A video verification workflow for newsrooms

A newsroom doesn't need a lab to verify video. It needs a routine that everyone on the desk runs the same way, a written log, and clear rules about what can be published. Here's a workflow that suits a two-person digital desk or a large fact-checking team, with an AI check placed where it actually helps.

Step 0: triage before you start

Not every clip deserves an hour. Decide the level first and write it in the log:

• Low: a light clip for a roundup, a cute animal, a weather moment. Source check and a quick scan.
• Medium: a local incident, a celebrity clip, anything you'll embed in a story.
• High: violence, elections, communal tension, disasters, health claims, or anything that names or shows a private person. Full workflow, two people, an editor's sign-off.

Writing the level down stops urgent clips from being waved through simply because they're urgent.

Step 1: preserve it first

Viral posts get deleted, edited or made private. As soon as a clip reaches the desk:

1. Save the post's link and screenshot the post with its account name, time stamp, caption and view count.
2. Download the best copy you can get, ideally from the earliest post rather than a forward.
3. Note how it reached you: tip line, WhatsApp group, a reporter, a social search.

Step 2: who held the camera?

The question isn't "who posted it?" but "who filmed it?" Work back to the earliest upload, then look at that account: where it usually posts from, what it posted before and after, whether it re-shares other people's clips.

When you can, contact the person who filmed it. Ask how and when they filmed it, ask for the original file, ask permission to use it and how they want to be credited. Think about their safety before naming them, especially in conflict, protest or communal stories.

Step 3: where was it filmed?

Match what you see against maps and street-level imagery: signboards and the script they're written in, road markings, number plates, shop names, skylines, hills, the shape of buses and autos, police uniforms. Write down each match in the log, not just the conclusion, so a colleague can repeat it.

Step 4: when was it filmed?

Check the weather records for the claimed day, the season (crops, clothing, festival decorations) and the light. Free tools such as SunCalc show where the sun was at a given place and time, which can confirm or rule out a claimed hour from the direction and length of shadows. Search for older copies too: one of the most common false claims is an old clip with a new date. Our guide on finding when a video was filmed covers the methods.

Step 5: is it AI-generated or altered?

Run an AI check once you have the best copy. Use the post's link where you can, because a link lets the checker read the caption and tags as well as the video. A useful check gives you more than a number:

• Which checks pointed which way: motion, frames, voice, watermarks, Content Credentials, metadata.
• Moments in the clip you can look at yourself.
• A plain "not conclusive" when the evidence is split.

On Video Verifier, a Deep Scan checks every second of picture and the whole soundtrack, with up to 12× more frames than a Quick Scan, and produces a full report and a PDF you can attach to the log. For links it also looks up published fact-checks. Our video fact-check page lists what it looks at.

Treat the result as one strand of evidence. Our published tests show where our video model is weaker, on polished stock footage and the newest generators, and those numbers are on our accuracy page.

Step 6: corroborate independently

Look for confirmation that doesn't depend on the clip: other angles of the same moment, local reporters, official statements, posts from emergency services, satellite or street imagery. Check whether fact-checkers have already covered it. Two independent sources that agree beat any single tool.

What a "real" result does and doesn't cover

Below: a real clip of a conference hall from Wikimedia Commons, which our video model read as 99% real. It's a good picture of what a "real" result covers. It says the footage looks like genuine camera footage. It says nothing about which event this is, when it happened or what was said there. Those answers come from steps 2 to 4. Real footage under a false caption is still one of the most common kinds of misleading video a desk will see.

Video VerifierReal video

Video is likely

Real

Test set · Oct 2026crackly-ai.comVideo model: 99% real

Step 7: sign off with careful wording

Decide what you can say, and say no more than that. Labels many desks find useful:

• Verified: source, place and time confirmed; no signs of AI generation or editing.
• Authentic footage, context unconfirmed: the video is real, but you couldn't confirm when or where.
• Misleading: real video, false claim (old, another place, selectively cut).
• AI-generated or altered: several independent signals agree.
• Unverified: you couldn't establish enough. Say so, or don't run it.

Avoid copy like "an AI detector proved it fake". Write what you found: where the clip first appeared, whether it carries an AI tool's watermark, what the checks showed. Detector percentages are estimates, never proof, and readers deserve to know that.

Step 8: keep a log a reader could audit

For every clip, record:

• The original post's link and your saved copy.
• Who checked it, when, and the triage level.
• Findings on source, place and time, with the evidence for each.
• AI check results, with the report link or PDF.
• Who signed off, and the final label.

When a reader, a lawyer or a rival outlet asks "how do you know?", the log is your answer. It also makes corrections fast: when new evidence turns up, you can see exactly which step it changes.

The desk's toolkit

• Keyframes and reverse image search, to find earlier copies.
• Maps and street-level imagery, for place.
• Sun position tools and weather archives, for time.
• Fact-check databases, to avoid redoing work.
• An AI check that shows its evidence.

Our free fact-check toolkit explains each of these with named tools.

For teams

If your desk checks videos every day, the pricing page has a teams and newsrooms option with Deep Scans and PDF reports at volume, and checks inside your own tools on request. A reporter can start right now with a free Quick Scan on the home page, no sign-up needed.