Who needs to detect election deepfakes?
Anyone whose job or reach means a fake could spread through them needs a routine for checking election content. That includes election officials, campaign staff of every party, fact-checkers, political reporters, civil society monitors, platform moderators, teachers and ordinary voters who share news with family.
- A video of a candidate saying something shocking, posted by an account with no history.
- A call or voice message in a candidate’s or official’s voice giving wrong information about voting.
- Images of incidents at polling stations, crowds at rallies or ballots being destroyed.
- A news-looking website that appears weeks before an election and publishes only political stories.
- Real footage that is cut, slowed down or given a false caption, sometimes called a cheapfake.
What does a political deepfake look like in practice?
A political deepfake is usually short, emotional and timed to spread before anyone can check it. Before the New Hampshire presidential primary in January 2024, voters received robocalls using an AI-generated voice that imitated President Biden and urged them not to vote in the primary.
Regulators responded. On February 8, 2024, the US Federal Communications Commission issued a declaratory ruling that AI-generated voices are “artificial” under the Telephone Consumer Protection Act, which brings robocalls using voice cloning under that law’s rules. In the European Union, the AI Act’s transparency obligations under Article 50, which include disclosing deepfakes, have applied since August 2, 2026.
Rules like these help after the fact. In the hours when a fake is spreading, verification habits matter most.
What are the risks of AI misinformation, and of false alarms?
Missing a fake can mislead voters about a candidate’s words, or about when and where to vote. Procedural misinformation is the most damaging kind, because a voter who misses an election day cannot get that vote back.
False alarms are harmful too. Labeling authentic footage as AI misleads voters as much as sharing a fake, and it gives anyone caught on camera an easy way to deny what they said. Fact-checkers keep their credibility across the political spectrum only by applying the same standard to every side and publishing their method.
How to verify a political video before you share or report it
Work outward from the clip to its source and to independent records of the event. This routine fits a newsroom desk or a voter checking a forwarded video.
- 01Pause before sharing. Strong emotion and urgency are what a fake is designed to produce.
- 02Find the earliest upload by searching key frames with a reverse image search and checking who posted first.
- 03Check official channels: the candidate’s or party’s accounts, the event organizer, and the election authority for anything about voting procedures.
- 04Look for other recordings of the same event from different angles, livestreams or press pool footage.
- 05Check the facts inside the clip: was the person really at that place on that date, does the setting match?
- 06Run the best copy through deepfake video detection and read the marked moments and signals.
- 07Publish or share only what you can support, and say clearly what you could not confirm.
AI Video & Deepfake Detector · Free plan
Paste a link to a public political video and see suspicious moments marked on a timeline.
Fact check AI video on social media
Social platforms recompress video and strip metadata, which weakens both provenance checks and detection. Look for the original upload rather than a re-share, and watch the mouth, teeth and the edges of the face during head turns. The guides on how to tell if a video is AI on social media and how to spot a deepfake video cover the visual checks.
Images of polling incidents and crowds
A photo of ballots in a dumpster or a packed rally can be generated, or it can be real but from another place or year. Check signs, uniforms, vehicles and the ballot design against the jurisdiction it claims to show. Run a reverse image search for older copies, then screen the image with the AI image detector, which checks content credentials first. Local election officials can often confirm or deny an incident within hours.
Election robocalls and voice messages
If a call in a familiar political voice tells you to vote on a different day, by a different method or not at all, treat it as false until the election authority confirms it. Save the recording if you can, check it with AI voice detection, and report it to your election authority.
News-looking sites that spread AI misinformation
Check who runs the site, how old it is and whether its articles and images look generated. Sites built quickly for an election often have a recent domain, no named editors, no corrections policy and articles that recycle other outlets’ reporting.
The AI website checker explores up to 20 public pages, runs texts and images through the matching detectors and reports domain age and Google Safe Browsing results where available. The guide on how to tell if a website is AI-generated lists the checks you can do by hand.
How can election offices, campaigns and newsrooms prepare?
Prepare before the campaign, because a fake spreads fastest in the first hours and a prepared team can respond while it still matters. The same steps work for an election office, a campaign of any party or a newsroom.
- One official source: publish voting dates, places and methods on a single page that voters learn to trust, and point to it in every message.
- Known channels: list your official accounts and phone numbers publicly, so voters can check whether a call or post really came from you.
- A response plan: decide in advance who verifies a suspected fake, who approves a public statement and how fast.
- Archive your own media: keep originals of speeches and events so you can show what was really said.
- Prebunking: tell voters ahead of time that fake calls and clips about voting may circulate, and what you will never ask them to do.
- A shared standard: fact-checkers and moderators apply the same checks to content about every candidate, and publish their method.
Which AnyFact detectors fit election verification, and what do they return?
Four detectors cover election media. Each returns a verdict of probably human, uncertain or probably AI (websites can also be mixed), the signals behind it, a plain-language explanation in English, French or Arabic, the limits of the method and a PDF report.
| Content | Detector | What it returns |
|---|---|---|
| Speech or interview clip | Video (up to 3 minutes, or a public video link) | Scores for one frame per second on a timeline, a face-swap check when the speaker faces the camera, and the audio checked for cloning |
| Robocall or voice message | Voice (up to 5 minutes) | Segment-by-segment scores for a synthetic voice, plus a transcript of what is said |
| Photo of an event or incident | Photo | Content credentials and metadata, a classifier verdict, and anomalies in hands, lettering, shadows and reflections |
| News-looking site | Website | Texts and images from up to 20 pages checked, with domain age via RDAP and Google Safe Browsing where available |
AI Voice Detector · Free plan
Check a political voice message or robocall recording for signs of cloning, segment by segment.
What AI detection cannot decide in an election
Detection cannot tell you whether a claim is true, whether a clip is shown in context or who made it. It answers whether media shows signs of generation, and it can be wrong in both directions, especially on short, compressed re-uploads.
AnyFact takes no political position and applies the same method to every file. It has no public API yet, so it fits case-by-case verification rather than monitoring every post during a campaign.
Detection also says nothing about intent. A clearly labeled parody made with AI and a fake built to mislead voters can get the same verdict. Whether to flag, label, debunk or leave content alone is a judgment about context and harm that people have to make.
Frequently asked questions
How can I tell if a political video is a deepfake?
Find the original source, check the candidate’s official channels, look for other recordings of the same event and watch for visual glitches around the face and mouth. A video detector adds a signal, but source checks usually settle it.
Are AI robocalls legal in the US?
In February 2024 the FCC ruled that AI-generated voices count as “artificial” under the Telephone Consumer Protection Act, so robocalls using voice cloning fall under that law’s restrictions on artificial-voice calls. Rules on election communications also vary by state.
What should I do if I get a deepfake about voting?
Do not share it. Check the official election authority’s website for voting dates, places and methods, keep a copy of the message, and report it to the platform and to your election authority.
Can a real video be wrongly called a deepfake?
Yes. Detectors make mistakes on compressed and short clips, and people sometimes call authentic footage fake to avoid accountability. That is why detection should always be combined with source checks.
Does AnyFact favor any party or candidate?
No. AnyFact applies the same detectors and method to every file it analyzes, whatever its political content, and explains the signals behind each verdict so you can judge the result yourself.