing Left executive director Yasmin Radjy told NBC News that Democrats generally focus too much on targeting specific groups, and that her approach is that you should knock every door.

Everybody Techworks said that its app can store information about everyone a user engages with, including people who do not appear in a voter file or contact database. Its privacy policy puts the organizations buying the software in charge of the data and makes them responsible for whatever disclosures the law requires of them.

For decades the complaint about campaign targeting was that it skipped these households entirely. Ground Truth is answering that complaint, and I want to be fair about it. Knocking every door is better organizing than knocking only the reliable ones. But the answer arrives with a pipeline attached. Ground Truth captures notes from conversations with the people traditional target lists leave off, and what comes out of those notes travels up to campaigns and state parties before anything travels back down to the voters.

Bond has less to check than you might assume. Pennsylvania’s AI disclosure bill, SB 1332, is sitting in the Senate State Government committee, so his state has no operative requirement to disclose AI use in political advertising. And Swing Left’s privacy policy governs what happens when you fill out a form on swingleft.org. I looked for a published policy covering what happens to what a person says at their own front door and could not find one.

Europe covers both ends of this. Its general data protection law already governs how political organizations handle personal data, and it treats anything revealing a person’s political opinions as sensitive. Ireland’s data protection authority spells out what that means at the door, telling campaigners that notes taken while canvassing are a data protection matter and that they need a retention period. The newer political advertising regulation goes further at the targeting stage. Since October 2025, using personal data to target or deliver online political advertising has generally been allowed only where the organization collected that data from the person directly and obtained separate, explicit consent for it. Profiling for that purpose using sensitive data such as political opinions is prohibited outright.

I trained in public diplomacy at USC, where the case that a government should listen to a population before messaging to it is the foundational argument in the field. It has a second half that gets skipped. Listening is a relationship, and relationships have terms. When an institution gathers what people say and the people saying it never learn where it went, the word for that is collection.

There is a version of this that would not require anyone to stop canvassing. A volunteer could say at the door that they will write up notes from the conversation and that the notes may be analyzed using AI. Swing Left could publish how long the notes are kept, whether a record exists for each person, and whether outside AI vendors can use any of it. The first would add one sentence to a canvass. The rest gets published off the doorstep, and Everybody Techworks already puts its customers in charge of the data.

November will be the broadest test these deepfake laws have faced, though the courts have been testing them for a while already. Michigan Rep. Penelope Tsernoglou, who passed two of them in 2023, told the Arizona Capitol Times that the laws will be tested and stretched to their limits this year. She is probably right, and that test will tell the country how well it regulates what campaigns broadcast. It will say nothing about what campaigns take home.