This isn't about being a cynic. It’s about being a skeptic.
Your Lying Eyes
There used to be a thing called “seeing is believing.” You watched it happen. You trusted your eyes. Your eyes had never lied to you before, at least not intentionally.
That’s over now.
What a Deepfake Actually Is
Deepfakes are AI-generated or AI-manipulated videos, images, and audio that make real people appear to say or do things they never said or did. The technology isn’t new. What’s new is that it got cheap, fast, and available to anyone with a laptop and a grudge.
A few years ago, making a convincing deepfake required a film studio budget and a team of specialists. Today, voice cloning apps are free. A decent gaming PC can generate a realistic fake video in minutes. And the tools only need a few seconds of real audio to clone someone’s voice well enough to fool most people.
That’s the part nobody in the panicked social media posts bothers to mention. You don’t need to be a nation-state or a tech genius to do this. You just need an afternoon and a reason to want to.
The Biden Robocall Was the Canary
In January 2024, thousands of New Hampshire voters got a phone call. The voice sounded exactly like President Biden. It told them not to bother voting in the Democratic primary, that their vote would count more in November. The voice was fake. A Democratic political operative had paid a magician he’d met through a mutual acquaintance to build the audio clip using a voice-cloning app. Total cost: about $500.
The FCC banned AI-generated robocalls in response. The operative claimed he did it to raise awareness. The League of Women Voters sued him. And between 5,000 and 25,000 people heard a fake version of the President of the United States tell them to stay home.
That’s the real headline. Not the fraud. Not the lawsuit. The headline is that it cost five hundred dollars and an afternoon, and it worked well enough to require a federal regulatory response.
The Thing Nobody Wants You to Know
Here’s the part that should actually make you feel a little better: in the 2024 election cycle, cheap fakes outperformed deepfakes seven to one.
Cheap fakes are the low-tech cousins. A real video clip stripped of context. A speech slowed down to make someone sound drunk. A jump cut that removes a single word and reverses the meaning of a sentence. Old footage with a new caption. The News Literacy Project tracked election misinformation throughout 2024 and found that this kind of basic manipulation was used seven times more often than AI-generated content.
Why? Because it’s faster. It’s deniable. And it works just as well on people who already want to believe it. The AI revolution in disinformation is real, but the boring old “take it out of context” trick still does most of the heavy lifting. The danger with deepfakes isn’t that they’ve taken over. It’s that they’re getting good enough, fast enough, that they soon won’t need to.
It’s Not Just Politicians
Public figures are the obvious targets, but the technology doesn’t care about your social media following. Scammers have used voice clones to impersonate family members in “grandparent scams,” calling elderly relatives pretending to be a grandson in legal trouble and needing cash immediately. In 2025, Americans reported $5 million in losses to these distress scams alone. That number is what got reported. The actual number is higher.
The difference between a political deepfake and a family scam deepfake is just the motivation. The tool is the same.
How to Actually Check Before You Panic
You don’t need a forensics lab. You need to slow down for about ninety seconds.
Watch the edges. Where the face meets the hair or the collar is where AI still tends to fall apart. Look for blurring, flickering, or that slightly waxy quality where the skin doesn’t quite behave like skin. Ears are a particular weak spot for current deepfake models.
Listen for the breath. Humans breathe between sentences. AI-generated audio often doesn’t. It also tends to be slightly too smooth, with an uncanny evenness that real human speech never has. If the voice never takes a breath in a thirty-second clip, that’s worth flagging.
Check if it’s been covered. Before sharing anything that seems explosive, run a quick search to see if any news organization has reported on it. If it’s real and significant, it’ll be covered somewhere. If the only place it exists is the account that sent it to you, that’s your answer.
Go to the primary source. The White House posts transcripts. Campaigns post schedules. If a politician supposedly said something in a speech, the speech exists somewhere. Find it.
Use the tools that already exist. PolitiFact, FactCheck.org, Reuters Fact Check, and Snopes all actively track and debunk synthetic media claims. It takes less time to search one of those than it does to write an angry comment about the thing you just watched.
The Real Problem Isn’t the Fakes
The actual damage deepfakes do isn’t just convincing people of things that didn’t happen. It’s giving everyone permission to dismiss things that did. If any video can be faked, then any video can be called fake. That’s the play. Create enough noise that nothing is verifiable, and then real evidence stops mattering.
Researchers call this the “liar’s dividend.” The technology doesn’t even need to fool you. It just needs to make you uncertain enough that you don’t trust anything.
The cure for that isn’t a better detection algorithm. It’s the same thing it’s always been. Slow down. Check the source. Ask who benefits from you believing this right now. Those tools have existed since before any of us had a smartphone. The AI just made using them more urgent.
References
- How to Detect Deepfakes in 2026 — missioncloud.com/blog/how-to-detect-deepfakes-in-2026
- Democratic operative admits to commissioning fake Biden robocall that used AI — nbcnews.com/politics/2024-election/democratic-operative-admits-commissioning-fake-biden-robocall-used-ai-rcna140402
- New Hampshire voters sue Biden deepfake robocall creators — nbcnews.com/politics/2024-election/new-hampshire-voters-sue-biden-deepfake-robocall-creators-rcna143662
- We Looked at 78 Election Deepfakes. Political Misinformation Is Not an AI Problem — knightcolumbia.org/blog/we-looked-at-78-election-deepfakes-political-misinformation-is-not-an-ai-problem
- How to Spot Deepfake Scams and Protect Yourself From AI Fraud — ncoa.org/article/understanding-deepfakes-what-older-adults-need-to-know
- How to Detect Deepfake Videos Like a Fact-Checker — politifact.com/article/2023/apr/19/how-to-detect-deepfake-videos-like-a-fact-checker
- Deepfakes and the Crisis of Knowing — unesco.org/en/articles/deepfakes-and-crisis-knowing
