How to check a suspicious video, voice message or call.

The principle

Check the request, the source and the context before you study the pixels. A scam is a scam even when the recording is real.

No tool delivers a final verdict of “fake” or “real”. Describe the outcome of a check with more careful wording:

• no sign of manipulation found;
• unclear;
• contradicted by independent sources;
• confirmed by independent sources.

Step 1. What is being asked of you?

1. 1If it involves money, urgency, secrecy, credentials or a change of payment details, treat the request as hostile until you confirm it through another channel. A demand for wire transfers or gift cards is a red flag ¹. The FBI gives examples including fake family emergencies and fake executive video chats ².

Step 2. Source and first upload

1. 1Find out who published it first, whether that is the original account, and whether the claimed speaker or organisation carries it on its own verified channels.
2. 2Look for the earliest and the longest version of the clip.

Step 3. Independent confirmation and context

1. 1Check whether independent coverage exists for the claimed event, date and place.
2. 2Reverse-search key frames and images. This is a common verification method; we do not recommend a specific tool here.

Step 4. Contact the person through a channel you know

1. 1Hang up and call back on a number you already have or look up yourself, not the number that called you ².

Step 5. What eyes and ears can and cannot tell you

Visual and audio tells are weak. A Ukrainian state counter-disinformation centre reported in November 2025 that some AI-generated videos cannot be debunked without additional verification tools ³. The FBI lists distorted hands, irregular faces and odd shadows as cues ², but the absence of such flaws proves nothing.

Step 6. Provenance, if it is present

1. 1Check whether the file carries Content Credentials (C2PA) metadata, meaning information about where the content came from. C2PA is a technical standard for certifying the source and history of media ⁴. Treat it as extra context, not a verdict.

Step 7. Detectors: last, and with care

On in-the-wild 2024 content, open-source detectors lost about half of their AUC compared with earlier benchmarks. Commercial tools do better but do not reach the level of forensic analysts ⁵. Common problems include false positives, language challenges and variable accuracy ⁶.

1. 1Record the tool, its version, the date, the input, the output and the confidence. Try more than one tool, and never base a public claim on a single result.

For journalists and fact-checkers

• Document each step so a reader could repeat it, and be transparent about how you reached your conclusions ⁶.
• The IFCN commitments on sources and methodology are a useful benchmark for how to write it up. NAK is not a signatory ⁷.

What next

• If it is about you or your company, see the guide “The first 24 hours after a deepfake”.
• If money is involved, see the guide “Voice clones and fake calls”.

Limits