PixelSeal
Experiment12 min ยท March 3, 2026

How Every Major Platform Strips Your Image Metadata

A systematic study of metadata preservation across 12 platforms โ€” Twitter/X, Facebook, Instagram, LinkedIn, Discord, Reddit, Imgur, Pinterest, Tumblr, WhatsApp, Telegram, and Signal.

metadataprovenanceplatformsexifiptc

The Metadata Problem

EXIF, IPTC, and XMP metadata is the traditional method for establishing image provenance. But what happens when you share an image? We tested 12 major platforms to find out exactly what metadata survives the upload/download cycle.

๐Ÿ”ฌMethod
This experiment uses PixelSeal's standard embed + verify pipeline. You can reproduce results using the developer tools.

Test Setup

We created a test image with comprehensive metadata: EXIF (camera model, GPS, date, artist, copyright), IPTC (headline, caption, byline, keywords), and XMP (creator, rights, description). We uploaded to each platform and downloaded the result.

Results: The Metadata Wasteland

11 out of 12 platforms stripped ALL EXIF data. 12 out of 12 stripped IPTC data completely. Only Telegram preserved some EXIF fields (camera model, date). No platform preserved copyright or creator information. GPS data was stripped by every single platform โ€” the one case where metadata stripping is actually beneficial for privacy.

What This Means for Creators

If you rely on metadata for image provenance, your attribution is destroyed the moment someone shares your work on social media. This is not a bug โ€” platforms strip metadata intentionally for privacy, storage, and performance reasons. Pixel-level watermarking is the only approach that survives this processing pipeline.

๐Ÿ’กKey Finding
Twitter/X: Strips all metadata, recompresses to Q85, caps at 4096px. Facebook: Strips all metadata, recompresses Q80-Q85, applies content-adaptive quality. Instagram: Strips everything, Q75-Q85, caps ...

Platform-by-Platform Breakdown

Twitter/X: Strips all metadata, recompresses to Q85, caps at 4096px. Facebook: Strips all metadata, recompresses Q80-Q85, applies content-adaptive quality. Instagram: Strips everything, Q75-Q85, caps at 1080px. LinkedIn: Strips metadata, minimal recompression. Discord: Strips EXIF GPS/artist, preserves some camera data for images under 8MB, compresses larger files.

Reproduce this experiment

Seal your own image and verify it after applying transforms. The same tools we used are available to you right now.

Frequently Asked Questions

Is metadata stripping always bad?

Can I re-embed metadata after downloading?

Does C2PA/Content Credentials solve this?