How to Tell If a Video Is AI Generated Before You Share It

Red Flags That a Video Was Made with AI

Check who posted a suspect video and what surrounds it before you judge the footage. Then look for the visual red flags below, and remember that no single sign settles the question. These checks help you spot video made with artificial intelligence (AI) before you share it.

Watch for these red flags:

• A short clip. Hany Farid, a UC Berkeley professor who studies manipulated media, told National Public Radio (NPR) that "the biggest tip is the video length." Farid says many AI clips run 8 to 10 seconds.
• Camera movement that does not make sense. Jeremy Carrasco, an AI literacy educator, says "there isn't a real person behind the camera."
• A subject framed too perfectly, with action that starts and stops cleanly. Farid says AI videos tend to look this way.
• A camera that is implausibly close, or that moves as smoothly as if it were on a gimbal. Farid says this can be a clue.
• Nonsensical text. The Associated Press (AP) lists nonsensical text as a sign of AI imagery.
• Distorted objects, or objects that are missing parts. The AP lists these too, but Carrasco says missing limbs and fingers "just aren't a dead giveaway anymore."
• An account whose profile says its content is AI generated. Kolina Koltai told NPR that accounts often say this in their profile descriptions.

These signs are hints. They do not prove a clip is fake. Farid says even experts get fooled: "I've been doing this every day for a long time, and it's really hard."

For the wider problem of low-quality AI content, see our guide to AI slop.

Start with the Account and the Context

Start with where the clip was posted, what the comments say, and the poster's history. NPR suggests these checks.

NPR also suggests a reverse image search on a screenshot of the clip. That search can find the original post or the news coverage of it.

Read the profile before the caption.

Carrasco suggests asking "who would've filmed that and why?" He links odd camera movement to the lack of a real person behind the camera, so ask the question when a shot moves oddly.

Platform Labels and Watermarks

Labels and watermarks are signs, and both can be removed. A missing label does not prove that a clip is real.

YouTube requires creators "to disclose AI-generated or altered content that seems realistic." Creators make that disclosure in YouTube Studio, under Attributes > "AI use", by choosing Yes or No.

YouTube may label content automatically when it was made with its generative AI (GenAI) tools, when it carries Content Credentials metadata, or when its systems detect AI generation or alteration. YouTube does not require disclosure for unrealistic content, such as a unicorn ride, or for minor edits such as beauty filters.

TikTok says it labels AI-generated content uploaded from certain other platforms when that content carries Content Credentials.

TikTok said on November 19, 2025 that its labeling had "helped label over 1.3 billion videos to date." The same day, TikTok described two tests: an AI-generated content control in Manage topics, and invisible watermarking that only TikTok can read.

Meta renamed its "Made with AI" label to "AI info" across its apps on July 1, 2024. Meta bases the label on industry-shared signals and on self-disclosure.

Since September 12, 2024, Meta has shown the label in the post's menu for content it detects was only edited with AI tools. Labels on content detected as fully AI generated stay visible.

On May 30, 2025, Google said it was adding a visible watermark to all Veo videos, except videos generated by Ultra members in Flow, Google's AI filmmaking tool.

OpenAI's Sora 2 System Card, dated September 30, 2025, lists a visible moving watermark on videos downloaded from sora.com or the Sora app. OpenAI's Help Center says the Sora web and app experiences were discontinued on April 26, 2026. The watermark therefore matters for Sora clips still circulating online.

Watermarks can be removed. 404 Media reported on October 7, 2025 that it tested three watermark-removal websites. All three removed the Sora 2 watermark "seamlessly" in "a matter of seconds," according to 404 Media.

Checking the File with Content Credentials and SynthID

Content Credentials Verify is a free web tool. It reads Content Credentials attached to a file and shows how the file was created or edited. A file with no credentials does not prove that it is real.

Since December 18, 2025, the Gemini app can check an uploaded video for SynthID. Google describes SynthID as an imperceptible watermark. Upload the clip and ask something like "Was this generated using Google AI?"

Google's example reply reported SynthID in the audio from 10 to 20 seconds and none in the visuals. The check accepts files up to 100 MB and up to 90 seconds long.

Google says image and video verification are now available in all languages and countries that the Gemini app supports.

Google also launched SynthID Detector, a verification portal, in 2025 at synthid.com. Google says it can check whether an image, video or audio file was made with AI from Google or its partners, including OpenAI, NVIDIA, Kakao and, soon, Apple. That coverage is Google's claim.

Carrasco says Google's invisible watermark cannot identify AI content that Google's tools did not make. So a clean result from a Google check still leaves the other checks open.

OpenAI's own detection tools are internal. The Sora 2 System Card doesn't mention any public tool that lets a reader check a video with them. The same document lists "C2PA metadata on all assets" as a provenance measure, and OpenAI wrote, "We recognize that there is not a single solution to provenance."

What to Look for in the Footage

Hany Farid calls video length the biggest tip, and many AI clips run 8 to 10 seconds. That run is not a hard limit. Farid says longer videos can be stitched together from short cuts.

Next, check whether the camera movement makes sense for a real person filming. Farid says AI videos tend to frame their subject perfectly, and a camera that moves as smoothly as if it were on a gimbal can be a clue.

Then check text and objects. The AP lists nonsensical text, and distorted objects or objects that are missing parts, as signs of AI imagery. Carrasco says missing limbs and fingers are no longer a dead giveaway, so a clip with normal-looking hands still needs the other checks.

No percentage accuracy exists for spotting AI video by eye. Farid says even experts find it hard.

AI Video vs Deepfakes

Face swaps and other deepfakes are covered in our guide to how to spot a deepfake.

If none of these checks confirms where the clip came from, hold it back before you share it. This advice would change if OpenAI released a public tool for checking Sora clips, because right now there isn't one. For still images, see our guide to how to tell if an image is AI generated.

Sources

• OpenAI, Sora 2 System Card (September 30, 2025)
• OpenAI Help Center, What to know about the Sora discontinuation
• 404 Media, Sora 2 Watermark Removers Flood the Web (October 7, 2025)
• Google, Veo 3 expansion post (May 30, 2025)
• Google, verifying Google AI videos in the Gemini app (December 18, 2025)
• Google DeepMind, SynthID
• SynthID Detector
• YouTube Help, Disclosing use of GenAI content
• TikTok Newsroom, partnering with our industry to advance AI transparency and literacy (May 9, 2024)
• TikTok Newsroom, More ways to spot, shape and understand AI-generated content (November 19, 2025)
• Meta, Our Approach to Labeling AI-Generated Content and Manipulated Media
• NPR, AI video slop is everywhere (November 30, 2025)
• NPR, Simple tricks to spot AI-generated videos in your social media feed (December 17, 2025)
• AP via PBS NewsHour, Tripped up by misinformation? Here's a refresher on identifying AI (April 2, 2026)
• Content Credentials Verify