Meta's "AI info" label on Facebook, Instagram and Threads: what triggers it and what it proves
Meta says it adds an "AI info" label when it detects industry-standard AI indicators in a file, or when the person posting says the content was made with AI. On Meta's account, the label reflects one of those two things. It does not tell you how much AI was used, or whether the post is true.
What does the "AI info" label mean on Instagram and Facebook?
The label is Meta's note that AI may be involved in a post. In its April 5, 2024 post on labeling, Meta says it adds the label to video, audio and images in two cases. The first is when it detects "industry standard AI image indicators". The second is when people disclose that they are uploading AI-generated content. Its February 6, 2024 post names Facebook, Instagram and Threads.
What triggers the label?
There are two triggers, and Meta's pages name both.
- Signals in the file. In its February 6, 2024 post, Meta says it was building tools to read invisible markers, specifically the "AI generated" information in the C2PA and IPTC standards. It named Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as companies whose images it planned to label as they add such metadata. That was a statement of plans in 2024.
- The poster's own disclosure. In the same February 2024 post, Meta said it would require people to use its disclosure tool for organic posts with photorealistic video or realistic-sounding audio that was digitally created or altered, and that it may apply penalties if they do not.
For its own tool, Meta said in February 2024 that photorealistic images made with Meta AI carry visible markers, invisible watermarks and embedded IPTC metadata, and that it labels them "Imagined with AI". Our guide to Meta's invisible watermark covers that part.
Why did "Made with AI" become "AI info"?
Meta renamed the label on July 1, 2024. It says labels based on industry indicators did not always match what people expected. Its example is content with minor AI changes, such as retouching tools, that carried indicators and was labeled "Made with AI". The new name can be clicked for more information.
On September 12, 2024, Meta added a second change. For content it detects was only modified or edited by AI tools, it moves the label into the post's menu. For content it detects was generated by an AI tool, it keeps the label on the post and says whether the label came from industry-shared signals or from self-disclosure.
What does the label prove, and what does it not?
By Meta's own description, the label means that Meta detected an industry-shared signal, or that the poster made a disclosure. In the pages we read, we found no passage that calls the label proof that a picture is true, false, edited or untouched.
Meta also states the limits. In its February 2024 post it said it was not yet possible to identify all AI-generated content, that people can strip out invisible markers, and that it was working on classifiers for content that lacks markers. It called this an area likely to become more adversarial.
Two consequences are ours, not Meta's. A label is a statement about a signal or a disclosure, not a test of the pixels. And no label may only mean that Meta found nothing it reads and the poster said nothing; it does not mean the picture is real or human-made.
What does Meta's own page say, side by side?
| Topic | What Meta says | Source and date |
|---|---|---|
| Triggers | Industry-standard AI indicators detected, or self-disclosure by the poster. | Meta, April 5, 2024 |
| Name change | "Made with AI" became "AI info" because labels were not always in line with expectations. | Same post, July 1, 2024 update |
| Placement | On the post for content generated by an AI tool; in the post's menu for content only modified or edited. | Same post, September 12, 2024 update |
| Limits | Cannot identify all AI content; invisible markers can be stripped. | Meta, February 6, 2024 |
| Ads | "About this ad" will include the "AI info" labels Meta already applies to ads made or significantly edited with its generative tools; Meta says it will begin detecting ads made with third-party AI tools through industry-standard signals. | Meta, February 3, 2025, updated June 1, 2026 |
How often is the label shown?
Meta's transparency page says labeling began in May 2024. For October 1 to October 29, 2024, it reports over 360 million labeled Facebook items and over 330 million labeled Instagram items with user label views. Meta calls this a snapshot in time and says the data may not capture labels on some content edited with AI.
What did we find in real files?
Meta says its label depends on industry-standard indicators. We can count how many of those indicators a set of real files carries, using what our checker reads. This is our measurement, made on October 27, 2026.
We ran 31 real files filed as AI-made: 29 from Wikimedia Commons categories named "Images generated by ..." and 2 vendor samples. 9 carry an indicator our checker can read: 8 signed Content Credentials records and 1 unsigned IPTC declaration. 22 carry neither: 16 show no signal our checker reads, and 6 carry only other metadata, such as a tool name. Both files filed under Meta AI are in the 16. 9 of 31 is 29%, so a label that relied only on those indicators would see fewer than a third of these files.
Commons files were uploaded by third parties, and we cannot know how each was handled before upload. Our checker does not read Meta's watermark, and we ran no upload through Meta's apps, so this is not a count of Meta labels. The full tables are in our accuracy notes.
What this means when you check an image
Save the picture and open the checker. It reads your file in your browser and never uploads it. Two headlines matter most here.
- "No supported provenance signals found". We say this does not mean the image is real or human-made, and that most images lose these signals when shared, screenshotted or re-saved. If Meta showed a label on the post, it may have used a signal that was removed from your copy, or the poster's disclosure.
- "Metadata declares AI use". The file carries an unsigned label. Anyone can add or remove it; it is a label, not proof. In February 2024 Meta said it was building tools to read the "AI generated" information in the C2PA and IPTC standards.
Our results always say these checks cannot determine whether an image is AI-generated, and that Meta's watermark was not checked here. If you can see an "Imagined with AI" label, tick the box on the checker to add a link to Meta's identification tool, which is reported as a preview. That link opens another company's site, you may need to sign in, and your image is uploaded to them, not to us.
Where to read next
This guide belongs to our page on visible AI watermarks. Related guides:
- Meta's invisible watermark
- Meta's AI info labels on ads
- Which platforms label AI images
- Which platforms show Content Credentials
Frequently asked questions
- Does an "AI info" label mean the whole image is AI-made?
- Not necessarily. Meta says minor AI changes such as retouching carried indicators that led to labels. In September 2024 Meta said content it detects was only edited by AI tools would get the label in the post's menu.
- Can a post be AI-made and have no label?
- Meta says so. In February 2024 it said it could not identify all AI-generated content, and that people can strip invisible markers. No label is not evidence that a picture is real or human-made.
- Does the New Watermark checker see what Meta sees?
- No. It reads Content Credentials, unsigned metadata and one open invisible watermark. It does not read Meta's watermark. Meta's pages name C2PA and IPTC information; we found no complete list of everything it reads.