How TikTok’s “AI-Generated” Label Works: C2PA, Watermarks, and Disclosure

TikTok puts an “AI-generated” label on content in three ways, and it says so directly. In July 2026 it reported labeling more than 3 billion videos as AI-generated content (AIGC) “using a combination of Content Credentials, creator labeling tools and our invisible watermarking technology.” Each of those is a different mechanism with different consequences for an image or video you post, so it is worth taking them one at a time.

#Three routes to the same label

#C2PA: the automatic label on upload

In May 2024 TikTok became, in its words, “the first video sharing platform” to implement Content Credentials, starting on images and videos. The mechanism is the C2PA manifest that generators such as ChatGPT and Adobe Firefly write into their files: “Content Credentials attach metadata to content, which we can use to instantly recognize and label AIGC.” Coverage at the time used a DALL·E 3 image as the example. If the manifest survives until upload, TikTok reads it; if an earlier step already dropped it, this route has nothing to read.

#The watermark only TikTok can read

In November 2025 TikTok announced a second layer: an invisible watermark on AI content made with TikTok tools such as AI Editor Pro and on “content uploaded with C2PA Content Credentials,” which it describes as “a robust technological ‘watermark’ that only we can read, making it harder for others to remove.” The practical effect is that a C2PA declaration read at upload gets re-anchored in the pixels. Stripping metadata from a later download would not remove it. By July 2026 TikTok was citing it among the methods behind its labels, but it has not named the watermarking method, given a rollout status, or offered a public way to check for it.

#What the Community Guidelines require

TikTok’s integrity and authenticity guidelines define AIGC as “any image, video, or audio made or changed by AI” and require creators to label AI-generated or significantly edited content “that shows realistic-looking scenes or people.” Unlabeled content “may be removed, restricted, or labeled by our team, depending on the harm it could cause.” Small edits such as color correction, reframing, or cropping do not need a label, and clearly stylized content is treated differently from realistic content. Some AIGC is not allowed even with a label, such as content made to look like a real news source or a fake political endorsement.

#Credentials on the way out, and viewer controls

TikTok also said in 2024 that it would attach Content Credentials to TikTok content so they remain after download, and its 2025 update describes adding them to AI-generated content made on TikTok. It has not confirmed that every download carries them, so check a downloaded file rather than assuming. On the viewing side, TikTok has been testing an AI-generated content control in “Manage topics” that lets people dial down how much AIGC appears in their For You feed. In July 2026 it also joined the C2PA steering committee.

#Before you post an AI image you own

1. Check the file at contentcredentials.org/verify so you know whether TikTok’s C2PA route will fire.
2. If the content is realistic, use the AI-generated label. The guidelines require it, and TikTok can apply it for you anyway.
3. Remember the layers are independent: a metadata strip removes C2PA from your copy, not an embedded signal such as SynthID or TikTok’s own watermark on content that already passed through TikTok.
4. If you are cleaning up provenance for privacy or a re-edit, our C2PA removal flow handles the metadata layer, and the platform labels overview compares TikTok with the rest.