Before you try to remove an "AI label": three different things people mean

Before you remove an "AI label", work out which of three things you are looking at.

A five-minute triage.

"Three things, one name"

(1) A badge or chip on a post ("AI info" on Instagram/Facebook, a CR icon on LinkedIn) lives in the platform. (2) Provenance tags inside the file (C2PA Content Credentials, IPTC DigitalSourceType, prompt text in a PNG) live in the file's metadata. (3) An invisible watermark (SynthID and similar) lives in the pixels.

"Step 1: why is the badge there?"

Meta says it adds "AI info" when it detects industry-standard AI indicators in the file (it names the "AI generated" information in the C2PA and IPTC standards), or when the person posting discloses the content as AI-generated. So either a signal in the file reached the platform, or a box was ticked at upload. Check the post settings before blaming the image.

"Step 2: read the exact file you uploaded"

Use the export you posted, not a screenshot. Run: exiftool -a -G1 -s photo.jpg | grep -i digitalsource . A value of trainedAlgorithmicMedia means generative AI made the image. For a C2PA manifest use the CAI's c2patool or their web verifier. If neither shows anything, the file did not trigger the badge and cleaning it again changes nothing.

"Step 3: decide what you want"

File carries a tag and you have a legitimate reason to publish without it: remove the metadata, re-scan to confirm, upload as a new post. No tag but badge still there: the source is the disclosure toggle, a classifier or a watermark; use the platform's appeal route if the label is wrong. If a rule, client or policy asks you to disclose AI use: disclose it.

"What this does not fix"

A label already on a published post stays until you replace the upload. A watermark in the pixels stays. No cleaner can promise a platform will not label something.

"Longer version"

I wrote a longer breakdown, with a free in-browser inspect-clean-verify tool: what removing an AI label from a photo can and cannot change . One image is free. Disclosure: this is my project. Not affiliated with Meta, Instagram or the C2PA.# Before you try to remove an "AI label": three different things people mean

Before you remove an "AI label", work out which of three things you are looking at.

A five-minute triage.

"Three things, one name"

(1) A badge or chip on a post ("AI info" on Instagram/Facebook, a CR icon on LinkedIn) lives in the platform. (2) Provenance tags inside the file (C2PA Content Credentials, IPTC DigitalSourceType, prompt text in a PNG) live in the file's metadata. (3) An invisible watermark (SynthID and similar) lives in the pixels.

"Step 1: why is the badge there?"

Meta says it adds "AI info" when it detects industry-standard AI indicators in the file (it names the "AI generated" information in the C2PA and IPTC standards), or when the person posting discloses the content as AI-generated. So either a signal in the file reached the platform, or a box was ticked at upload. Check the post settings before blaming the image.

"Step 2: read the exact file you uploaded"

Use the export you posted, not a screenshot. Run: exiftool -a -G1 -s photo.jpg | grep -i digitalsource . A value of trainedAlgorithmicMedia means generative AI made the image. For a C2PA manifest use the CAI's c2patool or their web verifier. If neither shows anything, the file did not trigger the badge and cleaning it again changes nothing.

"Step 3: decide what you want"

File carries a tag and you have a legitimate reason to publish without it: remove the metadata, re-scan to confirm, upload as a new post. No tag but badge still there: the source is the disclosure toggle, a classifier or a watermark; use the platform's appeal route if the label is wrong. If a rule, client or policy asks you to disclose AI use: disclose it.

"What this does not fix"

A label already on a published post stays until you replace the upload. A watermark in the pixels stays. No cleaner can promise a platform will not label something.

"Longer version"

I wrote a longer breakdown, with a free in-browser inspect-clean-verify tool: what removing an AI label from a photo can and cannot change . One image is free. Disclosure: this is my project. Not affiliated with Meta, Instagram or the C2PA.