Methodology
How the platform metadata results are produced: one control file per format with every metadata container populated, uploaded through each platform's own apps, downloaded again, and compared byte by byte. No results are published yet — this page is the method.
Status: no measurements published. Testing has not started. When results appear they will carry the date they were taken and the app version tested, and this page will carry a changelog. Until then there is nothing here to cite.
Why this exists
"Does Instagram strip EXIF?" has answers all over the internet. Most are undated, many are copied from each other, and a good number were true three redesigns ago. Platforms change their processing without announcing it, so an undated answer is not much of an answer.
The only fix is to test, publish what was found, date it, and test again.
The test corpus
One control file per format, with every container deliberately populated — including the ones tools commonly miss:
| Format | Populated with |
|---|---|
| JPEG | EXIF with GPS, make, model, body serial and lens serial; IPTC creator and copyright; XMP packet; ICC profile; an embedded thumbnail carrying its own GPS; MakerNotes |
| HEIC | EXIF and XMP, ICC profile, depth and auxiliary images |
| PNG | tEXt, iTXt, zTXt and eXIf chunks, ICC profile |
| WebP | EXIF and XMP chunks |
| TIFF | EXIF, IPTC and XMP |
The thumbnail carrying its own coordinates is in there on purpose. A platform — or a stripping tool — that clears the main EXIF and leaves that thumbnail intact has not removed the location, and the only way to catch it is to test for it deliberately.
The procedure
- Upload the control file to the platform
- Retrieve the image the way a viewer would — not through an export or a data download, which frequently returns a different file
- Read every container in the retrieved file and compare against the control
- Record what survived, what was dropped, whether the image was re-encoded and at what quality, whether it was resized and at what threshold, and whether the ICC profile survived
- Repeat for each upload path — web, iOS app, Android app — because they routinely differ and almost nobody documents that
- Record the date and the app or build version where it is visible
Cadence
Quarterly retests, with the date of the most recent test shown on every result. An undated claim about a platform is worth very little, and a claim that was true last year may be actively misleading now.
Corrections
If a result here is wrong, it should be corrected rather than defended. Changes will be logged on this page with dates rather than edited silently.