How to Spot Manipulated Images, Videos, and Audio

Updated 5 November 2026 · Standard formula, shown below

Use this guide to assess whether an image, video, or audio clip may have been edited, generated, or taken out of context. You will need the media itself, a device with internet access, and a few minutes to compare it with reliable sources; no single visual clue or detection tool can prove that something is fake.

What You’ll Need

The original post or file, if available, plus its URL, account name, date, and any caption.
A web browser and access to search tools for reverse image search, video keyframes, or audio transcripts.
Optional: headphones, a screenshot or saved copy, and a note-taking app for recording what you checked.

Step-by-Step

Pause before sharing. Treat surprising, emotionally charged, or urgent media as unverified until you have checked it. Avoid reposting it “to ask if it is real”: that still spreads the material. If it might put someone at risk, do not contact or publicly identify the person shown.
Find the earliest available source. Open the post and note who published it, when, and what it claims to show. Search a distinctive phrase from the caption or a few words from a transcript. Look for an original upload, a named eyewitness, or reporting that links to primary evidence. An old clip reposted with a new date or location is a common form of misleading media, even when the file itself has not been altered.
Check the context independently. Search for the location, event, people, and date using more than one reputable source. Compare visible landmarks, signs, weather, clothing, and shadows with the claim, but do not rely on these alone: lighting and camera settings can mislead. Check whether established local reporting, official statements, or direct records support the event. A missing news report is not, by itself, proof that a claim is false.
For an image, run a reverse image search. Use an image-search service to upload the picture or search with its URL. If that does not work, crop a distinctive area and try again. Look for earlier appearances, a fuller version, or a matching scene with a different caption. Search results can miss edited, cropped, or newly posted images, so try more than one service when the stakes are high.
Inspect the image at full size. Zoom in on edges where objects meet, reflections, repeated patterns, fingers, text, and fine details. Note inconsistent shadows, warped lines, or mismatched focus. These may suggest editing or image generation, but compression, resizing, motion blur, and ordinary camera processing can produce similar artifacts. Treat clues as reasons to investigate, not a verdict.
For video, inspect several moments. Watch the clip once for its overall claim, then pause at different points, especially around cuts and fast movement. Search a few distinct frames with a reverse image search or a video keyframe search tool. Compare the clip with the earliest upload and check whether omitted seconds, a different angle, or a longer version changes its meaning. A real video can still be deceptively edited or miscaptioned.
For audio, check the recording and the words. Listen with headphones at a comfortable volume, then replay short sections. Note abrupt changes in background noise, unnatural pauses, repeated sounds, or speech that seems detached from the room acoustics. These can result from edits, poor connections, or noise reduction as well as synthetic speech. Search distinctive spoken phrases and compare a transcript with the audio; automatic transcripts can mishear names and words.
Compare voices cautiously. If a clip claims to feature a specific person, look for a longer recording from a dependable source, or a direct confirmation through an established channel. Differences in pitch or accent are not enough to establish manipulation: microphones, emotion, illness, distance, and editing all affect a voice. Do not upload private or sensitive recordings to unknown analysis services.
Use detection tools only as supporting evidence. An image, video, or audio “deepfake detector” may flag a file, but results depend on the tool, file quality, and methods used to create or alter it. A clean result does not prove authenticity, and a warning does not prove fakery. Read what the tool actually tested, avoid sharing sensitive media, and seek independent evidence.
Make a measured conclusion. Record the claim, earliest source found, checks performed, and unresolved questions. Classify the result as supported, contradicted, or unverified rather than forcing a yes-or-no answer. If you post a correction, link to evidence and explain uncertainty; do not accuse a person based only on artifacts or an automated score.

Common Mistakes

Assuming strange hands, blurry text, or uneven lighting prove a picture is generated. Compression and normal editing can create visual defects.
Trusting a verified-looking account, a large number of shares, or a confident caption as proof. Accounts can be mistaken, copied, or compromised, and popularity is not verification.
Checking only the current post instead of searching for older copies and the original context.
Treating metadata as conclusive. Metadata can be stripped or changed, and a file’s date or device information does not establish what it depicts.
Believing a detector’s percentage is a universal probability that media is fake. Scores are tool-specific and may not be calibrated for the file being assessed.

Tips

Keep a record of source URLs and search terms so you can retrace your checks.
Compare the exact claim with the evidence: a genuine recording can be paired with a false date, place, speaker, or interpretation.
Use independent sources rather than several posts that all trace back to the same upload.
Be especially careful with private people, children, and sensitive events. Avoid downloading, forwarding, or publishing material that could expose or endanger someone.
When a claim is time-sensitive, state what is known and what remains uncertain, and update your conclusion if better evidence appears.

Troubleshooting

Reverse search finds nothing: Try a crop with a clear landmark or object, another search service, or several video frames. A lack of matches is not proof that the media is original.

The file is too blurry or compressed: Look for the earliest, highest-quality upload. If none is available, report that the evidence is inconclusive rather than relying on tiny details.

Sources disagree: Compare their evidence, publication dates, and links to original material. Separate confirmed facts from interpretation, and wait for stronger sourcing if the claim is consequential.

A detector gives a result that conflicts with your research: Do not let either one settle the question alone. Check the tool’s limitations and prioritize verifiable provenance, context, and independent corroboration.

You cannot establish the source: Label the clip unverified, avoid sharing it as fact, and seek a trustworthy primary source or qualified fact-checking coverage.

Frequently Asked Questions

Can I tell whether an image is AI-generated just by looking?
Sometimes clues are visible, but there is no reliable visual shortcut. Editing, compression, and camera effects can resemble generation artifacts.

Do reverse image searches prove a picture is real?
No. They can reveal earlier appearances or different context, but a match does not verify the caption, and no match does not establish authenticity.

Can a real video be misleading?
Yes. It may be old, filmed somewhere else, selectively cut, or paired with an inaccurate description without being digitally altered.

Are deepfake detection tools reliable?
They can help identify material worth investigating, but accuracy varies. Treat their output as one clue, not a final judgment.

What should I do if I shared a misleading clip?
Remove or correct your post where possible, explain what changed, and link to reliable evidence. Avoid repeating an unsupported claim in the correction.

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