Suspicious videos, images, and audio deserve a pause before you share them, send money, or make a decision. This practical deepfake detection checklist shows how to examine visual and audio clues, verify the source and context, search for earlier versions, use detection tools carefully, and document uncertainty without overstating what you know.
Overview
A deepfake is manipulated or AI-generated media designed to make a person appear to say or do something that did not happen. Some fakes are obvious, while others are convincing enough that a single visual clue cannot settle the question. Compression, poor lighting, subtitles, camera movement, and reposting can also make genuine footage look unusual.
The safest approach is not to ask, “Can I find one sign that proves this is fake?” Instead, use a repeatable workflow:
- Preserve the original claim. Save the link, account name, caption, and upload date. Do not rely only on a screenshot.
- Inspect the media. Look for visual, audio, and editing inconsistencies, but treat them as clues rather than proof.
- Check the source and context. Identify who posted it, when it first appeared, where the event supposedly happened, and whether independent reporting or primary evidence supports it.
- Search for earlier versions. Reverse-image searches, video keyframes, and distinctive quotations can reveal an older clip with a different caption.
- Assign a confidence level. Mark the item as verified, likely manipulated, unverified, or misleading rather than forcing a yes-or-no conclusion.
This process is especially important when a clip creates urgency. A deepfake scam may imitate a public figure promoting an investment, a company representative requesting payment, or someone you know asking for immediate help. Do not transfer money, disclose a code, or click a linked site based only on the media. Verify through a separate, trusted channel.
Checklist by scenario
When checking a suspicious video
- Watch the clip several times at normal speed, then pause at moments involving the face, hands, text, or a change in lighting.
- Look for lip movements that do not consistently match the words, unnatural blinking, shifting facial edges, inconsistent teeth, or skin texture that changes between frames.
- Check hands, jewellery, glasses, hair, and objects near the face. AI-generated details may change shape or position, although ordinary video compression can create similar defects.
- Listen for abrupt changes in tone, breath, room sound, pronunciation, or background noise. A clean voice over a visibly noisy scene may warrant closer review.
- Inspect cuts and transitions. A short excerpt may remove the part of the event that would explain what was really said or shown.
- Capture several clear keyframes and search them separately. A reverse-image search may find the original event, an older upload, or a fact-check that provides missing context.
When checking a suspicious image
- Zoom in on text, signs, logos, fingers, reflections, shadows, and repeated patterns. Errors in these areas can be useful clues, but their absence does not establish authenticity.
- Compare the image with photographs from the claimed event or location. Check weather, clothing, architecture, camera angle, and the position of notable objects.
- Run a reverse-image search using the full image and a distinctive crop. Cropping can help when unrelated page elements obscure the main subject.
- Search a distinctive caption or quoted statement in quotation marks. Look for the earliest traceable version rather than assuming the most popular post is the original.
When checking suspicious audio
- Ask whether the speaker’s identity is independently confirmed. A familiar voice is not sufficient proof of who created the recording.
- Listen for unnatural pauses, repeated sounds, clipped consonants, inconsistent emotion, or a voice that changes around edits.
- Find the full recording if possible. A short clip can be rearranged, taken out of context, or combined with another person’s voice.
- Contact the supposed speaker using a phone number or account you already trust, not contact details supplied with the recording.
When a post asks for money or action
Slow down even if the media seems authentic. Confirm the request through a separate channel, visit an official website by typing its address yourself, and check payment details independently. For related checks, see our message verification checklist and guide to checking whether a website is legitimate.
What to double-check
Source, date, and location
Review the account’s history, username, profile details, and links. An account may be impersonating a person or organisation, or a genuine account may have reposted material without checking it. Confirm the stated date and location using independent sources. A real video can become misleading when paired with a false caption.
Metadata and file history
Metadata can sometimes show a file’s creation software, timestamp, or device information, but social platforms and messaging apps may remove or change it. Metadata is therefore supporting evidence, not a reliable authenticity certificate. A lack of metadata does not prove manipulation.
AI detection tools
An AI-generated image checker or deepfake video checker can be useful as one input, particularly when comparing several tools or reviewing a large set of files. Treat results as indicators rather than verdicts. Detection systems can produce false positives and false negatives, especially after resizing, cropping, re-encoding, adding subtitles, or recording a screen. Do not upload private family, client, or identity documents to an unfamiliar tool without understanding its privacy terms.
Independent confirmation
Give more weight to evidence that does not simply repeat the same post. Look for original footage, a direct statement from the person involved, records from a credible primary source, or reporting that explains how the media was authenticated. Several accounts repeating identical wording may represent one unverified source, not independent confirmation.
Record your reasoning in a simple note: original link, account, date found, key observations, searches performed, tools used, and unresolved questions. This makes it easier to correct your conclusion if better evidence appears.
Common mistakes
- Relying on appearance alone: Modern manipulation may not contain obvious glitches, and genuine low-quality footage may look strange.
- Assuming a watermark proves origin: A logo, badge, or platform label can be copied or added during editing.
- Confusing popularity with verification: Views, likes, and reposts show distribution, not accuracy.
- Searching only the exact claim: Try names, locations, dates, distinctive phrases, and image crops to uncover earlier context.
- Uploading sensitive media everywhere: Use caution with detection services, browser extensions, and unknown apps. Our fake app safety guide explains how to examine an app before installing it.
- Sharing while adding a warning: A warning can still amplify a false claim. Save evidence privately and share only what is necessary.
- Making a high-stakes decision before verification: If the content involves money, account access, health, safety, or reputation, pause and use a trusted second channel.
When to revisit
Revisit this checklist whenever the claim changes, a new version of the media appears, or the original account edits its caption. Return to the source before sharing during elections, emergencies, major product announcements, seasonal shopping periods, and other moments when misleading media may spread quickly.
Detection tools and platform features also change. Before relying on a tool for professional publishing, review its current privacy practices, supported file types, limitations, and explanation of results. Keep this compact checklist nearby:
- Save the original link, file, caption, and date.
- Inspect face, hands, text, lighting, sound, edits, and background continuity.
- Search full media, keyframes, crops, names, dates, and distinctive phrases.
- Check the uploader and confirm the event through independent sources.
- Use AI detection tools only as supporting evidence.
- Verify money or access requests through a trusted separate channel.
- Label the result accurately: verified, likely manipulated, misleading, or unverified.
- If you cannot resolve it, do not share it as fact.
If you already acted on a suspicious deepfake scam, secure affected accounts, contact the relevant payment provider, preserve messages and links, and follow a recovery plan such as our step-by-step guide for what to do after a scam.
