Clear answers about sharing paths, AI watermarks, content provenance, and regulation—built from official documentation, technical standards, primary law, and clearly labeled independent research.
Three synthetic images, real cleanup results. Compare removed metadata, decoded pixels and file hashes, then repeat the test yourself.
Explore the JPEG, PNG and WebP experiment →Each verdict separates what the sources support from what remains uncertain—and ends with a safer action you can take now.
Messaging apps
A 2025 independent study found that WhatsApp image-mode transfers removed the tested EXIF fields, while sending the same images as documents preserved their metadata and file hashes. Treat that result as mode- and version-specific, not as a permanent WhatsApp privacy guarantee.
Social platforms
A 2025 independent study found that Instagram uploads were re-encoded and the tested critical EXIF fields were removed from the resulting image. That is useful evidence about the published copy, but it is not proof that Instagram collected no location, account, device, or upload metadata separately.
Yes—when a photo is attached as a file, email normally transports the file data rather than sanitizing it. MIME specifies reversible transfer encodings, and a 2025 controlled study found identical hashes and complete tested EXIF retention for email attachments. Photo-app resize options can be an exception.
Apple sharing
Yes. Apple’s current iPhone guide says sharing a photo includes associated metadata such as date, time, location, device, and captions. The share-sheet Options screen lets you turn Location off; enabling All Photos Data for AirDrop sends the original file with edit history and metadata.
Cloud storage
Do not rely on Google Drive to remove embedded metadata. Google’s current help documents uploading, sharing, downloading, versions, and file conversion as separate actions; it does not describe link sharing as a metadata sanitizer. The conservative privacy assumption is that a downloaded uploaded file can retain its embedded metadata.
AI provenance
OpenAI, Google, Amazon, Microsoft, Adobe, and Meta document provenance signals for at least some generated media, but they do not all use the same mechanism. C2PA is an open metadata standard; SynthID and Amazon watermarks need compatible provider detection. Anthropic says Claude currently produces text and text-to-speech output and does not claim an equivalent deployed watermark.
AI regulation
The EU AI Act Article 50 marking and detection obligations apply from 2 August 2026, with separate disclosure duties for deepfakes and certain public-interest text. China’s explicit and implicit AI-content labelling measures have applied since 1 September 2025. The United States still lacks one equivalent general federal watermark mandate; provenance bills remain proposals unless enacted.
BeforeShare did not run the platform tests cited in these briefs. We synthesize current sources, show who produced each one, and mark the gap between a documented fact and a conservative inference.
Product-owner instructions or disclosures. Best for current controls and declared behavior.
A protocol specification. Best for what a format or transport is designed to do.
Third-party controlled testing. Useful evidence, always bounded by its dataset, versions, and method.
Platforms change. Inspect the exact recipient-facing copy when the answer matters.
Inspect the hidden fields first, then create a cleaned share copy locally in your browser.