User-generated video has become one of the most powerful forms of evidence in modern journalism. It has also become one of the most dangerous to mishandle. When a clip lands in our inbox or appears on social media tagged to a breaking news event, the pressure to move quickly is real. So is the risk of publishing something fabricated, misattributed, or recycled from an unrelated incident. This checklist is designed to slow that process down, just enough to get it right.
Step 1: Establish basic provenance
Before any technical analysis, ask the most basic questions. Who uploaded this video? When was it first posted? On which platform? Tools like InVID/WeVerify, developed with support from the European Commission, allow journalists to break a video into key frames and run those frames through reverse image search engines including Google Images, TinEye, and Yandex. This single step regularly reveals footage that has been recirculated from a prior conflict, natural disaster, or entirely unrelated event. The Yandex reverse image search is particularly useful for footage originating in Eastern Europe or Central Asia, where its index is deeper than Google's.
Step 2: Extract and read the metadata
Video files carry embedded metadata that can reveal the device used to record them, the software used to edit them, and sometimes GPS coordinates. Tools such as ExifTool, a free command-line utility maintained by Phil Harvey, allow editors to read this data directly. Be cautious here: metadata can be stripped or spoofed, so its presence is a signal worth noting, not a guarantee of authenticity. Its absence, however, especially in footage that claims to be raw and unedited, is itself a red flag that warrants further investigation.
For a broader view of how synthetic media complicates this process, our field guide to synthetic media for newsrooms lays out the landscape editors need to understand before tackling any individual clip.
Step 3: Geolocate the footage
Geolocation is the practice of confirming that a video was recorded in the location it claims to show. This involves identifying fixed landmarks in the footage, including buildings, street signs, terrain features, shadows, and vegetation, and cross-referencing them against satellite imagery in Google Earth, Bing Maps, or Maxar's archive. The open-source intelligence community, in particular groups like Bellingcat, has documented this methodology extensively. Bellingcat's online guides provide step-by-step instruction in matching camera angles, shadow direction, and architectural detail to confirmed map coordinates.
Shadow direction and length can also help establish the approximate time of day, which you can then test against the claimed recording date using sun-position calculators such as SunCalc. If the shadows in the video are inconsistent with the sun's position at the stated time and location, the footage's provenance is in serious doubt.
Step 4: Listen and look for signs of manipulation
As AI-generated and AI-manipulated video becomes more accessible, visual and audio anomalies are increasingly worth scrutinising. Unnatural blinking patterns, inconsistent lighting on faces, hair that behaves oddly at the edges, or audio that does not quite sync with lip movement are all indicators that have been cited by researchers at MIT Media Lab and the Partnership on AI. None of these signals is conclusive on its own, but a cluster of them justifies escalating to dedicated detection software.
Our buyers guide to deepfake detection tools covers the current landscape of software options, their accuracy rates, and their pricing. And given that manipulated audio is increasingly used in combination with genuine video, our piece on voice cloning and audio deepfakes is essential reading before drawing conclusions from a clip's soundtrack alone.
Step 5: Contact the source directly
If the video was posted by an identifiable account, contact that person. This is not optional. Platform verification badges mean very little, accounts can be hacked or impersonated, and a brief exchange over a direct message or email can surface inconsistencies that no automated tool will catch. Ask them to describe what they saw, where they were standing, and what happened immediately before and after the clip ends. Inconsistencies in that account are meaningful. So is a refusal to engage.
When the source cannot be contacted, triangulate. Find other people who were present, look for other footage of the same scene from different angles, and consult local journalists or stringers who have ground knowledge. Wire services including Reuters, AP, and AFP have each built internal verification desks precisely for this purpose, and their practices are worth studying. We have reported on how those organisations are building institutional responses to deepfakes and synthetic media.
Step 6: Document your verification process
Whatever conclusion you reach, document how you reached it. Record which tools you used, what the metadata said, which landmarks you matched, and who you contacted. This documentation serves two purposes. First, it protects your newsroom if the footage is later disputed. Second, it builds institutional knowledge so that the next editor facing a similar clip does not start from scratch.
Newsrooms developing a broader editorial response to AI-generated content will find strategic context in our piece on how newsrooms are responding to AI-generated video.
A note on speed versus accuracy
The checklist above takes time. We recognise that. But the reputational cost of publishing a fabricated or misattributed video is far higher than the cost of being second. Publishing wrong is not being first. It is simply being wrong, and in an environment where trust in journalism is already under sustained pressure, that distinction matters more than ever.
Sources
- InVID/WeVerify project, European Commission-funded media verification toolkit (weverify.eu)
- ExifTool, Phil Harvey (exiftool.org)
- Bellingcat, open-source geolocation and verification guides (bellingcat.com)
- SunCalc sun-position calculator (suncalc.org)
- MIT Media Lab, research on deepfake detection indicators (media.mit.edu)
- Partnership on AI, Synthetic Media Framework documentation (partnershiponai.org)