How to Check Whether a Viral Video Is AI-Generated Before Sharing It

A startling clip arrives in a family group: a public figure makes an extraordinary claim, a street appears flooded or a celebrity seems to endorse a product. Video used to feel like direct evidence. Today it may be authentic, edited, out of context or generated with AI. The safest first question is not whether a face looks strange. It is where the clip came from, when the underlying event happened and who else can independently confirm it.

Start with the original uploader

A forwarded clip often loses its source. Search for the earliest account or website that posted it and see whether the account has a record of original reporting. Look at the caption and upload date, but remember that a video can be reuploaded years after an event. A verified social-media account is not proof of every claim it shares. A suspicious source should lower your confidence without automatically proving that the video is fabricated.

Separate the image from the claim

Sometimes a real video is attached to a false location, date or explanation. A crowd scene could be from a different city; footage of an old storm could be presented as an incident this morning. Write the claim in one sentence: who, what, where and when. Then search for each element separately. This approach can reveal misleading captions even when no AI has been used. It also prevents a debate about deepfake detection from distracting from ordinary misinformation.

Look for independent confirmation

For a major public incident, look for official statements, direct eyewitness records and independent credible reporting. Do not rely on several accounts that all copied the same original post. Confirmation is stronger when different sources describe consistent facts and can show their own evidence. For a consumer scam, consult the brand’s official channels rather than an influencer clip claiming to reveal a secret offer. Absence of reporting does not by itself prove a claim false, but it is a reason to pause.

Technical clues can help, but are not verdicts

AI-generated video can show odd hands, changing jewellery, inconsistent reflections, drifting signs or movements that do not match physical surroundings. Audio may sound unnaturally smooth or lack expected environmental noise. However, compression, filters and poor lighting can create similar artifacts in real footage. Newer systems can also avoid many of the old errors. Treat visual clues as prompts for further investigation, not as a reliable yes-or-no test.

Examine the audio and the edit

Listen for sudden changes in ambient sound, speech that does not match the mouth and awkward cuts around a key sentence. A genuine recording can be edited to remove context. If a politician appears to announce a sweeping policy, look for the complete speech or a transcript from an official source. If a product testimonial sounds too perfect, check whether the person has actually published the endorsement. The missing minutes before and after a clip can matter more than one strange frame.

Use reverse-search and frame comparison carefully

Capture one distinctive frame and look for earlier appearances using reputable image-search tools. Search visible landmarks, event banners or unusual objects as text. If you find an older version with a different caption, compare both. Reverse searching does not always find new or private material, and a negative result is not proof of authenticity. Do not upload private family footage to unknown detection websites simply to test it. Protect the people in the recording.

What about AI-detection tools?

Automated classifiers can provide clues, but false positives and false negatives occur. Performance may drop after a clip has been compressed, cropped, screen-recorded or altered. Some services make sweeping confidence claims without showing how they were tested. Use detection as one input alongside source history and independent corroboration. If the stakes are high, avoid a definitive public accusation based only on a detector score. The question is what the evidence supports, not what a single percentage suggests.

Teach children a repeatable checklist

A family can use four simple questions: Who posted this first? When was it recorded? What exactly is being claimed? Can we check it elsewhere? Demonstrate with an ordinary entertainment clip rather than frightening material. Explain that even people we like can share mistaken posts. Encourage children to ask before reposting personal images or insulting someone shown in a questionable video. Curiosity and patience work better than ridicule.

When the clip could cause harm

If a video appears to involve fraud, impersonation, non-consensual intimate imagery or a threat, avoid spreading it in the name of verification. Save the URL and relevant evidence, use the platform’s reporting mechanism and consider official reporting channels where appropriate. Do not contact an alleged victim through a viral thread simply to demand proof. The most responsible answer is sometimes: this claim is unverified, so I will not amplify it.

Questions readers often ask

Is an odd-looking hand proof of a deepfake?

No. Image artifacts can arise from compression, motion and editing. Check the source and context.

Does a blue check mean a clip is authentic?

No. Account verification alone does not verify a particular claim.

A practical example

Imagine a ten-second clip supposedly showing a sudden closure of an airport. A careful viewer notices that no date appears in the video. Searching for the terminal architecture reveals footage from an older construction project. The video itself may be genuine, while the caption is misleading. The example shows why verifying an event’s place and time often produces a clearer answer than guessing whether every frame was generated by AI.

Should I repost it to ask if it is fake?

If it may harm someone, avoid amplifying it. Use a private description or a link to trustworthy fact-checking instead.

Build a verification habit

Before forwarding an emotional video, wait long enough to identify its original source. Search the core claim, not just a sensational phrase in the caption. Compare dates and locations and ask whether independent evidence exists. If the video turns out to be wrongly labelled, correct the claim in the same group where it circulated. Explain the process without embarrassing the person who shared it. A useful verification culture makes it easier for others to slow down the next time.