Most viral fakes are real footage with a false story
After a disaster or a protest, fact-checking teams such as AFP Fact Check, Reuters Fact Check and Full Fact see the same pattern again and again. The footage is often genuine, but it is old, or it was filmed somewhere else, or it is cut so that the meaning changes. Deepfakes are growing, but a miscaptioned clip is still the cheapest way to mislead.
People know the problem exists. In the Reuters Institute Digital News Report 2025, 58% of respondents across all surveyed markets said they were concerned about their ability to tell what is real from what is fake in online news. Concern was highest in the United States and in Africa (73% each) and lowest in Western Europe (46%). Online influencers and personalities and national politicians were each named by 47% as among the biggest threats for false or misleading information online.
Four checks before you trust a clip
Reverse search the key frames
Screenshot two or three distinct frames, such as a wide shot and a sign. Search each one with Google Lens, Bing Visual Search, Yandex Images and TinEye, because each index finds different copies. The free InVID-WeVerify plugin, built by partners including AFP, extracts key frames for you. If an older copy appears, the "breaking" clip is recycled.
Trace the first uploader
Look for the earliest upload. Check how old the account is, what else it posts, and whether it was plausibly at the scene. Reposts by aggregator pages tell you nothing about the source.
Confirm the place and the date
Match landmarks, signs, road markings and the language people speak against satellite and street-level imagery of the claimed location. Compare the weather, the season and the direction of shadows with records for the claimed date. A video cannot be from an event if it was online before that event happened.
Look for signs of manipulation
The MIT Media Lab Detect Fakes project suggests watching the face closely: skin that looks too smooth or too wrinkly, odd shadows around the eyes and eyebrows, glare on glasses, blinking too little or too much, and lip movements that do not look natural. Also check hands, teeth and background text.
What file metadata can and cannot tell you
A phone or camera file can carry a creation date, a device model and sometimes GPS coordinates. Free tools such as ExifTool read these fields. But most social platforms remove much of this metadata when a file is uploaded, and anyone can edit it. Metadata is most useful when the person who filmed the clip sends you the original file.
A newer signal is Content Credentials, based on the open C2PA standard. Some cameras and editing apps attach a signed record of how a file was captured and edited. A valid credential is strong evidence; a missing one proves nothing, because most video online still has none.
Red flags in the caption
- Urgency such as "share before they delete this".
- No place, no date and no named source.
- A clip that is cropped, mirrored or filmed off another screen.
Viral Video Verification Checklist
Answer what you have checked. Unchecked items count as half risk, because an unchecked clip is not a verified clip.
Method: each warning sign adds points (no original uploader 20, older copies found 25, place mismatch 15, date mismatch 10, artefacts 20, cropped or re-filmed 5, urgent caption 5; the maximum is 100). An unchecked item adds half its points. Independent confirmation subtracts 20, and the score never drops below 0. Bands: 0-24 low, 25-49 medium, 50-74 high, 75-100 very high. The weights are our judgement of how strongly each sign points to a false claim; the score is a guide, not a verdict.
Viral video verification questions
Can metadata prove that a video is real?
No. Most social platforms remove much of a file's metadata on upload, and metadata is easy to edit. Use it as one clue next to the source, the location and the date.
What is the most common kind of fake viral video?
Fact-checkers regularly debunk real footage shared with a false caption: an old clip presented as new, or a clip from one country presented as another. A reverse search of key frames finds many of these in minutes.
If I see no deepfake artefacts, is the video real?
Not necessarily. Generation tools improve quickly, and compression hides small flaws. Treat visual checks as a warning sign when they fail, not as proof when they pass.