Political Deepfakes on Social Media and How to Spot Them
A clip of a candidate saying something outrageous lands in your feed two days before an election, and it already has 400,000 views. You cannot tell if it is real, and by the time anyone official responds, the vote may be over.

A clip of a candidate saying something outrageous lands in your feed two days before an election, and it already has 400,000 views. You cannot tell if it is real, and by the time anyone official responds, the vote may be over. Political deepfakes are built for exactly that gap between a fast share and a slow correction.
This guide covers what political deepfakes look like in practice, the cases that have already shaped the rules, what campaigns and platforms do about them, and a practical routine for checking a suspicious clip before you pass it on.
What counts as a political deepfake
A deepfake is audio, video or an image generated or heavily altered by machine learning so that a real person appears to say or do something they did not. In politics, the forms tend to fall into three groups:
- Synthetic audio. A cloned voice reading a script. This is the cheapest and most common type, because a convincing voice clone can be made from a minute or two of public speech.
- Face-swapped or lip-synced video. A real clip with the mouth movements changed to match new words, or a different face mapped onto a body.
- Generated images. A candidate at an event they never attended, or a crowd that never gathered.
It helps to separate these from "cheapfakes": real footage slowed down, cropped or stripped of context. A clip of a politician slowed to 75 percent speed so they sound slurred needs no AI at all, and it often spreads further than a polished synthetic one.
The cases that changed the rules
For years, election deepfakes were mostly a warning in think tank reports. That shifted in 2023 and 2024.
Days before Slovakia's parliamentary election in September 2023, an audio recording circulated in which a party leader appeared to discuss rigging the vote. It surfaced during the country's 48-hour pre-election media moratorium, which made it harder for news outlets and the campaign to respond in public. Fact-checkers flagged signs of synthetic audio, but the recording had already spread widely.
In January 2024, voters in New Hampshire received robocalls using an AI-generated voice that sounded like President Biden, telling them not to vote in the primary and to "save" their vote for November. The fallout was concrete. In February 2024, the Federal Communications Commission ruled that AI-generated voices in robocalls fall under existing rules on artificial voices in the Telephone Consumer Protection Act. The political consultant behind the calls later faced a proposed $6 million FCC fine and state criminal charges.
At the state level, Texas passed one of the first election deepfake laws in 2019, and California followed the same year. By the 2024 cycle, roughly twenty states had laws targeting deceptive synthetic media in elections, most of them requiring a disclosure label within a set window before voting rather than an outright ban.
How platforms and campaigns respond to political deepfakes
Platform policy has settled into a few common patterns. Major video and social platforms now ask uploaders to label realistic synthetic content, and political advertisers are generally required to disclose when an ad contains digitally altered or generated people or events. Content that is likely to mislead voters about when, where or how to vote is usually removed outright. Labeling is the default for most other altered content.
The gap is in organic posts. A paid ad goes through review. A clip reposted by an anonymous account with 30 followers does not, and it can be screen-recorded and reuploaded faster than any moderation queue clears it.
Campaigns have adapted with a playbook that looks a lot like the classic rapid response desk:
- Monitor. Staff or vendors watch for the candidate's name and voice across platforms, especially in the final two weeks.
- Verify internally. Check the candidate's schedule, the original recording if one exists, and whether the location or clothing matches.
- Respond with the real thing. The strongest rebuttal is authentic footage of the actual event, posted quickly.
- Report and document. File platform reports and keep a record, since state laws often require proof of the timing and spread.
Moderation on the campaign's own pages matters too. Fake clips often appear first as replies under a candidate's posts, which is one reason campaign comment moderation has become a core job rather than an afterthought.
How to spot a political deepfake before you share it
Detection tools exist, but none is reliable enough to trust on its own, and most people will not run a clip through one anyway. A short manual routine catches far more:
- Find the source. Who posted it first? A news outlet, the campaign, or an account created last month? Search the exact quote in quotation marks and see who else reported it.
- Look for the full event. Real speeches and interviews have longer versions. If a 20-second clip exists with no wider footage anywhere, be suspicious.
- Check the timing. A shocking clip appearing within 72 hours of polls opening, with no coverage from established outlets, fits the pattern of the cases above.
- Listen carefully. Synthetic audio often has flat breathing, odd pauses between phrases, or a consistent background with no room echo change.
- Watch the mouth and edges. In video, look for blurring around the jaw and teeth, earrings or glasses that flicker, and blinking that seems off.
- Ask what it wants you to do. Messages telling you not to vote, that your polling place moved, or that the election date changed are the highest-harm category. Check with your local election office directly.
None of these checks takes more than two or three minutes. That is usually faster than the time it takes for a correction to catch up.
Why spotting tricks are not the real defense
The standard advice is to train people to spot visual glitches: six fingers, warped backgrounds, strange teeth. I think that advice is aging badly. Each new generation of tools fixes the obvious tells, and audio, the most common political deepfake, has no fingers to count. Teaching people to hunt for artifacts can even backfire, because a clean-looking fake then feels verified.
The more durable skill is source checking. A clip that cannot be traced to an original event, reported by anyone credible, or matched to a candidate's known schedule should be treated as unconfirmed no matter how real it looks. There is a second risk in the other direction too. Once deepfakes are common, real recordings get dismissed as fake. Researchers call this the "liar's dividend," and it is one reason authentic, timestamped footage from campaigns and journalists matters more now, not less.
Deepfakes also borrow the same sharing mechanics that make political memes on social media travel: humor, outrage and a sense of being in on something. The format is new. The psychology is old.
What voters can do in the final week
The highest-risk window is the last seven days, when there is the least time to correct anything. A few habits help:
- Get voting logistics only from your state or county election office website or official mailers.
- Follow the candidates' verified accounts, so you know what their real channels look like.
- Hold off sharing any shocking clip for a few hours. Real news rarely disappears in that time; fakes often get debunked.
- Be wary of paid creators pushing a sudden clip. Our explainer on how political influencer campaigns work covers the disclosure rules to look for.
Frequently asked questions
Are political deepfakes illegal?
It depends on where you are. In the United States, about twenty states have election deepfake laws, and most require a disclosure label near an election rather than banning the content outright. AI-voiced robocalls are covered by federal rules after the FCC's 2024 ruling.
Can AI detection tools tell if a video is fake?
They can help, but they are not reliable enough on their own. Detection accuracy drops when clips are compressed, cropped or reuploaded, which is exactly what happens on social media. Use tools as one signal alongside source checking.
What should I do if I see a deepfake of a candidate?
Do not share it, even to criticize it, since that spreads it further. Report it on the platform using the misleading or manipulated media option. If it gives false voting information, tell your local election office as well.
More in Social Media and Politics
Social Media and Politics
Campaign Social Media Toolkits and Why Volunteers Skip Them
A congressional campaign I talked to in October had a 41 megabyte folder on Google Drive called Supporter Toolkit.
Social Media and Politics
Campaign Discord Servers and How Volunteers Organize
A state house campaign I followed last spring set up a Discord server on a Tuesday, posted the invite link on the candidate's...
Social Media and Politics
Campaign Reddit AMAs and What Candidates Get Wrong
A candidate agrees to a Reddit AMA, the staff post a verification photo, and within twenty minutes the thread has two thousand...