A convincing video is not automatically trustworthy. Real footage can be miscaptioned, old clips can be presented as current events, audio can be replaced, and synthetic video can now look highly realistic.
The safest approach is to verify the video as evidence rather than trying to “spot the fake” by eye.
A practical video verification workflow
1. Identify the exact claim
What is the video supposed to prove? A location, event, statement, identity or action? Separate the footage from the caption and ask what evidence would be needed to support the claim.
2. Find the earliest source you can
Look for the original uploader, publisher or organisation. Reposts often remove useful context and provenance. If the clip is supposedly from a broadcaster, government body, company or public figure, check that source directly.
3. Search key frames
Pause the video at clear moments and search individual frames using visual or reverse-image search. This can reveal older versions, the original event, a different country or a previous use of the footage under another caption.
4. Check context and continuity
Compare landmarks, weather, language, clothing, shadows, signage, vehicle plates, timing and surrounding reports. For edited clips, check whether cuts change the meaning of what was said or shown.
5. Check the audio separately
Audio can be cloned, dubbed or replaced while the video remains genuine. If a voice or statement matters, search for the original recording or an independent source. Treat voice identity and video identity as separate questions.
6. Look for provenance signals
Where supported, Content Credentials can provide signed provenance information about creation or editing. Watermarks such as SynthID can provide another signal for content generated with participating AI systems. Google says SynthID is used across image, video, audio and text, and its verification tools can detect those watermarks in supported media. Read about SynthID.
7. Use detector results cautiously
Deepfake detectors can help, but video compression, frame-rate changes, screenshots, re-encoding and new generation models can affect performance. Use detection as supporting evidence rather than the final answer.
Why checking one frame is not enough
Video contains temporal information: movement, lip synchronisation, blinking, lighting changes and frame-to-frame consistency. A suspicious frame may be caused by motion blur or compression, while a convincing still frame does not prove the full sequence is authentic. Verification should consider both individual frames and the sequence as a whole.
What if the video has no Content Credentials?
Keep verifying. Missing provenance does not prove a video is fake. Credentials can be stripped or may never have been added. Source, context, corroboration and forensic analysis still matter.
What if a watermark is found?
A valid watermark can be strong evidence that a participating system generated or altered the media, but it does not tell you whether the surrounding claim is accurate. A synthetic video may be satire, an advert, a demonstration or a scam. Context still matters.
Community Smart Hub’s verification rule
Do not ask only: “Is the video fake?” Ask: “What can I independently establish about its source, history and claim?”
Use our Verify & Trust pathway, read The Five-Layer Shield, and explore the AI Trust Lab.
Quick checklist
- What exactly is this video claiming?
- Can I find the original source?
- Do key-frame searches find an older or different context?
- Does independent reporting match?
- Is the audio independently verifiable?
- Are provenance or watermark signals available?
- Do detector results agree with the wider evidence?
Bottom line: video verification is an evidence-gathering process, not a visual guessing game.
