There is a lot of confident commentary about which platforms preserve Content Credentials and which destroy them. Very little of it is measured. We decided to measure it.
Method
We produced 4,000 signed assets: 3,200 stills across four formats, and 800 short videos. Each was uploaded to eight platforms through the normal consumer path, downloaded again, and checked for an intact manifest, a recoverable soft binding, or nothing at all.
Every upload was repeated three times over six weeks, because platform behaviour changes without announcement and a single run tells you about a single day.
Results
- Stills: full manifest survived on two of eight platforms. Soft binding recoverable on five.
- Video: full manifest survived on none. Soft binding recoverable on two, and only for clips over twenty seconds.
- One platform preserved manifests on stills consistently across all three runs, which was not the one we expected.
- Two platforms changed behaviour mid study, both in the direction of stripping more.
The variance between runs was larger than the variance between platforms. That is the finding that should worry people.
What surprised us
The assumption that image formats behave similarly is wrong. One format lost its manifest on six platforms while a visually identical export in another format survived on four. The difference is entirely in how each platform normalises on ingest.
Short video is close to hopeless today. If your work is video first, a signed original on a domain you control is not a nice extra, it is the only durable claim you have.
Limitations
Consumer upload paths only. No business APIs, which often behave differently and usually better. Results are a snapshot of a moving target, and we will rerun this in six months.
The dataset and the harness are available to researchers on request.