A viral clip can look convincing and still be edited, misleading or AI-generated. This practical deepfake checklist helps you assess suspicious videos before sharing them, using visual, audio, source and corroboration clues rather than relying on one giveaway.
Overview
When a video appears in your feed, the first question should not be “does this look real?” It should be “what can I verify?” Generative tools and conventional editing can both produce clips that seem authentic at a glance. A video may also be genuine but presented with the wrong date, location, caption or context.
No single sign proves that a video is AI-generated. Warped hands, strange reflections and unnatural speech can be useful clues, but compression, poor lighting, fast movement and reposting can create similar artefacts. Treat the process as a series of checks. The more independent clues point in the same direction, the more confident you can be.
Before investigating, save the link and note the account that posted it, the caption, the date shown and any claims made about the footage. If the clip may disappear, follow a careful archiving process such as the one explained in how to save a viral video before it disappears. Preserve the original context where possible; a screen recording without the caption or account details may remove useful evidence.
Checklist by scenario
1. Start with the source
- Is this the original uploader, or a reposting account?
- Does the account provide a date, location, event name or explanation?
- Has the caption changed as the clip moved between TikTok, Instagram, YouTube or X?
- Do other posts from the account suggest satire, fan editing or synthetic content?
An account with little context is not automatically unreliable, just as a verified or popular account is not automatically correct. Look for the earliest accessible version and compare captions across platforms. A useful next step is to trace the clip or related image back to its source, using the same approach as a viral meme source check.
2. Inspect faces, hands and movement
Pause the video at several points, especially during quick gestures or close-ups. Look for facial features that subtly change shape, teeth that merge or disappear, earrings that move inconsistently, hair that blends into the background, or skin texture that appears unnaturally smooth. Hands may show irregular finger lengths, fused fingers or objects that seem to pass through them.
Movement matters more than a single frozen frame. Watch whether a face changes naturally when the person turns, whether glasses remain correctly aligned and whether a hand maintains the same anatomy from one frame to the next. Fast cuts and low-resolution reposts make these checks harder, so mark the result as uncertain rather than treating a blurry frame as proof.
3. Check lip-sync and audio
Listen while watching the speaker’s mouth. Possible warning signs include consonants that do not match the lips, a voice that remains unnaturally even during emotional speech, breaths that appear in the wrong places, or room sound that cuts out around edits. Pay attention to names, numbers and unusual words, where synthetic speech or dubbing may sound less natural.
Audio problems can also come from ordinary editing. A clipped microphone, background noise reduction or a translated voice-over does not by itself establish that the video is fake. Compare the sound with the visible environment: do footsteps, applause, traffic or music change when the image changes?
4. Examine lighting, reflections and shadows
Follow the light across the scene. Faces, clothing and nearby objects should generally respond to the same light source. Be cautious if a person’s shadow points in a different direction from the shadows of surrounding objects, or if reflections in windows, mirrors, water or polished surfaces do not match the scene.
These clues require care. A camera’s exposure, artificial lighting and reflective surfaces can produce confusing results. Use them as supporting evidence, not a verdict. Try watching the clip at normal speed and then frame by frame before drawing a conclusion.
5. Look for abrupt edits and missing context
A short clip can be real but misleading because the beginning or ending has been removed. Ask what happened immediately before the first frame and after the last. Sudden changes in image quality, background position, audio tone or camera angle may indicate a cut, a stitched compilation or a reposted excerpt.
Search distinctive phrases, captions or visible signs rather than only searching “is this video real?”. If the video concerns a celebrity, public event or breaking story, look for independent reporting, official statements or longer footage. This is also why a weekly roundup should separate verified developments from unconfirmed social chatter; see how to build a reliable viral stories roundup.
6. Check labels, provenance and technical details
Some platforms or editing tools may display an AI, altered-content or provenance label. Treat a label as useful context, not as a complete authenticity judgement. Labels can be missing, removed during reposting or applied differently across services. Conversely, the absence of a label does not prove that footage is untouched.
If you have access to the original file, inspect its metadata for creation time, editing software and file history. Metadata is supporting evidence only: social platforms commonly strip or rewrite it, and it can sometimes be altered. A downloaded copy is not necessarily the same file that was first recorded.
| Clue type | What to check | How much weight to give it |
|---|---|---|
| Visual | Faces, hands, edges, reflections and shadows | Useful when several inconsistencies repeat across frames |
| Audio | Lip-sync, breathing, room sound and unnatural edits | Helpful, but compression and dubbing can mislead |
| Source | Original uploader, caption history and date | Often essential for identifying missing context |
| Technical | Labels, provenance information and metadata | Supporting evidence, not a final answer |
| Corroboration | Longer footage, eyewitness material and reputable reporting | Stronger when sources are genuinely independent |
What to double-check
Reverse image search can help locate earlier frames, related reports or an older version of a clip. Take several clear screenshots from different moments rather than relying on one frame. Search distinctive visual details and exact wording from the caption. Results may be incomplete for video, newly created content or material from private accounts, so a failed search is not evidence that the clip is authentic.
Compare the claim with reliable, independent sources. For a supposed UK incident, check whether local reporting, emergency services, public bodies or people connected to the event have addressed it. Do not count ten accounts repeating the same post as ten separate confirmations if they all copied one another.
Consider the possibility that the video is neither fully real nor fully synthetic. Common alternatives include an authentic recording with an AI voice-over, a real person placed in a different setting, a genuine event shown with an incorrect date, or a compilation combining several unrelated clips. The most accurate conclusion may be “edited”, “misleadingly captioned” or “not verified”, rather than simply “AI” or “real”.
AI-detection websites and browser tools can be useful for further investigation, but their results should not be treated as conclusive on their own. Detection methods change as generation and editing tools change, while short, compressed social clips provide limited material to analyse. When the stakes are high, wait for confirmation instead of sharing an uncertain accusation.
Common mistakes
- Trusting your eyes alone: A polished or unsettling appearance is not a reliable authenticity test.
- Using one giveaway as a verdict: Odd hands, a strange shadow or robotic audio can result from editing or compression.
- Confusing popularity with confirmation: Views, likes and reposts measure distribution, not truth. See what viral metrics really mean.
- Sharing before checking the caption: A real clip can become misinformation when its location, date or context is changed.
- Accusing a real person too quickly: Deepfake claims can cause harm. Use cautious language when evidence is incomplete.
- Ignoring copyright and privacy: Investigating or reposting a clip does not automatically give permission to publish it. Review this UK guide to reposting viral video before using someone else’s footage.
When to revisit
Use this checklist whenever a suspicious clip becomes part of the social media trends, celebrity news or breaking conversation you are following. Recheck it before posting a reaction, embedding the video in an article or repeating a claim on a podcast. If the clip concerns a serious allegation, public safety, an election, a crime or a private individual, raise your verification threshold and wait for stronger corroboration.
The practical workflow is simple: save the original link, record the claim, inspect the visuals and audio, trace the source, check labels and metadata where available, search for earlier versions, and compare independent reporting. If those checks do not resolve the question, share the uncertainty rather than the clip. Say “unverified” or “appears edited” when that is what the evidence supports.
Revisit this guidance when platforms introduce new provenance labels, editing features or verification tools, and when detection methods change. For everyday browsing, keep a short version beside you: pause, inspect, trace, corroborate, then share. That habit is more dependable than trying to memorise one permanent sign of an AI-generated video.
