How to Spot a Deepfake Video: A Practical Checklist

Learning how to spot a deepfake video comes down to a simple habit: slow down, look at a few specific parts of the frame, and check where the clip actually came from before you believe or forward it. A deepfake is a video in which a person’s face, voice or body has been synthetically altered or generated by artificial intelligence, usually to make them appear to say or do something they never did. The good news is that most fakes in wide circulation still leave visual and contextual clues. The harder truth is that the technology is improving fast, so no single trick works forever. This guide from newsreverse com walks through the cues that still hold up, and the verification steps that matter more than any one visual tell.

What is a deepfake, and why does it spread so easily?

A deepfake uses machine-learning models trained on real footage to swap a face, clone a voice, or fabricate a talking-head clip from scratch. Because these clips are cheap to make and emotionally charged, they travel quickly on WhatsApp forwards, short-video apps and social feeds. In India the most common patterns are fake endorsements of investment schemes using the likeness of well-known business figures, doctored clips of politicians during election season, and non-consensual sexual imagery targeting private individuals. The motive is usually money, influence or harassment. Understanding that motive helps you stay skeptical: ask who benefits if you believe this clip, and why it reached you the way it did.

How to spot a deepfake video by looking at the frame

Full-screen a suspicious video and, if you can, step through it slowly. Deepfakes tend to break down at the boundaries and in fine detail. Watch these areas:

• Face edges. The line where the face meets hair, ears, neck or the background is where models struggle. Look for a faint blur, a flicker, or an outline that seems pasted on.
• Lip sync. Watch the mouth on hard consonants like b, m and p, where the lips must fully close. If the lips do not quite meet the sound, or the movement looks slightly delayed or too smooth, be suspicious.
• Eyes and reflections. Check whether both eyes catch light the same way. Mismatched reflections, glassy stares, or eyes that do not track naturally are common artefacts.
• Teeth and jewellery. Fine, repetitive detail is hard to fake. Teeth can look like a smeared block, and earrings or spectacles may warp between frames.
• Lighting and shadows. The direction of light on the face should match the rest of the scene. A face lit differently from the room around it is a red flag.
• Skin and texture. Overly smooth, waxy skin, or a face that is oddly sharper or softer than the background, can signal manipulation.

One outdated tip deserves a warning. For years people were told that deepfakes fail to blink naturally. That was a genuine weakness around 2018, but modern models fixed it. Treat blinking as one weak signal at best, and never as proof either way. The same caution applies to any single visual tell: the reliable approach is to weigh several cues together.

Which cues are reliable and which are outdated?

Because the quality of fakes keeps climbing, it helps to know which signs still earn their place. The table below sorts the common cues by how much weight they deserve today.

How do I verify the source instead of just the pixels?

Visual inspection catches many fakes, but source-checking catches more, and it keeps working even as the models improve. The single most important question is: where did this video first appear, and who is standing behind it?

1. Find the original. A convincing fake is often a real clip with a new face or new audio bolted on. Take a clear still from the video and run a reverse image search across more than one engine, since each indexes a different slice of the web. If the footage exists elsewhere in a different context, you have your answer.
2. Check who is reporting it. If something this dramatic were real, established newsrooms would carry it. If only anonymous accounts and forwarded messages are pushing the clip, treat it as unverified until a credible outlet confirms it.
3. Read the account, not just the post. Newly created profiles, mismatched handles, and accounts that only post inflammatory content are warning signs.
4. Mind the context. Deepfakes are often paired with urgent captions designed to make you share before you think. Financial “opportunities”, sudden scandals just before a vote, and shock clips are the usual bait.

This source-first instinct is the same one that protects you from doctored photos and AI-generated stills. If you have not already, it is worth reading our companion guide on how to spot an AI-generated image, because many of the same habits apply across formats.

Can provenance tools and Content Credentials help?

A promising development is content provenance: a way of attaching a tamper-evident record to a file that says where it came from and how it was edited. The best-known standard is C2PA, backed by a coalition of technology and media companies, and often surfaced to the public as Content Credentials. When a photo or video carries a valid credential, you can inspect a cryptographically signed manifest describing its origin.

Understand the limits, though. Provenance proves authenticity at the point of creation; it does not detect fakes after the fact. A credential records what the signer asserts, so its value depends on the signer being honest. And the absence of a credential does not mean a video is fake, because most everyday clips carry none. Think of Content Credentials as a green tick you can sometimes rely on, not a fraud detector that flags the bad stuff.

The same goes for automated deepfake detectors. They return a probability score, not a verdict, and they can be fooled or can misfire on genuine footage. Use them as one more input, never as the last word.

A quick everyday routine

You do not need a lab to be a careful viewer. The following routine takes under a minute and catches most manipulated clips before they mislead you:

• Pause and resist the urge to forward. Strong emotion is the point of the fake.
• Full-screen it and scan the face edges, mouth and lighting.
• Grab a still and reverse-search it.
• Ask whether any trusted newsroom is reporting the same thing.
• Look for Content Credentials if the platform shows them, but do not treat their absence as proof.

Digital literacy sits alongside basic account safety, because the same bad actors who spread fakes also try to hijack real accounts. It is worth pairing this habit with strong login protection, which we cover in our explainer on two-factor authentication, and with an understanding of secure connections in the difference between HTTP and HTTPS. You can browse more guides in our technology section.

How are deepfakes made, and why does that help you spot them?

You do not need to be an engineer to benefit from a rough sense of how these clips are built. Most face-swap deepfakes are produced by training a model on many images of a target person and then mapping that learned face onto a source video frame by frame. The model is excellent at the centre of the face and much weaker at the boundaries, at fine texture, and at keeping everything consistent from one frame to the next. That is precisely why the tells cluster where they do: at the hairline and jaw, in the teeth and eyes, and in flickers between frames. Fully synthetic clips, generated from a text prompt rather than a source video, have their own signature weaknesses, such as physically impossible details in the background, hands with the wrong number of fingers, or text on signs that dissolves into nonsense. Voice clones, meanwhile, are trained on samples of a person speaking; they can nail the timbre but often miss the natural rhythm, breathing and hesitation of real speech. Knowing that the technology is strong in the middle and weak at the edges tells you exactly where to point your attention.

Why is India a particular target?

India’s scale, its many languages, and its heavy reliance on forwarded video messages make it fertile ground for manipulated media. A single convincing clip can be re-captioned in a dozen languages and pushed into millions of chats within hours, often faster than any correction can follow. Election periods see doctored clips of candidates; festival and market seasons see fake celebrity endorsements of trading apps and quick-money schemes; and private individuals, disproportionately women, are targeted with fabricated intimate imagery. The common ingredient is emotional pressure, a clip engineered to make you angry, greedy or afraid so that you share it before you check it. Recognising that pressure as a manipulation tactic in itself is half the defence. When a video seems designed to provoke an instant reaction, that is the moment to slow down rather than speed up.

What if a deepfake targets you or someone you know?

If a manipulated video is being used to defame, extort or sexually exploit a person, treat it as a crime, not just a nuisance. Preserve the evidence first: save the URLs, take screenshots with visible dates, and note the accounts sharing it. In India you can file a complaint on the National Cyber Crime Reporting Portal or call the cyber-crime helpline 1930, and you can report the content to the platform hosting it so it can be taken down. Acting quickly matters, because early removal limits how far a fake travels.

The technology behind deepfakes will keep getting better, and one day the visual tells in this guide may fade. What will not change is the value of a skeptical, source-first mindset. Slow down, check where a clip came from, ask who benefits, and confirm with people you trust before you believe or share. That habit protects you far better than any single trick for spotting a seam in the pixels.