How to Detect Deepfake Audio and Video Before Sharing

Updated 10 November 2026 · Standard formula, shown below

Use these checks to assess whether an audio clip or video may be manipulated before you pass it along. No single visual or audio clue can prove a recording is fake, so the practical goal is to pause, check its source and context, compare it with independent evidence, and avoid sharing it as authentic when uncertainty remains.

What You’ll Need

The original post, file, or link, if available; avoid relying only on a cropped repost.
A web browser and access to more than one independent, credible source.
Optional: a reverse image search, video key-frame search, or audio-analysis tool. Treat results as clues, not verdicts.
A few minutes to check the date, source, and context before sharing.

Step-by-Step

Pause before forwarding. Do not repost, quote as fact, or send the clip to a group while checking it. Viral claims and emotionally charged recordings benefit from rapid sharing. If the content could expose someone to harm, avoid downloading or reposting it; preserve the link or basic details only if needed to report it.
Find the earliest available source. Open the post and note who uploaded it, when, and where it claims to have been recorded. Look for a link to a full-length video, original statement, or named source. Search a distinctive phrase from the caption or transcript, and search the person’s name alongside a few words from the claim. Reposts without a traceable source are not automatically fake, but they are weak evidence.
Check the claim independently. Search reputable news organizations, the person or organization’s verified channels, and relevant official statements. Look for independent reporting that confirms the specific event, not merely articles repeating the same clip. Check whether the date, location, language, and surrounding events fit. A real recording can be given a false caption or placed in a misleading context.
Inspect the video at normal speed, then replay short sections. Watch the face while the person speaks and then look at the whole frame. Note possible mismatches in lip movement, unusual blinking, facial edges, teeth, hair, lighting, shadows, reflections, or motion between frames. Also check whether background objects warp or details change abruptly. These may indicate editing or compression, but poor lighting, low resolution, filters, and ordinary encoding can produce similar defects. A smooth-looking video is not proof of authenticity.
Listen to the audio separately if possible. Pay attention to whether the voice’s rhythm, breath, emphasis, and pauses sound consistent throughout. Notice abrupt changes in room tone, background noise, volume, or pronunciation, and whether the spoken words align with visible mouth movement. Listen for unnatural phrasing or a voice that seems oddly flat, but do not treat an accent, emotion, or speaking style as proof. Editing, a bad microphone, dubbing, and noise reduction can all create suspicious-sounding artifacts.
Check the file and context without overtrusting metadata. If you have a legitimate copy of the original file, note its filename, duration, and available creation details. Metadata can be removed, changed, or lost when platforms process a file; it cannot by itself establish who recorded it or whether it was altered. Compare the clip with other footage of the same event, such as a longer recording from another angle, and check whether the scene and timeline match.
Use reverse search or detection tools as supporting checks. For video, take a clear frame and search it with a reverse-image tool, or use a video key-frame search if available. This can reveal an older source or a different event. For audio, speech-to-text may help you search a distinctive sentence, while specialist detectors may flag likely synthetic media. Results can be wrong, especially for short, noisy, compressed, dubbed, or edited clips. Check how a service handles uploads before submitting private or sensitive material.
Make a careful sharing decision. If strong, independent evidence supports the clip and its context, share the original source and describe what is confirmed. If evidence is mixed or missing, do not present it as genuine; you can wait, ask for a source, or share a correction with context. If it appears manipulated, report it through the platform’s available process and avoid repeating the claim in a way that amplifies it.

Common Mistakes

Trusting one tell. A strange blink or imperfect lip sync is not a reliable verdict. Real recordings can look odd, and convincing manipulation may have no obvious flaws.
Assuming a detector settles the question. Automated classifiers estimate likelihood and may produce false positives or negatives. Their performance can vary with file quality, language, editing, and the tool itself.
Confusing authenticity with context. A real clip can be old, clipped, dubbed, or captioned misleadingly. Verify what it shows and when, not just whether the pixels or voice seem real.
Using a repost as the original. Cropping, subtitles, filters, and platform compression can hide useful context or create artifacts. Trace the earliest source you can find.
Sharing first and correcting later. A correction may not reach everyone who saw the original. Hold off until you can describe the evidence accurately.

Tips

Separate observations from conclusions: “the audio changes abruptly here” is more precise than “this is definitely a deepfake.”
Look for corroboration from sources that do not all point back to the same upload.
When asking others to check a clip, share a source link and a neutral question rather than a sensational caption.
Keep sensitive recordings out of unfamiliar upload tools. If a clip involves a private person, consider the privacy impact before forwarding even for verification.
If you cannot establish authenticity, say so plainly—or choose not to share.

Troubleshooting

The original link is gone: Search the exact caption or a distinctive sentence, and check whether reputable outlets or the relevant organization have preserved or discussed the clip. Do not treat a cached repost as original evidence.

The clip is too short or low quality: The available material may not support a confident assessment. Look for a longer version or independent recording. If none exists, label it unverified and do not share it as fact.

Checks disagree: Give more weight to independently verifiable context and reliable primary sources than to a single visual impression or detector score. Keep the conclusion open if evidence remains inconclusive.

A tool flags the clip: Check its explanation, limitations, and whether it analyzed the original file or a compressed copy. Seek corroboration; a flag is not proof.

You already shared it: Remove or correct the post where possible, clearly state what changed, and link to reliable evidence. Avoid repeating unsupported details in the correction.

Frequently Asked Questions

Can I identify a deepfake just by looking or listening?
No. Sensory clues can prompt further checks, but real recordings can contain similar artifacts and sophisticated fakes may look or sound natural.

Are deepfake detectors accurate?
They can help prioritize a review, but no detector is definitive. Results depend on the content and tool, and false positives and negatives are possible.

Does a verified account prove a clip is real?
No. Accounts can share misleading material, be compromised, or repost content without confirming its origin. Check the source and the claim independently.

Can metadata prove when a recording was made?
Not reliably. Metadata can be edited or stripped during sharing, so use it only as one clue alongside independent evidence.

What should I do if I cannot verify a clip?
Do not present it as authentic. Wait for credible confirmation, ask for the original source, or leave it unshared.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.