Recognize AI Images, Video and Audio: A Verification Guide

Updated 5 November 2026 · Standard formula, shown below

This guide helps you assess whether an image, video or audio clip may have been generated or altered with AI. You will compare the media with its context, look for corroborating evidence and judge uncertainty; no single visual clue or detector can prove how a file was made. You’ll need the original or highest-quality copy you can access, a web browser, and time to check sources.

What You’ll Need

The media file or a link to its earliest known source. Avoid relying only on a screenshot or compressed repost.
A browser for reverse-image searches, frame searches and source checks.
Headphones and, if useful, a way to slow playback without changing pitch.
A notes app or document to record where the file came from, what you checked and what remains uncertain.

Step-by-Step

Pause before sharing. Treat a surprising or emotionally charged clip as unverified until you have checked it. Save the URL and note when and where you found it. If the claim could cause immediate harm, look for reliable confirmation rather than amplifying the clip.
Find the earliest available version. Trace reposts back to an original post, publisher, witness or official source. Compare dates, captions and surrounding material. A post date is not proof of when the media was recorded, and a credible account can still share an inaccurate clip.
Check the claim against independent evidence. Search distinctive phrases, names, locations and event details. Look for reputable reporting, original statements, other recordings or records that could confirm or contradict the claim. Prefer independent sources over multiple accounts repeating the same upload.
Inspect images at full resolution. Zoom in on small details such as hands, teeth, jewellery, text, reflections, shadows and object edges. Check whether lighting, perspective and anatomy make sense throughout the image. These may reveal editing or generation, but unusual details can also come from blur, resizing, motion or ordinary photography. Their presence is a clue, not a verdict.
Search for the image elsewhere. Use a reverse-image search with the full image, then try a crop of a distinctive subject or background. Look for earlier appearances, different captions, or a source image that has been edited. A match can establish that an image existed earlier; it does not by itself establish who made it or whether the current context is truthful.
Check video in sections. Watch once at normal speed, then review key moments and transitions. Look for changes in face shape, clothing, background, lighting or object position; inspect whether movement and shadows stay consistent. Extract a few clear frames from different points and reverse-search them. A single frame may be misleading, and ordinary compression can create visual glitches.
Listen to audio carefully. Replay short sections and note abrupt changes in room tone, unnatural pauses, inconsistent pronunciation, odd breaths or a voice that does not fit the scene. Compare the voice with a reliable recording, if available, while remembering that microphones, editing and stress alter how people sound. Do not treat a voice match as proof of identity.
Review file information and provenance. If you have the original file, inspect its metadata and any content credentials or edit history provided with it. This may offer useful clues about a file’s origin or changes, but metadata can be stripped, altered or absent. A missing credential does not mean a file is fake, and a credential does not guarantee that every caption or claim is accurate.
Use detection tools only as supporting evidence. If you try an AI-media detector, check what kind of media it supports and whether it explains its result. Save the result, but do not call a clip authentic or fake based on a score alone. Tools can produce false positives and false negatives, especially on compressed, cropped, translated or edited media.
Write a cautious conclusion. Separate what you observed from what you infer. For example: “The earliest copy I found is dated later than the claimed event; the source and recording date remain unconfirmed.” If evidence is insufficient, say so. Revisit the assessment if stronger evidence appears.

Common Mistakes

Relying on one tell. Strange fingers, smooth skin, odd blinking or robotic speech are not definitive. Real cameras and editing can create similar effects.
Assuming a detector settles the question. Detector output is an estimate, not a certificate of authenticity. Different tools may disagree.
Confusing “AI-generated” with “false.” A generated illustration may be clearly labelled and harmless; a genuine recording can be misleadingly cropped or captioned.
Trusting metadata as proof. File details can be removed or changed, and screenshots often contain little useful provenance information.
Checking only the subject. The person or object may be real while the place, date, quote or surrounding context is wrong.

Tips

Preserve the original file and link when possible; repeated downloads and messaging can reduce detail.
Search exact wording from a caption or alleged quote as well as visual material.
For video, compare several frames across the clip rather than judging one paused moment.
Keep a brief evidence log: source, date, searches performed, findings and unresolved questions.
Use precise language: “unverified,” “possibly altered” or “evidence supports” is more accurate than “definitely fake” when proof is incomplete.

Troubleshooting

The reverse search finds nothing: Try a different frame, crop or image orientation, and search distinctive text or landmarks. No match is not evidence that the media is original; search indexes are incomplete.

The clip is too blurry or compressed: Seek a higher-quality copy from the earliest source. Do not enhance it and then treat invented-looking details as original evidence; processing can introduce artifacts.

Sources disagree: Compare their evidence, dates and access to the original. Look for independent confirmation and state the disagreement instead of choosing the most confident post.

A detector flags genuine-looking media: Check whether the file was compressed, edited or captured from a screen, and try a second independent line of verification. Report the detector’s result as a limited signal, not a conclusion.

You cannot establish whether it is AI-made: You may not be able to. Focus on whether the specific claim is corroborated, avoid sharing it as fact, and explain what remains unknown.

Frequently Asked Questions

Can I tell from one visual clue that an image is AI-generated?
No. A clue may justify further checks, but editing, low resolution and normal photographic effects can look similar.

Are AI image or audio detectors reliable?
They can help identify material for review, but accuracy varies by tool and file. Treat results as supporting evidence, not proof.

Can metadata prove a photo or video is genuine?
No. Metadata can support a provenance check, but it can be missing, changed or disconnected from the claim being made.

What is the best way to check a viral video?
Find its earliest source, verify the date and location independently, inspect multiple frames, and search those frames for earlier appearances.

What should I do if a clip could put someone at risk?
Do not repost it as fact. Seek confirmation from reliable, independent sources and use appropriate reporting channels if there is an immediate threat.

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