Does TikTok watermark AI content? Its invisible mark, Content Credentials and the AI-generated label
TikTok said in November 2025 that it was testing invisible watermarks for content made with its own AI tools, and that it keeps reading and adding C2PA Content Credentials. TikTok says only it can read the watermark, and our checker does not look for it. TikTok's public checker covers content from five named apps.
What does TikTok say it adds to AI-made content?
TikTok's newsroom post More ways to spot, shape and understand AI content is dated November 19, 2025. We read it on October 30, 2026. It describes three things.
- An invisible watermark. TikTok says it was testing invisible watermarking. It calls the mark robust and says it is one that "only we can read". It says it planned to add the mark, over the coming weeks, to content made with TikTok tools such as AI Editor Pro and to content uploaded with Content Credentials.
- Content Credentials. TikTok says it reads C2PA Content Credentials and adds them to AI-made content created on TikTok.
- A label. TikTok says it requires people to label realistic AI-made content. It says it applies the rule through creator labelling tools, its own detection models and Content Credentials, and that these efforts have helped label over 1.3 billion videos to date.
The post also says labels may be removed when content is reuploaded or edited elsewhere, and presents the invisible watermark as an added layer of safeguards that is harder for others to remove. It does not claim the watermark survives every edit. We did not open a later TikTok page on the rollout, so we cannot say how far the watermark has reached.
Can you check a TikTok mark yourself?
TikTok runs its own page, Identify AI-Generated Content. It opens another company's site, and your file goes to them, not to us. We opened it on October 30, 2026. This is what the page says.
| Question | What TikTok's page says |
|---|---|
| Which content does it cover? | Content made or edited with AI tools from TikTok, CapCut, Dreamina, Hypic or Pippit. |
| What does it look for? | Two signals added by those apps: C2PA Content Credentials and invisible watermarks. |
| Which files? | One file at a time. Images up to 20 MB (JPEG, PNG, WebP). Video up to 100 MB (MP4, MOV, WebM, AVI). Audio up to 100 MB. |
| What happens to the upload? | TikTok says the file is processed only to check for signals and deleted right after the result. |
| What does a hit mean? | The content was "likely" made or edited with AI tools from those apps. If the signal is intact, the page says it also shows which tool. |
| What does no hit mean? | The file may not come from those apps, or may not be AI-made at all. TikTok adds that the Content Credentials may have been stripped, the watermark degraded, or the file may predate these signals. |
TikTok's page also says its tool covers only Content Credentials added by those five apps, and points to the Content Authenticity Initiative's Verify page for credentials from other sources. It says that if other platforms later modified a file, the tool cannot identify those changes.
Would our checker trust a record signed by TikTok?
Our checker compares a record's signing certificate with two trust lists. On October 30, 2026 we read the subject of every certificate in our copies of both. Our copies were fetched October 27, 2026.
| List and file | Date of our copy | Certificates | Naming TikTok |
|---|---|---|---|
| C2PA Trust List, signer anchors | October 17, 2026 | 32 | 0 |
| C2PA Trust List, time-stamp certificates | October 17, 2026 | 23 | 0 |
| Interim Trust List, anchors | January 8, 2026 | 27 | 0 |
| Interim Trust List, allowed certificates | April 21, 2026 | 115 | 1 |
The one match is a certificate whose subject names "TikTok Inc." and an email address at tiktok.com. By its own dates it is valid from May 13, 2024 to May 15, 2027. The Interim Trust List was frozen on January 1, 2026, according to the Content Authenticity Initiative's trust list page, and no new certificates are added to it.
So, by the rules our checker applies, a record signed with that certificate would read "Content Credentials found; legacy trust-list match", not as trusted on the current C2PA Trust List. That is a prediction from the list contents. We hold no TikTok-signed file and have not tested one, so we cannot say which certificate TikTok signs with, or whether it signs every file.
A name on a list does not show which products sign by default. We also do not read TikTok's invisible watermark.
What happens to these signals when a video is saved or re-uploaded?
TikTok says labels may be removed on reupload or editing elsewhere, and its checker page lists stripped metadata and a degraded watermark as reasons for a missed signal. We can add one measurement of our own, with limits. On October 30, 2026 we drew each of the 9 real files that carry a Content Credentials record onto a canvas in Chrome and saved it again as a PNG and as a JPEG. After both, 0 of 9 still showed a record. Our accuracy page shows how we measure.
That was a canvas, not TikTok. We did not upload anything to TikTok or any platform, and the files were images, not videos. It shows what re-drawing pixels does to metadata in files we hold. It does not show what TikTok does when a video is uploaded.
What this means when you check an image
Open the checker and choose an image file. If it carried a record whose signer is on the Interim Trust List only, you would see the headline "Content Credentials found; legacy trust-list match". The line under it says who the certificate names and that the signer is on the frozen list, not the current C2PA Trust List. It means a signed record is present and the signer is on a list we load. It does not say the claims in the record are true.
The checker reads images only. A video file such as an MP4 or MOV reads "Unable to check this file", as it did for the short test clips we made on October 30, 2026. That headline says nothing about the video. For an image with no record and no label, "No supported provenance signals found" does not mean it is real or human-made: these checks cannot determine whether an image is AI-generated.
For something you saw on TikTok, the maker's own checker is the place to look for TikTok's invisible mark. Our Privacy and method page explains what our checker does with your file.
Guides on this topic
- What is SynthID? is the main page on invisible watermarks.
- TikTok's AI-generated label covers how the label is applied.
- Invisible AI watermarks compared sets the different marks side by side.
- Which platforms read or show Content Credentials? compares TikTok with other platforms.
- Can AI watermarks be removed? covers what survives saving and re-uploading.
Frequently asked questions
- Does TikTok put a watermark on every AI video?
- We cannot say. TikTok's post says it was testing an invisible watermark for content made with its own tools, and its checker covers content from five apps only.
- Is TikTok's invisible watermark the same as SynthID?
- TikTok's post does not link its mark to SynthID, and we found nothing on the pages we read that does. It describes its own mark that only TikTok can read.
- Does our checker read TikTok's watermark?
- No. We do not check it. The only invisible watermark our checker decodes is the pattern added by Stable Diffusion's reference software.
- If TikTok's checker finds nothing, what does that mean?
- TikTok says it could mean the file is not from its apps, or that the signal was lost or never added. It is not a statement about whether the content is real.