How to Spot an AI Face Swap: A Checklist for Teachers and Students

A step-by-step check for a suspicious photo or video: look closely, check the file, search for the original and read around it. No single sign proves a face swap, so the checklist asks for evidence from more than one place.

We make face-swap software. That is why we know where swaps tend to fail, and why we think people should be able to check what they see. Where a point comes from our own experience rather than a cited source, the text says so. Every source is listed at the end with the date we last read it.

Key takeaways

• No single clue settles it. Treat every visual sign as a reason to look further, not as a verdict.
• The strongest evidence is outside the image. Finding the original, or finding who made it, beats staring at the face.
• Missing metadata proves nothing. Most images online have been re-saved; a blank result is the normal result.
• "Can't tell" is a valid answer. It is often the honest one, and a good reason not to share.

Start here: no single clue settles it

MIT Media Lab's Detect DeepFakes project puts it plainly: there is no single tell-tale sign of a fake, only artifacts worth looking for 1. Detection apps don't settle it either. Poynter's MediaWise lesson notes that detection tools are not 100% accurate and people still need to review the content themselves 2.

So treat every visual clue as a reason to look further, not as a verdict. A real video can look smeared after heavy compression, and a careful swap can show none of the signs below. The strongest evidence usually comes from outside the image: finding the original, or finding who made it.

1. Stop before you share

The SIFT method begins with Stop: notice your emotional reaction, because headlines and posts are often built for clicks, and think about what you already know about the topic and the source before you share 3. A clip that makes you angry or delighted in the first second is exactly the one to slow down on.

2. Look closely at the face

Watch at full screen, pause, and step through the parts where the face is clearest. MIT's list of things to pay attention to is a good classroom starting point 1:

• Cheeks and forehead. Is the skin too smooth or too wrinkly? Does the age of the skin match the hair and eyes?
• Eyes and eyebrows. Do shadows fall where you would expect them?
• Glasses. Is there glare, too much glare, and does it change angle when the person moves?
• Facial hair and moles. Do a beard, sideburns or a mole look real, and do they stay put?
• Blinking. Does the person blink too little or too much?
• Lips. Do the lip movements look natural and match the sound?

Poynter's lesson adds, for video, blurring around edges, flickering, shadows and lip syncing; for images, odd hands, reflections, shadows and backgrounds; and for audio, the tone and the background sounds 2.

From our own experience building face-swap software

A face swap replaces the inner face and blends it into the original head, so the seams are where to look. These are the moments swaps handle worst, in our experience:

• The edge where the face meets the hairline, jaw and neck, especially if the skin tone changes across it.
• A head turning to a side profile, or tilting far up or down.
• A hand, microphone, hair or food passing in front of the face.
• Fast movement and motion blur, where the face can lag or flicker for a frame.

None of these proves anything on its own. Real footage has motion blur too, and better tools keep reducing these artifacts.

3. Check the file: credentials and metadata

Some files carry Content Credentials, built on the open C2PA standard for recording where content came from and how it was edited. Where they are present, you can see the method of creation and an editing history 4. You can check a file at contentcredentials.org/verify.

Google's "About this image" can also show how an image was made or edited, but Google warns that this metadata can be changed with editing tools and may not be accurate 5. The InVID plugin's metadata tab reads Exif data in JPEG photos and metadata in MP4 videos 6.

Missing credentials or metadata prove nothing. The C2PA specification itself notes that a file can become separated from its credentials when its metadata is removed or corrupted 7. Many images and clips you meet online have been re-saved or re-shared, so a blank result is common. A credential that says "made with AI" is a strong clue; no credential is no clue at all.

4. Search for the original

A face swap needs a source photo or video. If you can find it, you have your answer: the same scene, clothes and background, with a different face.

• Google Lens. On a computer, use Search by image on google.com to upload a file or paste an image link, or right-click an image in Chrome and choose Search with Google Lens. On a phone, touch and hold an image in the Google or Chrome app 8. "About this image" shows when Google may have first seen similar versions, and earlier pages that used it, though it is only available in some regions 5.
• TinEye. Upload, paste or drag in an image, or give it a URL. Sort results by Oldest to see the earliest copies TinEye found, or by Most Changed to surface heavily edited versions 9. No result only means TinEye hasn't indexed a match, not that none exists.
• Video. Reverse search works on still images, so search frames. Take screenshots where the face is clear and where the background is distinctive. The InVID plugin's keyframes tool splits a video into frames you can search on Google, Yandex, TinEye and Baidu 6.

5. Check the context, not just the pixels

Professional fact-checkers read laterally: they leave the post and search the open web to find out who is behind it before trusting it 10. The Digital Inquiry Group frames this as three questions: Who's behind the information? What's the evidence? What do other sources say? 11

• Investigate the source. Who posted it first, and what do trusted sources say about that account or outlet? 3
• Find better coverage. If a famous person really said or did this, would reputable outlets or fact-checkers such as FactCheck.org, Snopes or PolitiFact have covered it? 3
• Trace it to the original context. Follow links back to the first version and check whether it was clipped, relabelled or presented as something else 3.

The checklist

Print it or put it on a slide. Students tick a step only when they can say what they found.

1. Stop before sharing. Note how the post makes you feel.
2. Pause and look at the face edges, skin, eyes, glasses, blinking and lips.
3. Watch the moments a head turns or a hand crosses the face.
4. Check for Content Credentials and metadata, knowing both can be missing.
5. Reverse-search the image, or still frames from the video.
6. Look for the same scene with a different face.
7. Find who posted it first, and what other sources say about it.
8. Decide with evidence from the file, the search and the context, not the face alone.

Using it in class

• Ask for a claim plus evidence: "I think this is (or isn't) a face swap because…", with at least one clue from the file, one from the search and one from the context, not only from the face.
• Allow "can't tell" as an answer. It is often the honest one, and it is a good reason not to share.
• Use public examples that fact-checkers have already resolved, so the answer can be checked against a reliable source. Don't make swaps of classmates or staff to use as test material.

A note on responsible use

Face-swap tools have legitimate uses in film, memes, parody and personal projects, and they can also be used to deceive or harass. Our rules of thumb: swap only faces of people who have agreed, label edited media as edited, and never use a swap to impersonate someone, mislead people about real events, or make sexual or humiliating images of anyone. Depending on where you live, some of those uses are also illegal.

If you find a harmful fake of yourself or a student, don't reshare it to warn others. Save the link, a screenshot and the date, report it to the platform, and tell a trusted adult or your school.

If you want to see how a swap is made, so you know what to look for, the online photo tool on this site is free and needs no account; the Windows app handles video on your own PC, and its desktop page explains what it can and cannot do.

Sources

Each source was read on October 29, 2026. The MIT Media Lab page was read through its October 23, 2026 Internet Archive copy, because the live page would not load for us.

1. MIT Media Lab — Detect DeepFakes: How to counteract misinformation created by AI
2. Poynter MediaWise — Recognizing AI-generated content (AI Unlocked lesson, updated March 24, 2025)
3. University of Chicago Library — The SIFT Method
4. Content Credentials / C2PA — Content Credentials
5. Google Search Help — Learn more about an image
6. InVID project (maintained within WeVerify) — InVID verification plugin
7. Coalition for Content Provenance and Authenticity — C2PA Technical Specification 2.3, section 2.4.2
8. Google Search Help — Search with an image on Google
9. TinEye — TinEye tutorial
10. Digital Inquiry Group — Lateral Reading on the Open Internet
11. Digital Inquiry Group — Civic Online Reasoning