Is This Video AI? How to Spot AI-Generated Videos on TikTok, Instagram and YouTube
A year ago, Americans typed “is this AI” into Google about 40,500 times a month. In August 2026 that number reached 135,000. The narrower question “is this video AI” grew almost five times over the same period. People are not asking out of curiosity. They are scrolling TikTok, Instagram and YouTube, seeing something that looks slightly off, and trying to work out whether to believe it.
This guide covers what actually helps: the labels each platform uses, the signs that give AI video away, the tools worth trying and the ones that promise more than they deliver.
The numbers: how fast the question is growing
We pulled monthly US search volumes from DataForSEO, comparing September 2025 with August 2026.
| Search | Sep 2025 | Aug 2026 | Change |
|---|---|---|---|
| is this ai | 40,500 | 135,000 | ×3.3 |
| ai slop | 27,100 | 74,000 | ×2.7 |
| is it ai | 9,900 | 60,500 | ×6.1 |
| ai video detector | 1,900 | 4,400 | ×2.3 |
| is this video ai | 590 | 2,900 | ×4.9 |
| how to tell if a picture is ai | 720 | 1,900 | ×2.6 |
Two things stand out. The generic “is it AI” grew fastest, which suggests the doubt is no longer limited to obvious fakes; people now question ordinary-looking clips. And searches for “sora watermark” jumped from about 70 a month to 22,200 in October 2025, the month OpenAI’s Sora app spread across social feeds, before settling back to around 600. A single app launch was enough to teach millions of people that some AI videos carry a mark you can look for.
Start with the label the platform gives you
Before you study pixels, check whether the platform has already told you. The major apps now label AI content, although none of them catches everything.
TikTok
TikTok asks creators to label realistic AI-generated content, and it adds an “AI-generated” label automatically when a video carries Content Credentials, the industry metadata standard also known as C2PA. Look under the creator’s name and in the caption area. The label is missing when a creator skips it and the file has no metadata, which is common for clips that were downloaded and re-uploaded.
Instagram, Facebook and Threads
Meta shows an “AI info” label on posts it detects as AI-generated or that the creator marks as such. Tap the label to see why it was applied. Meta reads the same kind of industry metadata, so a video that went through a screen recording or several re-uploads may arrive without it.
YouTube
YouTube requires creators to disclose realistic altered or synthetic content. The notice “Altered or synthetic content” appears in the expanded description, and for sensitive topics such as elections or health it is also shown on the video player. Open the description and look for the “How this content was made” section.
Videos from the Sora app
Videos generated in OpenAI’s Sora app carry a moving visible watermark and embedded metadata. If you see a clip with the Sora logo drifting around the frame, it came from the app. If the edges of the frame look cropped or a corner is oddly blurred, someone may have tried to remove the mark.
A missing label proves nothing. It only means nobody declared the video as AI and no metadata survived. Treat the label as a quick yes, never as a reliable no.
Seven signs you are watching an AI video
Generators improve every few months, so no single sign is decisive. Look for several at once.
- Length and cuts. Most AI clips still run for a few seconds to about a minute and are often stitched from short shots. Hany Farid, a digital forensics professor quoted by the BBC, puts it simply: “The three things to look for are resolution, quality and length.”
- Soft or smeared detail. Real phone footage is sharp in places and noisy in others. AI video often looks evenly smooth, with skin that has no pores and edges that blur slightly when they move.
- Text that changes. Signs, labels, jersey numbers and license plates are hard for generators. Pause the video and read any text in the background, then play a second and read it again.
- Hands, teeth and jewelry. Count fingers, watch rings and earrings as the head turns, and look at teeth when someone speaks. These details tend to merge or reshape between frames.
- Physics that almost works. Water, hair, fabric and crowds give AI away. Watch whether objects keep their size and shape, whether a ball bounces the way it should, and whether people in the background repeat the same motion.
- Sound that does not belong to the room. Check whether the voice matches the space. A person speaking outdoors should not sound like they are in a studio, and lip movements should match hard consonants such as “p” and “b”.
- A story with no source. A dramatic clip of a real event usually appears from several angles, uploaded by different people. If you can find only one version and it comes from an account that posts nothing but viral clips, be skeptical.
Check where the video came from
Most misleading AI videos are not caught by looking harder. They are caught by checking the source.
- Look at the account. When was it created, what else does it post, and does it ever show the same people or places in other videos? Pages that publish several “unbelievable” clips a day with no location are a warning sign.
- Take a screenshot and search it. Grab a clear frame and run a reverse image search with Google Lens or a similar tool. You may find the original clip, an earlier upload with a different caption, or a fact-check.
- Read the comments. Other viewers often spot the problem first. Search the comments for “AI” or “fake” before sharing.
- Check Content Credentials. If you can download the file, upload it to the Content Credentials verify tool. It shows whether the file carries C2PA metadata and which tool created or edited it. Metadata is easy to strip, so an empty result is not proof either way.
Do AI video detectors work?
Searches for “ai video detector” more than doubled in a year, and there are now dozens of sites promising a verdict in seconds. Services such as Hive and Deepware analyze a file frame by frame and return a probability score.
They are useful as one more signal, with three caveats:
- They are trained on known generators, so a new model can slip through until the detector catches up.
- Compression hurts them. A clip that has been downloaded from TikTok, re-uploaded to Instagram and forwarded on WhatsApp has lost much of the detail detectors rely on.
- They produce false positives, especially on heavily filtered or edited real footage.
Google also embeds an invisible watermark called SynthID in media made with its own AI tools, and its Gemini app can check files for that mark. This only confirms content from Google’s generators; it cannot clear a video made elsewhere.
Use a detector the way you would use a second opinion: helpful when it agrees with what you already see, not something to trust on its own.
What to do when you are not sure
Often you will not reach a firm answer, and that is fine. The practical rule is simple: if a video is designed to make you angry, scared or amazed, and you cannot confirm it, do not share it. If you do want to share it, say that you are not sure it is real. On most platforms you can also report content that should carry an AI label but does not, which helps the label system work for the next viewer.
Frequently asked questions
Is there a free way to check if a video is AI?
Yes. Check the platform label first, then take a screenshot and run a reverse image search, and upload the file to the Content Credentials verify tool. Several AI video detectors offer free checks with limits on file size or the number of uploads.
Can TikTok tell if a video is AI-generated?
Partly. TikTok labels content automatically when the file carries C2PA metadata and relies on creators to label the rest. Clips without metadata or a creator label can appear with no AI marking at all.
What does “AI info” mean on Instagram?
It is Meta’s label for content that was detected as AI-generated or marked as AI by the person who posted it. Tap it to see more detail about why it appears.
What is AI slop?
“AI slop” is a popular term for low-effort AI-generated content made in bulk to collect views, such as surreal animal clips, fake disaster footage and invented celebrity moments. Searches for the term almost tripled over the past year.
Can AI videos be detected with 100% accuracy?
No. Neither people nor detection tools can identify every AI video, and the gap changes as generators improve. The most reliable approach combines platform labels, visual checks and verifying the source.
