85.6% of Social Media Users Want AI Content Labeled. How Do the Labels Actually Work?
Most AI labels on social platforms come from provenance metadata inside the file, which the platform reads and turns into a label. If that information is missing or stripped along the way, the label does not appear.
How many users want AI content labeled?
In Opensurvey's Social Media Trend Report 2026, fielded September 3 to 9, 2026 with 900 recent users of YouTube, Instagram and TikTok in Korea (300 each), 85.6% said AI-generated content on social media should be labeled. 54.1% said labeling is essential.
Platform | Say AI content should be labeled |
|---|---|
YouTube | 88.0% |
85.3% | |
TikTok | 83.3% |
What are users worried about?
The most cited concern was the spread of misinformation and fake news presented as fact (60.3%). Asked how AI content feels, respondents most often said it seems artificial or awkward (49.9%) or that they doubt whether it is factual (39.0%). Both point to authenticity, not to AI use itself.
How do platforms apply AI labels?
There are two routes. The uploader can self-disclose, or the platform can read provenance metadata embedded in the file. The main standards for the second route are C2PA Content Credentials and IPTC metadata. Meta announced in February 2024 that it would read both to label AI images on Facebook, Instagram and Threads, and renamed the label "AI info" in July 2024.
Self-disclosure | Automatic metadata reading | |
|---|---|---|
Who labels | Uploader | Platform, from the file |
Evidence | A checkbox | C2PA or IPTC metadata |
Weak point | Skipped if the uploader does not tick it | Lost if metadata is stripped |
Verifiable | Hard to verify | Tool and origin can be checked |
Why do AI labels disappear?
Metadata is attached to the file rather than to the pixels. Screenshots, re-encoding and re-saving in another app often remove it, so a copy that has been shared a few times may carry no trace of its origin. C2PA addresses this with soft binding: an invisible watermark in the content that lets a verifier look up the original provenance record even after the metadata is gone. How the two approaches differ is covered in C2PA vs watermarking.
Who is legally required to label AI content in Korea?
Under Korea's AI Basic Act, in force since January 22, 2026, the generative AI labeling provisions apply to AI business operators, not to individual creators. Korea's Broadcasting, Media and Communications Commission is reportedly pursuing legislation that would extend labeling duties to platforms that distribute AI-generated content (reported September 11, 2026; no effective date announced).
What is MuseBlossom Contents Defence?
MuseBlossom Contents Defence is a content provenance product listed on the C2PA Conforming Products List for images (June 11, 2026), audio (August 11, 2026) and, as of September 21, 2026, video, covering both generation and validation of Content Credentials. MuseBlossom's soft binding algorithm is listed on the C2PA soft binding algorithm list for images.
FAQ
Who adds AI labels on social media?
Either the uploader, through self-disclosure, or the platform, by reading C2PA or IPTC metadata in the file. Meta has used both standards for its "AI info" label since 2024.
Why doesn't my AI-generated image show a label?
The metadata may have been removed by a screenshot or re-save, or the tool may never have written provenance data. A platform can only read what is still in the file.
What is C2PA?
C2PA (Coalition for Content Provenance and Authenticity) is an open standard for recording and verifying who created or edited content and with which tools.
See how MuseBlossom Contents Defence attaches and verifies provenance: go.contentsdefence.com/blog
Sources
Opensurvey, Social Media Trend Report 2026, September 2026
Meta, announcement on labeling AI-generated images, February 6, 2024
Reporting on the Broadcasting, Media and Communications Commission's AI user protection bill, September 11, 2026
C2PA, Conforming Products List, as of September 2026