How to Teach Students to Spot AI-Generated Videos

Trying to spot AI-generated content with the naked eye is "basically a guessing game, even for seasoned reporters."

What holds up better is a habit. Teach students to ask who made a video, to look for other coverage of the same claim, and to treat any label as a clue rather than a verdict. Below is a 40-minute lesson you could run during U.S. Media Literacy Week, October 26 to 30, 2026, and the reasoning behind it.

Why It's Hard to Teach Students to Spot AI-Generated Videos by Eye

Looking closely is an unreliable test, and studies disagree on how unreliable: how well people do depends on the kind of media they are shown. One study of 1,276 participants found average detection "close to a chance level performance of 50%" across images, video, audio and audiovisual media. A University of Florida team reported in February 2026 that people correctly identified real and fake videos about two-thirds of the time, though they did no better than chance on still photos of deepfake faces. The Florida tests used specific kinds of faces and videos under controlled conditions.

Tips that expire as the technology improves can't anchor a lesson, and tips for spotting AI visuals are "becoming less useful as the technology advances."

Kids who get most of their news through social apps often assume that if it's video, "it must be real." A checklist of looks can leave students surer than they should be, which is why the lesson below has them rate their confidence before and after.

Teach the Habit: Who Made This, and Who Else Reports It?

Teach students to pause before sharing, find where a video came from, and look at "the source, evidence and surrounding context." Then see whether other credible sources report the same thing.

The habit also fits a standard you may already teach to, ISTE 1.3.b: students evaluate "the accuracy, validity, bias, origin, and relevance of digital content." "Origin" is exactly what this habit practices.

Match the Method to the Age

For ages 8 to 10, try a two-source rule: before believing or sharing, find two real news outlets that say the same thing. For older students, Common Sense Education offers a 45-minute lesson for grades 9 to 12 on tracing a claim back to its original source, though it is built on a written article, not a video.

The step of leaving the video to see what other sources say is often called lateral reading. A study we read taught it to eighth graders using live websites, and we found none on AI video.

Treat Content Credentials as a Positive Signal Only

Some cameras and apps attach Content Credentials, a provenance record that the C2PA standard describes as working "like a nutrition label for digital content." The Content Authenticity Initiative's FAQ says, about images, that they "provide a positive signal about the origin and history of an image" but "don't provide a negative signal about the authenticity" of one.

So a credential is a clue worth reading when it's there, and its absence proves nothing. C2PA's FAQ on removed metadata adds that the metadata can be removed "inadvertently or intentionally." For most classroom clips, the signals students can actually see are the creator's note and YouTube's label. Teach students that "no credentials" means "keep checking," never "fake."

Treat YouTube's Labels as a Clue, Not a Verdict

A YouTube label tells students something about how a video was made, not whether what it shows is true, and a missing label is not proof of anything. As of October 28, 2026, YouTube's help page says it requires creators to disclose "AI-generated or altered content that seems realistic". A label in the player "may also appear" for photorealistic AI content, and animated content may carry its disclosure in the expanded description.

Automatic labels are conditional, and the wording differs by page. Since May 2026, YouTube has said that when a creator doesn't say whether they used AI and its systems "detect significant photorealistic AI use, we will now automatically apply a label," directly below the player on long videos and as an overlay on Shorts. A separate YouTube help page on automatic labels says that for undisclosed content, "in some cases, YouTube may take action to reduce the risk of harm to viewers by proactively applying a label."

The "Captured with a camera" label appears in the expanded description, and only for video recorded with built-in C2PA support and no significant edits to sound or visuals. As of October 28, 2026, YouTube's help page on that label says that if the label is absent, "it doesn't mean the content has modified audio or visuals." No YouTube page we read gives a figure for how complete the labeling is.

A 40-Minute Lesson With Three Clips (This Post's Own Design)

Before class, watch every clip in full, and prefer clips whose origin you can verify yourself. Clip A should be real, with an origin you can verify, such as a video your school recorded or one from an outlet you can confirm. Clip B should be AI-made with a visible disclosure, such as a creator's note in the description or a YouTube label. Clip C should be unlabelled, with an origin you have traced yourself and will reveal at the debrief. Our suggestion: label clips A, B and C and name no channel on slides or handouts, so the lesson stays about the habit.

1. Warm-up (5 minutes). Students rate from 1 to 5 how sure they are that they could tell whether a video was made with AI. Collect the ratings.
2. Three clips (15 minutes). Show clips A, B and C. After each, students record the evidence they used on an evidence sheet: what they noticed, where it came from, what any label said, and what they still don't know.
3. Lateral reading (10 minutes). Students leave the video. In pairs, they look for other sources on Clip C: who posted it, whether anyone else reports the same thing, whether a credible outlet covers it. "We couldn't confirm it" counts as a valid finding.
4. Debrief (5 minutes). Reveal what you traced. Compare it with their evidence sheets: which evidence held up, which misled. Students re-rate their confidence.
5. Class checklist (5 minutes). The class writes its own short checklist, such as "Who made this? Who else reports it? What does the label say, and what doesn't it say?"

Our suggestion for upper elementary: run step 3 as a teacher-led search that finds two sources. Middle school students can search in pairs, and older students can trace the claim back to its original source.

Getting the Clips to Students

SafeShare creates a dedicated page for a YouTube or Vimeo video that focuses on the video content, instead of sending viewers to the original platform where they might see comments and suggestions. The page hides comments and suggested content.

Set a custom start time and an optional end time so students see only a segment, the moment you picked for each example. Each page comes with a built-in QR code, and you can share the page's link instead. If the lesson lives in a learning platform, a page can be embedded with the provided embed code, but the site's FAQ says classroom tools do not accept embed codes from every website, so keep the link as your fallback.

Pages can show some of the video's own details, such as its title, description and uploader. You can set a custom title, but check what yours reveals before you use it for the "who made this" step.

Make the pages ahead of time with a free account, which keeps links active. As of October 28, 2026, pages created without one are temporary and may expire within a few hours. Some videos won't play: the page supports public videos not restricted by region or age.

Get Your Three Clips Ready

Choose your three clips, then Create your first SafeView free; a free account keeps your links active, so each page and QR code is ready on lesson day.