← All posts

How Meta's “AI info” label works — and what it reads

November 18, 2026

If you've posted an AI-generated image to Facebook or Instagram and seen an "AI info" label appear, you might wonder how the platform knew. The answer isn't magic — it's metadata.

The two signals platforms read

When you export from ChatGPT, Grok, Midjourney, Adobe Firefly and similar tools, the image usually carries provenance data:

Meta and other platforms scan for these on upload and attach the label automatically. The image content itself isn't analyzed for the label — the tags do the talking.

Removing the tags

Because these are metadata, they can be stripped. A tool like Barefile reads the C2PA and DigitalSourceType tags, shows them to you, and removes them along with the rest of the hidden data. See removing AI metadata & C2PA.

The honest limit: invisible watermarks

Here's what most "remove AI detection" tools won't tell you: some generators also embed an invisible, in-pixel watermark — Google's SynthID is the best known. That watermark lives in the image's pixels, not its metadata, so no metadata tool can remove it. Anyone promising your AI image will be completely undetectable is overselling. Stripping metadata removes the tags that trigger platform labels today; it does not guarantee an image can never be identified as AI.

We think the honest version is more useful: remove what's actually in the metadata, and be clear about where the line is.

Clean a file right now

See what's hidden in your images, video and PDFs — and strip it, in your browser.

Open Barefile
How Meta's “AI info” label works — and what it reads · Barefile