AI for Political Campaigns: What It Actually Does, and What It Shouldn't

Most of what gets written about AI for political campaigns is about the frightening part: deepfakes, fake robocalls, doctored ads. That part is real and worth watching. But if you're running for city council or the Assembly, or you run a committee or an advocacy group, the useful question is plainer than that. What does AI actually do for a campaign on an ordinary weekday in the last month before an election?

The short answer is that it takes on the drafting, sorting and research that used to eat a campaign's hours, and a person still decides what goes out. Here is what that looks like, one job at a time.

Digital ads, and the sites they point to

A digital ad program is a pile of small pieces. The same message has to be cut for Facebook, Instagram, Google and X, in several sizes each, every one carrying the committee's disclaimer, and most of them pointing to a simple page where a voter can check the facts behind the ad.

That is where AI earns its keep. It drafts the versions and the copy, resizes the artwork for each platform, builds the page the ad links to, and pulls the week's spending and results into a report a candidate can actually read.

What it doesn't do is pick the message. That stays with the campaign. The platforms have their own rules on top of that: Meta and Google both require political advertisers to disclose realistic images, video or audio made or altered with AI, and California requires its own label on those ads under AB 2355.

Email and fundraising

A lot of campaign email programs run exactly as often as somebody has time to write them, which means they slow down right when they matter most.

AI changes the drafting side of that. It can take the morning's news and the candidate's real positions and have a fundraising draft in the candidate's voice waiting before the finance director's day starts. I wrote about the first time my own machine pitched me a donor email from the morning's news; that moment is what this looks like in practice. The finance director still reads it, fixes what's wrong, and decides whether it goes. The program stops depending on whoever has a free hour, and it goes out from the campaign's own domain to the campaign's own list.

If a donor asks whether the candidate wrote it, there's an honest answer, and it's a good one.

Donor research

Before a candidate calls someone, they should know who they're calling: what the person does, what causes and candidates they've supported, and who they know in common. Much of that sits in public records, and AI can pull it into a one-page profile before the call, instead of after an afternoon of searching.

There's a legal line here worth knowing. Federal law bars using contributor names copied from FEC reports to solicit contributions. Donor research is for understanding the people already in front of you, like the person who handed the candidate a card last night. It is not a way to build a call list out of somebody else's donors.

Government affairs support

This one isn't only for campaigns. Officeholders, advocacy groups and organizations that deal with a legislature or a local board all have the same problem: bills move, agendas post, meetings happen, and the one item that matters to you is buried somewhere in a long packet.

AI reads the packets and the bill text, flags what touches your issue, writes up what happened at the meeting, and drafts the letter or the follow-up. A person decides what to do about any of it, and signs whatever goes out.

Video and content

Short video used to be something only well-funded campaigns could keep up with. AI makes explainers, social clips and ads affordable for a campaign that never had a video budget.

Two rules keep that honest. The video is disclosed as AI wherever the law or the platform asks. And in political ads we don't use AI-generated images of real people; political creative uses real footage and real documents.

What it shouldn't do: answer the phone as you

AI can now answer a campaign's phone line, or place calls, in a voice that sounds human. We can build it, and I advise candidates against it.

Part of the reason is the law. In February 2024 the Federal Communications Commission ruled that AI-generated voices count as "artificial" voices under the federal robocall law, so an AI voice call carries the same consent rules as any other robocall. The bigger reason is the voter. Someone who calls a campaign is trying to reach a person, and if what answers is a machine speaking for the candidate, the first thing they learn about your campaign is that it would rather not talk to them.

Where AI stops

AI is good at the structured, repeatable work. It is not good at judgment. The donor who wants to talk about a vote, the endorser who needs reassurance, the decision about which message leads the last week: those belong to people, and they always will.

It also can't be better than what you give it. It works from the candidate's real positions, accurate notes and a current list. Give it guesses and it will hand you confident-sounding guesses back.

And nothing it produces should reach a voter or a donor without a person saying yes first. That approval step isn't overhead. It's the reason the rest of it is safe to use.

Where to start

Don't start with everything. Start with the one job in your week you'd most like to never do by hand again, get that running well, and add the next one after it.

The technology changes every month. Knowing which parts of a campaign a machine should touch, and which ones it never should, is the part that doesn't. If you want to see what that looks like for your race, here's how we work.

Eric Linder served in the California State Assembly (60th District, 2012 to 2016) and founded AutomatedTeams, a political AI consulting firm for campaigns, committees, PACs and advocacy groups.