Does sharing an image remove its metadata? What each route does
It depends on the route, and no single answer covers every app. Vendors say screenshots, conversions, edits and re-uploads can remove provenance metadata. In our test, a browser re-save removed every Content Credentials record from the 9 real files that carried one. We have not measured any platform's upload.
Does sharing always remove metadata?
No, and nobody can say so for every route. Metadata is information stored inside the file: a Content Credentials record, EXIF camera data, XMP or IPTC labels, or text chunks. Some routes copy the file as it is. Others decode the picture and write a new one, and a new file holds only what the software chooses to write.
That is why one answer for "social media" or "messaging apps" does not hold up. Each app, and each way of getting a file into and out of it, can behave differently. We found no vendor page that lists what its upload does to each metadata type, so we do not give a table of platforms.
What do vendors say about sharing and metadata?
Four statements, each from a page we opened on October 30, 2026. Three are vendors' own words about their own files or labels, so they do not describe other tools. The fourth is IPTC's report about Google's guidance.
| Source | What it says |
|---|---|
| OpenAI, help article on provenance signals | Content Credentials can be lost through everyday actions such as a screenshot or a file conversion. If no signal is found, the file could still have come from OpenAI, for example if metadata was stripped during upload, download, editing, conversion or sharing. |
| Midjourney, Content Authenticity help page | The hidden ID tag in its images is easily lost. The page says "Screenshots, edits, and most social media sites remove it" and advises checking the original downloaded file. |
| TikTok, newsroom post of November 19, 2025 | TikTok says a common industry challenge is that labels like its own, which include ones read from C2PA metadata, may be removed when content is reuploaded or edited elsewhere. It says it is testing invisible watermarks as another safeguard. |
| IPTC, news post of February 20, 2024 (reported) | IPTC reports that Google reminded merchants who want their products listed in Google results not to strip embedded metadata, in particular the Digital Source Type field that can signal generative AI. IPTC says it has been saying for years that website publishers should not strip metadata from images. |
Sources: OpenAI's help article on provenance signals, Midjourney's Content Authenticity page, TikTok's newsroom post and the IPTC post about Google's reminder.
Why do routes differ?
A file can travel in three broad ways. These are general mechanics, not findings about any app.
- The same bytes. A copy or download that does not touch the file keeps whatever it holds.
- Metadata rewritten. Software that edits a file may keep some blocks and drop others. OpenAI's help article lists editing, conversion and sharing among the ways metadata gets stripped.
- The pixels only. A screenshot, or an export that decodes the picture and encodes it again, carries pixels and leaves the old file's blocks behind.
The third route matters most for signed records. A Content Credentials record, or the pointer to one stored online, sits in the file's metadata and not in the pixels, so a new file made from the pixels carries one only if the software writes it.
OpenAI also says the SynthID watermarks it uses are designed to withstand common changes, which it says can include screenshots, while heavy cropping or file compression can make a watermark undetectable. We do not check SynthID, so we cannot test that claim on our files.
What happened when we re-saved real files?
On October 30, 2026 we took the 9 real files in our test set that carry a Content Credentials record: 3 Adobe Firefly, 2 Microsoft Image Creator, 3 ChatGPT and the C2PA test image. Of our 32 real files, 29 came from Wikimedia Commons, and 6 of those 29 carry a record; the other three are two vendor samples and the test image. Their original headlines are counted on our accuracy page.
We decoded each file in headless Chrome 156, drew it on a canvas of the same size, and saved it again twice: once as PNG and once as JPEG (quality 0.92, ordinary 4:2:0 chroma subsampling). This is a stand-in for a screenshot or a browser export: the pixels carry over and the old file's blocks do not. Our built checker then read all 18 outputs.
- All 18 read "No supported provenance signals found".
- 0 of 9 records survived the PNG re-save, and 0 of 9 survived the JPEG re-save.
- No metadata of any kind was found in the outputs. A canvas holds pixels only, so it carries no record, XMP packet or text chunk across.
For the invisible Stable Diffusion pattern, the one invisible watermark our checker decodes, we used 20 fixture images that matched before. After the PNG re-save 19 of 20 still matched, and after the JPEG re-save 0 of 20 did. The one PNG miss carries a colour profile, and we did not test why it was lost.
The limits are firm. This is one browser, one JPEG quality and no resizing. It is a canvas, not a platform, and no platform was contacted. A real service might resize, convert colour, keep some metadata or add its own, so do not read these numbers as what any app does.
What this means when you check an image
Open the checker, choose the file, and look for the headline "No supported provenance signals found". The line under it says: "This does not mean the image is real or human-made. Most images lose these signals when they are shared, screenshotted or re-saved." It means we found no signed record, no source label and no Stable Diffusion pattern. It does not say where the file came from.
If a record does survive, you would see a headline such as "Content Credentials found; signer trust not established" or "Content Credentials found; legacy trust-list match". Those mean the file kept a signed record, and the line under each says how far the signer is vouched for.
A shared copy is a poor test of the original, so check the original downloaded file where you can. Our Privacy and method page explains what our checker does with your file.
Guides on this topic
- How to tell if an image is AI is the main page on this topic.
- Why Content Credentials disappear covers the ways a record gets lost.
- Does a screenshot keep metadata? looks at the screenshot route.
- Can EXIF data be faked? explains why unsigned metadata proves little.
- How to view image metadata shows what a file holds.
Frequently asked questions
- Do Instagram, WhatsApp or other apps remove metadata?
- We cannot say. We found no vendor page for those apps that states what an upload does, and we did not post test files to any platform. Our only measured route is a browser re-save, which is not a platform.
- Does a screenshot keep a Content Credentials record?
- OpenAI says Content Credentials can be lost through a screenshot or a file conversion. In our browser re-save test, 0 of 9 records survived.
- If metadata is gone, does that show the image has no source?
- No. A missing record or label means the file no longer carries one. It does not show that the image is real, human-made or from any particular maker.
- Is removing metadata always something a site chooses to do?
- Not always. OpenAI says metadata can be stripped during upload, download, editing, conversion or sharing. IPTC reports that Google reminded merchants not to strip embedded metadata, and IPTC says publishers should not strip it from images either.