# Why does Instagram show “AI info” on my photo?

> What Meta says triggers the “AI info” label, why edited real photos get it, and what removing C2PA and IPTC metadata does and does not change.

- Last reviewed: 2026-11-16
- Evidence confidence: Moderate
- Canonical page: https://beforeshare.com/en/metadata-lab/why-instagram-shows-ai-info-label

## Short answer

Meta says it applies “AI info” when it detects industry-standard AI indicators in a file or when the person posting discloses AI use. The indicators it names are C2PA Content Credentials and the IPTC digital source type, alongside invisible watermarks. Because editing apps can write those fields after a small generative edit, a real photo can be labelled. Since September 2024 Meta says content it detects as only edited with AI tools carries the label in the post’s menu, while content it detects as AI-generated keeps a visible label.

**Verdict:** Usually metadata an editing or AI tool wrote into the file, or your own disclosure.

## What the evidence supports

Meta’s February 2024 announcement says it reads the “AI generated” information defined by the C2PA and IPTC technical standards, and names Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as companies adding that metadata. In July 2024 it renamed “Made with AI” to “AI info”, saying labels based on these indicators “weren’t always aligned with people’s expectations” when a photo had only been retouched. IPTC’s guidance defines the value AI generators should write, trainedAlgorithmicMedia, and the C2PA specification defines the signed manifest that travels with the file.

## Important distinction

A label that comes from file metadata and a label that comes from the content are different things. C2PA manifests and IPTC fields sit beside the pixels and are absent from a cleaned copy. Invisible watermarks such as SynthID, Meta’s own detection and a disclosure you make yourself do not depend on that metadata, so removing it cannot promise that a label will not appear.

## Evidence ledger

1. **Official documentation — Meta.** [Labeling AI-Generated Images on Facebook, Instagram and Threads](https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/)
   - Published: 2024-02-06
   - Finding: Meta says it detects the “AI generated” information in the C2PA and IPTC standards, names the companies adding it, acknowledges that invisible markers can be stripped, and requires people to disclose photorealistic video or realistic-sounding audio.

2. **Official documentation — Meta.** [Our Approach to Labeling AI-Generated Content and Manipulated Media](https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/)
   - Published: 2024-04-05 (updated 2024-07-01 and 2024-09-12)
   - Finding: The 1 July 2024 update renames “Made with AI” to “AI info” after retouched photos were labelled. The 12 September 2024 update moves the label to the post’s menu for content detected as only edited with AI tools and keeps it visible for content detected as AI-generated.

3. **Technical standard — IPTC.** [IPTC publishes metadata guidance for AI-generated “synthetic media”](https://iptc.org/news/iptc-publishes-metadata-guidance-for-ai-generated-synthetic-media/)
   - Published: 2023-05-09
   - Finding: IPTC recommends that software creating images with trained AI algorithms writes the Digital Source Type value trainedAlgorithmicMedia into the file’s metadata, or carries it in a C2PA manifest.

4. **Technical standard — C2PA.** [C2PA Technical Specification 2.4](https://spec.c2pa.org/specifications/specifications/2.4/specs/C2PA_Specification.html)
   - Published: 2026-04
   - Finding: The specification defines Content Credentials as a signed manifest stored with the asset that can record the tools and actions used, including generative AI; it is separate from the image pixels.

## What remains uncertain

- Meta does not publish the full list of signals it uses or how it weighs them, and its detection can change without notice.
- BeforeShare has not tested Instagram’s labelling by posting files; this brief reports what Meta and the standards bodies state.
- Feed posts, Stories, Reels, ads and Threads may be handled differently, and the label’s name and position have both changed since 2024.

## Safest action

1. Inspect the exact file you plan to post. If it lists a C2PA manifest or an IPTC digital source type, you have found the most likely trigger.
2. If the photo is real and only retouched, export a copy without Content Credentials or clean the metadata, and keep the original.
3. If the image is AI-generated, disclose it where Meta’s rules or the law require it; cleaning metadata does not remove that duty.
4. Do not treat a clean file as proof that an image is not AI: missing metadata proves nothing.

## Questions people ask

### Does removing metadata remove the “AI info” label?

It removes the C2PA and IPTC fields Meta says it reads, so a newly posted copy no longer carries that trigger. It cannot affect invisible watermarks, Meta’s own detection, or a label already applied to an existing post. Treat it as removing one signal, not as a guarantee.

### Why did Instagram label a real photo I only edited?

Some editing tools write AI indicators into the file when a generative feature is used, even for a small retouch. Meta acknowledged in July 2024 that such labels did not match what people expected and renamed the label; since September 2024 it says edited-only content shows the label in the post’s menu.

### Do I have to label AI content on Instagram myself?

Meta requires people to disclose organic content with photorealistic video or realistic-sounding audio that was digitally created or altered, and says it may apply penalties when they do not. Laws in some places add their own duties, which the regulation brief covers.

## Verify your own file

Inspect the exact file you plan to share: https://beforeshare.com/en/view-metadata

Check a file for AI metadata: https://beforeshare.com/en/remove-ai-metadata

This brief is a source-labeled synthesis, not a claim that BeforeShare independently tested every platform, client, or version.
