Deepfakes and How to Spot Them
A deepfake is media in which a person’s face, voice, or actions have been fabricated or altered using machine learning. The results range from obviously playful to unnervingly convincing. Understanding how they are made and how to question them is a basic piece of modern media literacy.
How they are made
Many face-swap methods train a model on images of two people so it can render one person’s face with the expressions of the other. Voice cloning systems learn the characteristics of a speaker from recordings and then read new text in a similar voice. Other tools generate entire videos or alter lip movements to match different audio. Quality depends on the amount of source material and the care of the creator.
Why they matter
Fabricated clips can be used for fraud, harassment, political manipulation, or reputational damage. Even when not believed, their existence enables a second problem: real evidence can be dismissed as fake. Societies therefore need both skepticism toward suspicious media and confidence in verified material.
Visual clues that sometimes appear
- Odd blinking patterns or fixed gaze
- Blurring or flickering at the hairline, ears, or jaw
- Teeth that look smudged or unnaturally uniform
- Mismatched skin tone between face and neck
- Lighting and shadows that disagree with the scene
- Jewelry, glasses, or hands that distort
- Lip movements that do not quite match speech
Audio clues
Unnatural rhythm, flat emotion, strange breathing, background noise that cuts in and out, or pronunciation that differs from the person’s usual style can hint at synthesis. Phone calls with cloned voices are particularly dangerous because the listener has little else to evaluate.
Why clues are unreliable
Detection by eye is a moving target. As tools improve, old giveaways vanish, and compression on social platforms can add artifacts to genuine videos that look suspicious. Do not treat any single sign as proof either way.
A better approach: verify the source
- Check where it came from. Who first posted it? Is there an original upload?
- Look for corroboration. Are reputable outlets or the person themselves confirming it?
- Search for the original footage. Reverse image or frame search can reveal reuse or alteration.
- Consider the motive. Who benefits if you believe or share it?
- Check context. Date, location, and surrounding events should line up.
- Slow down. Emotional, shocking content is designed to spread quickly.
Protecting against voice scams
Agree on a family or workplace code word for urgent requests. If someone calls asking for money or sensitive information, hang up and call back using a number you already know. Treat unusual urgency as a warning sign.
Tools and standards
Detection software exists but is imperfect and produces both false alarms and misses. Some organizations are developing provenance standards that attach verifiable records to media showing how it was captured and edited. Adoption is still developing.
Responsibility when sharing
Pause before forwarding an alarming clip. Adding “is this real?” to a post still amplifies it. If you find a likely fake targeting someone, report it to the platform rather than spreading it.
If you are targeted
Save evidence, report to the platform, seek support from trusted contacts, and consider legal advice or reporting to authorities, as many places have introduced rules covering harmful synthetic media.