1. Who is the source and what's the context?
Before you even analyse the pixels, check the video’s origin. Who posted it? Is it a reputable news organisation, an official government channel, or an anonymous account with no history? Misinformation often thrives on new or unverified accounts. Take
a moment to investigate the source. Furthermore, consider the context. A quick search can often reveal if a video is old but being presented as new, or if the situation it depicts is plausible. For example, if a video shows a public figure endorsing a dubious financial product, a quick check of their official social media channels or news reports will likely debunk it. Verification tools like InVID-WeVerify can help by extracting keyframes from a video and running a reverse image search to see where else it has appeared online, often revealing its original context.
2. Are there any visual glitches or inconsistencies?
AI video generators are getting better, but they still make mistakes, especially with fine details. Pay close attention to the person in the video. Do they blink naturally, or is it too often or not at all? AI models often struggle with creating realistic eye movements. Look at the edges of the person, especially around their hair, where you might see a shimmering or blurry effect as the fake element is overlaid onto the background. Hands and fingers are another common giveaway; look for an incorrect number of fingers, unnatural bending, or hands that seem to merge with objects they are holding. Also, check for consistency in appearance. Do accessories like glasses or jewellery appear and disappear between frames? Does clothing or hair style subtly change without a cut in the video? These small errors, often called 'drift,' happen because the AI doesn't have a persistent memory of the subject from one frame to the next.
3. Do the lighting and shadows make sense?
One of the most difficult things for AI to replicate accurately is the physics of light. Scrutinise the lighting on the subject's face and see if it matches the environment they are in. For instance, if the person is supposedly outdoors on a sunny day, but their face is lit from an angle that suggests an indoor lamp, that's a major red flag. Shadows can also reveal a fake. Do they fall in the correct direction based on the light sources in the scene? Are they present at all? A generated face might be pasted into a scene without the corresponding shadow it should cast, making it look like it's floating. Unnatural skin texture, which may appear too smooth or have a strange sheen, can also be a sign that the lighting is artificial.
4. Does the audio match the visuals and sound natural?
Often, your ears can pick up on a fake before your eyes do. One of the clearest signs of a deepfake is poor lip-syncing, where the person's mouth movements are not perfectly aligned with the words being spoken. Beyond syncing, listen to the quality and tone of the voice itself. Does it sound robotic, flat, or lack the normal emotional inflections of human speech? AI-generated audio might have an unnatural rhythm or pacing, and it often lacks the small, subtle sounds of a real human speaking, like breaths between sentences or slight mouth noises. Sometimes, the audio is unnaturally clean, with no background noise, which can be suspicious if the video is set in a busy environment. A voice that doesn't seem to match the person's age or physical presence is another warning sign.
5. What emotion is the video trying to make me feel?
This final question moves beyond technical analysis to critical thinking. Maliciously manipulated media is almost always designed to provoke a strong emotional reaction. Is the video trying to make you angry, scared, or outraged? These emotions can bypass our skepticism and encourage us to share content impulsively without verification. Scammers use deepfakes to create a sense of urgency or familiarity, like a fake video call from a loved one in trouble to solicit money. If a video prompts an intense emotional response, it's a good signal to pause, take a breath, and apply the other four questions. The goal of much disinformation is not just to deceive, but to erode trust in what we see and hear. By asking what the creator wants you to feel and do, you can better protect yourself from being manipulated.
