You enable Content Credentials in your camera or attach them in export, upload the photo, and then what? For most platforms the honest answer is: the provenance record is read by the platform, used for its own labeling decisions, and deleted from the file everyone actually sees. A couple of platforms do better. Knowing which is which changes how much weight you should put on credentials reaching your audience, and what to rely on instead. Status as of mid 2026; platform behavior changes, and we update this page when it does.
The pattern: read at the door, strip on the way in
Social platforms re-encode nearly every image and video at upload. Re-encoding produces a new file, and metadata that is not deliberately carried over dies in the process, which is how EXIF data has been vanishing from social uploads for a decade. C2PA manifests live in the same layer of the file and meet the same fate.
What has changed since 2024 is the reading half. The major platforms now inspect provenance metadata at ingest, because C2PA and IPTC fields are the most reliable signal they have for labeling AI content. So the typical pipeline is: your metadata informs the platform's label, then the platform discards it. The label survives as the platform's own annotation; your verifiable record does not. As of this writing, only LinkedIn displays inbound Content Credentials as credentials, and only TikTok has announced re-attaching credentials to content so provenance survives download.
Platform by platform
| Platform | Strips C2PA from served files | Reads it at upload | Shows anything |
|---|---|---|---|
| Instagram / Facebook / Threads | Yes | Yes | "AI info" label on AI-flagged content |
| Yes (standard send) | No labeling | Nothing; sending as document preserves the file | |
| TikTok | Historically yes; re-attachment announced | Yes | Auto "AI-generated" label from C2PA markers |
| YouTube | Yes (full transcode) | Yes | "Captured with a camera" and synthetic-content disclosures |
| Displays credentials | Yes | The CR icon with a provenance panel | |
| X (Twitter) | Yes | Reads EXIF at upload | No provenance display shipped |
| Largely undocumented | Yes (IPTC) | "AI modified" label since April 2025 |
Meta: Instagram, Facebook, Threads, WhatsApp
Meta joined the C2PA steering committee in September 2024 and reads both C2PA and IPTC markers at upload across Instagram, Facebook, and Threads. What it does with them is labeling: content flagged as AI-generated gets the "AI info" label introduced in February 2024. The label's history is instructive for photographers. It launched as "Made with AI" and was renamed within months after real photographs, lightly retouched with tools that write AI-related metadata, started getting branded as AI. That episode is a case study in the false positive problem: the pipeline reads a marker, not the truth, and a Photoshop generative-fill dust removal can label an authentic photo. Since September 2024 the label sits less prominently, in the post menu, for content that was AI-edited rather than AI-generated.
What Meta does not do is preserve your credentials. Files served or re-downloaded from its apps carry neither EXIF nor C2PA manifests, and Meta keeps the provenance record server-side for its own purposes. WhatsApp strips metadata in a standard photo send as well, with one useful exception: sending an image as a document transmits the original file untouched, credentials included. That makes WhatsApp-as-document one of the few mainstream ways to hand someone a file with its provenance intact.
TikTok and YouTube
TikTok became the first video platform to implement C2PA reading in May 2024. Content arriving with AI-generation markers, from tools like DALL-E 3, gets an automatic "AI-generated" label. TikTok also said it would begin attaching Content Credentials to content made in the app, so provenance survives download, and in November 2025 it added something more robust: an invisible watermark applied to AI content, explicitly because, in its own framing, C2PA metadata can be removed when content is re-uploaded or edited. A platform building a second marking system because it expects the first to be stripped tells you what the platforms themselves think about metadata survival.
YouTube transcodes everything, so inbound file metadata does not survive into the streams viewers watch. But since October 2024 it reads C2PA at upload for two disclosures in the video description: a synthetic-content disclosure, and its inverse, a "Captured with a camera" note for footage whose C2PA record (version 2.1 or later) shows an unedited camera capture. That inverse label matters: it is the first large-platform use of provenance to positively mark authenticity rather than to flag AI, which is the direction photographers should want this whole system to move.
LinkedIn, the exception
LinkedIn displays the official CR icon on images that arrive with Content Credentials, and clicking it opens the provenance panel: what created the image, when, whether AI was involved. It is the one major network where credentials attached to your photo are visible to viewers as credentials, rather than being consumed and discarded. Whether the manifest also survives inside re-downloaded files is not officially documented, so treat the display as the confirmed part. For photographers whose clients live on LinkedIn, this is the platform where enabling credentials pays off visibly today.
X and Pinterest
X strips EXIF from posted photos and has shipped no provenance display. The company joined C2PA back in 2021 and prototyped credentials display before its acquisition, but nothing reached users. In January 2026 Elon Musk teased an edited-visuals labeling feature; no details or rollout have followed as of this writing, and third-party claims that credentials display already exists on X are not supported by any primary source. Assume your metadata dies at X's door.
Pinterest reads IPTC metadata and has shown an "AI modified" label on image pins since April 2025, supplemented by its own classifiers for unmarked content. Its handling of C2PA manifests in served files is undocumented; the safe assumption is that they do not survive.
What this means for photographers
Three practical conclusions fall out of the matrix.
Credentials are for your channels, not for reach. Enable them if your camera or phone offers signing, but the audience that sees them is on your website, in professional workflows, and on LinkedIn. On every other platform in the table, attaching credentials mostly feeds the platform's own labeling machine.
Labels are not your record. A platform label is the platform's opinion, derived from markers it read and sometimes misread; the "Made with AI" mislabeling wave proved how that lands on real photos. Your defensible record is the one you keep: the camera original, and where you want it, the signed file itself, checkable in any inspector regardless of what platforms display.
The record that survives every pipeline is the one that never enters it. Re-encoding cannot strip what does not travel in the file. A RAW original in your archive, compared against the published JPEG, establishes that the image is a faithful development of a real capture, and that comparison works on the mangled, metadata-free copy a platform serves, because it examines pixels against your original rather than metadata that was long since deleted. That is the verification model Lumethic runs, and platform stripping is precisely the situation it was built for.
Frequently Asked Questions
Does Instagram remove Content Credentials from photos?
Yes. Instagram re-encodes uploads and the served files carry neither EXIF nor C2PA manifests. The metadata is read first and can trigger the "AI info" label, but the record itself does not survive into what viewers see or download.
Which platforms actually display Content Credentials?
LinkedIn displays the CR icon with a provenance panel on images that arrive with credentials. YouTube surfaces C2PA-derived disclosures, including "Captured with a camera", in video descriptions. TikTok shows an AI-generated label based on C2PA markers. The Meta apps show their own "AI info" label without exposing the underlying credentials, and X shows nothing.
Why did real photos get labeled "Made with AI" on Instagram?
Because the label was triggered by metadata markers that editing tools write even for minor AI-assisted retouching. Photographers who removed a dust spot with a generative tool carried the same marker as fully generated images. Meta renamed the label to "AI info" in July 2024 and later made it less prominent for edited content.
How can I share a photo without losing its credentials?
Serve the file from your own site, use C2PA-aware delivery in professional workflows, or send it as a document rather than a photo in messengers like WhatsApp. Any pipeline that re-encodes the image will almost certainly remove the manifest.
If platforms strip everything, how do I show a disputed photo is real?
With the original. Metadata-based provenance dies in platform pipelines, but a comparison between your camera original and the published image does not depend on anything the platform preserved. Verify the image against its RAW and the resulting report answers the question no matter which platform mangled the copy.
The system that is supposed to carry trust across the internet currently loses its cargo at most borders. Until that changes, attach credentials where they survive, expect labels rather than provenance everywhere else, and keep the one record no pipeline can touch: your original, and a verification that ties the published image back to it.