Here is the uncomfortable truth first: independent research has found that most people can no longer reliably tell a real video from an AI-generated one just by looking. The visual tricks that worked against 2019-era deepfakes, such as counting blinks, mostly fail against modern generators. That does not mean you are defenceless. It means the checks that matter most have shifted from spotting pixels to verifying sources.
The Checks That Still Work, Sometimes
Watch the Edges, Not the Centre
AI models spend most of their computational effort on the main subject’s face, which means backgrounds and boundaries get comparatively less attention. Watch where a face meets the hairline, ears and neck, particularly as the person turns their head. Look at straight lines in the background, such as door frames or bookshelves, for subtle warping as the subject moves.
Check Hands and Small Details
Hands remain a genuine weak point for AI generation, particularly complex poses such as interlaced fingers or holding an object. Jewellery, glasses and earrings can also flicker, shift position or blend into skin unexpectedly across a clip.
Listen as Carefully as You Watch
AI-generated speech sometimes inserts breathing sounds at the wrong moments, or the audio sounds unnaturally clean for the visible setting, such as studio-quality audio over supposedly outdoor, windy footage. Slight lip-sync mismatches, where mouth movement does not quite match the words, remain a genuine tell.
For Still Images: Look at Text and Repeating Patterns
AI image generators still struggle to render readable text accurately, so signs, labels and jersey numbers often melt into near-letters under close inspection. Repeating patterns such as fences, brickwork or crowds of people can merge or repeat unnaturally when examined closely.
The Check That Matters Most: Verify the Source
Before analysing any pixel, ask the more basic questions first. Who posted this, and is that account established or brand new? Has a recognised, trustworthy news source reported the same event? If the content features a public figure, has their own official account addressed or shared it? A shocking clip from an anonymous, newly created account deserves scepticism regardless of how convincing the footage looks.
This approach, often called lateral reading, means checking other sources about the same claim rather than only scrutinising the suspicious content itself. It consistently outperforms visual analysis, particularly as generation quality keeps improving.
Can Detection Software Help?
Tools such as browser-based deepfake detectors and mobile security apps can flag some suspicious content, but none currently claim to catch everything, and detection technology is generally understood to lag behind generation technology. Treat detection tools as one signal among several, not a definitive verdict.
Frequently Asked Questions
Can most people still tell a deepfake from a real video?
Independent research suggests most people cannot reliably do this anymore, since modern AI generation has moved past the obvious visual flaws that older detection advice relied on, such as unnatural blinking.
What is the single most reliable way to check if a video is real?
Verifying the source is generally more reliable than visual analysis. Check whether an established, trustworthy account or news source has reported the same event before trusting a shocking or unusual clip.
Are deepfake detection tools reliable?
They can help flag some suspicious content, but detection technology is widely considered to lag behind generation technology, so no current tool guarantees a definitive answer on its own.
What are the most common giveaways in AI-generated still images?
Text within an image, such as signs or labels, often renders incorrectly, and repeating patterns such as fences or brickwork can merge or repeat unnaturally when examined closely.
Should I share a suspicious video before verifying it?
No. Sharing unverified content, even with a caveat, contributes to its spread. Verify the source and check whether trustworthy outlets have reported the same event before resharing anything that seems shocking or unusual.
Want to understand more about how AI detects patterns like this itself? Read our explainer on what agentic AI actually means or browse AI News & Updates.
