1. Scrutinise the Face and Eyes
The face is often where deepfakes reveal themselves. Pay close attention to the eyes; AI models often struggle to replicate natural blinking, which can result in a person blinking too much, too little, or at a weirdly regular pace. Unnatural facial expressions
that don't match the tone of what's being said are another red flag. Look for inconsistencies around the edges of the face, like a flickering or blurry jawline, or hair that seems to shift unnaturally. Even skin texture can be a clue; AI-generated skin may appear too smooth and plastic-like, lacking the normal pores and blemishes of a real person. While technology is improving, these subtle errors in facial rendering are still common giveaways. By pausing the video and looking closely, these small imperfections can become more obvious.
2. Watch for Unnatural Body Movement
Beyond the face, the overall movement of the person can signal a fake. AI often has trouble with the complex physics of the human body. Look for a head and body that seem disconnected or move awkwardly, especially during quick turns. Posture might appear overly rigid or, conversely, too smooth and lacking the minor, imperfect corrections real people constantly make. One of the most notorious weak spots for AI generators is rendering hands. Examine them closely for anything that looks off, such as extra or missing fingers, unnatural bending, or hands that seem to pass through objects instead of gripping them properly. Background objects can also be a telltale sign; watch for furniture that seems to warp or patterns on walls that shift as the camera moves.
3. Listen for Audio Flaws
Your ears can be just as important as your eyes when detecting a deepfake. AI-cloned voices have become incredibly sophisticated, but they aren't perfect. Listen for speech that sounds robotic, flat, or has an unnatural, metronome-like cadence. Real human speech is filled with small imperfections like filler words and breaths, which are often absent in AI-generated audio. Another clue is a mismatch between the audio and the visual. Check if the lip movements are perfectly synced with the words being spoken; sometimes there is a noticeable lag. Pay attention to hard consonants like 'p' and 'b' that require the lips to close fully—if they don't, it’s suspicious. Finally, consider the background noise. A video filmed in a busy setting that sounds perfectly clean and quiet may indicate the audio was synthetically created.
4. Investigate the Source and Context
This might be the most crucial step of all. Before you even analyse the pixels, ask yourself: where did this video come from? Malicious deepfakes are often designed to provoke a strong emotional reaction, clouding your judgment. Slow down and question the source. Was it posted by a reputable news organization or an anonymous account that was created recently? Perform a quick search to see if established media outlets are reporting on the same story. If a shocking video of a public figure is circulating but no credible news sources are covering it, you are likely dealing with a fake. You can also use a reverse image search on a screenshot from the video to trace its origin and see if it has appeared in a different, legitimate context before.
5. Slow It Down and Look for Glitches
Deepfakes that look convincing at normal speed can fall apart under closer inspection. Slowing down the playback speed or pausing to go through the video frame-by-frame can reveal artifacts you would otherwise miss. Look for flickering around the edges of the subject, particularly where the hair meets the background or along the jawline. This 'shimmer' is a common artifact of face-swapping technology. Inconsistencies in lighting are also a major giveaway. Shadows on the face should match the lighting of the surrounding environment; if they don’t, it suggests the face was digitally added. Sometimes objects in the background may appear to morph, duplicate, or briefly disappear. These small technical glitches are signs that the video has been manipulated. No single clue is definitive, but if you spot several of these issues, it's a strong indicator that the clip is not authentic.
