How Instagram's "AI info" label works, and when removing metadata changes it
Short answer: Instagram adds the "AI info" label when Meta detects industry-standard AI signals in a file (C2PA or IPTC metadata written by the tool that made or edited it), or when the person posting discloses AI use. Since September 2024, posts that were only edited with AI show the label in the post's menu, not on the post. Metadata signals disappear when the metadata is removed. Your own disclosure, and Meta's other detection methods, do not.
Where the label comes from
Meta has described its approach in two newsroom posts: Labeling AI-generated images (6 Feb 2024) and Our approach to labeling AI-generated content (5 Apr 2024, updated since). Put together, there are three routes to a label.
1. Metadata written by the AI tool
Many image generators and editors write a note into the file saying AI was involved. Meta says it looks for the "AI generated" information defined by two standards:
- C2PA (Content Credentials): a signed manifest that records which tool created or edited the file, and what it did.
- IPTC: a
DigitalSourceTypefield in the photo's XMP metadata, with values such astrainedAlgorithmicMediafor a fully generated image.
In February 2024 Meta named Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as companies whose images it planned to label as they added this metadata. This route explains most surprise labels on real photos: an editor's AI feature (generative fill, object removal) writes the same kind of record as a generator does, even for a small retouch.
2. Your own disclosure
When you post, Instagram lets you add an AI label yourself. Meta says it requires this for realistic video or audio made or altered with AI, and may apply penalties to accounts that don't. That disclosure is a choice you make in the app, not something in the file.
3. Meta's own detection
Images made with Meta's own AI features carry visible markers, invisible watermarks and metadata. Meta has also said it is building classifiers to detect AI content that has no markers at all, precisely because metadata can be stripped. Meta's posts don't say how widely these run today, so assume they can label a file regardless of what its metadata says.
"Made with AI" vs "AI info": what changed
| Date | Change (from Meta's own update notes) |
|---|---|
| Feb 2024 | Meta announces labels for AI images based on C2PA and IPTC signals. |
| May 2024 | Labels start appearing on a wider range of images, video and audio. |
| 1 Jul 2024 | "Made with AI" is renamed "AI info", after photographers complained that minor AI retouching earned a label. Tapping it shows more information. |
| 12 Sep 2024 | Content only edited with AI tools gets the label in the post's menu instead of on the post. Content detected as generated keeps the visible label, which says whether industry signals or self-disclosure triggered it. |
So if you only see "AI info" behind the three-dot menu, Meta's systems read your post as edited, not generated.
Which labels removing metadata affects
| Why the label appeared | Does removing metadata change it? |
|---|---|
| C2PA manifest from a generator or editor | Yes, for a new upload of the cleaned file. The manifest is gone. |
IPTC DigitalSourceType in XMP | Yes, for a new upload. XMP is removed with the rest. |
| You switched on the AI label when posting | No. It isn't in the file. |
| An invisible watermark in the pixels | No. A watermark is part of the image itself. |
| A classifier judged the image AI-made | No. Classifiers look at the pixels. |
| A post that is already live | No. Instagram labelled the copy it received. |
To see which case you are in, check the file before you post it. The AI info tag checker looks for C2PA markers in the browser, and the image metadata viewer lists the XMP and EXIF fields, including the software that wrote them. If neither finds an AI signal and the post still gets a label, it came from the disclosure setting or from Meta's own detection, and no metadata tool will change it.
When removing metadata is the right fix
The legitimate case is a real photo carrying a record from a small AI edit, where the label tells viewers something that isn't true of the picture. The AI info tag remover runs in your browser: it draws the image to a canvas and saves a new file, which drops EXIF, XMP, IPTC and C2PA together. Nothing is uploaded. JPEGs are re-saved at quality 0.95, so keep your original. The tool accepts JPEG, PNG and WebP; convert iPhone HEIC photos first.
The opposite case matters too. If an image is generated, or a realistic video was made with AI, the label is accurate, and Meta's rules expect you to disclose it. Removing the metadata doesn't change what the image is, and it removes the record that shows where it came from. C2PA content credentials explained covers what that record contains.
What this doesn't protect against
- Labels that don't come from metadata. Self-disclosure, invisible watermarks and classifiers are untouched by any metadata tool. See how to check if an image is AI-generated for what metadata can and can't tell you.
- Changes to Meta's system. Meta calls its labelling methodology "still evolving". What triggers a label today may not be what triggers it next year.
- Posts already published. Cleaning your copy doesn't change a post Instagram has already labelled.
Common questions
Why does my real photo say "AI info" on Instagram?
Most often because an editing app wrote C2PA or IPTC metadata after you used an AI-powered tool, even for a small fix. Check the file in the image metadata viewer and look at the software and XMP fields.
Is "AI info" the same as "Made with AI"?
Yes. Meta renamed "Made with AI" to "AI info" on 1 July 2024, and in September 2024 moved the label into the post menu for content that was only edited with AI.
Can I see why Instagram added the label?
Tap "AI info" on the post. Since September 2024 Meta says the label indicates whether it came from industry signals or from self-disclosure.
Does Instagram remove the metadata when I upload?
Instagram reads AI signals in the file you upload; that's how the label is added. What it keeps in the copy it serves is a separate question, covered in which platforms strip EXIF metadata.