Meta's AI label relies on two different trigger paths

The test

Meta's own newsroom published 'Our Approach to Labeling AI-Generated Content and Manipulated Media' on 5 April 2024, announcing a revised labeling approach for Facebook, Instagram and Threads ahead of a major election year, later updated within the same post on 1 July and 12 September 2024 as the rollout continued.

What the evidence says

The original announcement states the change was made 'based on feedback from the Oversight Board that we should update our approach to reflect a broader range of content,' replacing a narrower 2020 manipulated-media policy that the post describes as covering only videos altered to make someone appear to say something they did not say. The revised approach applies a 'Made with AI' label to 'a wider range of video, audio and image content' when Meta detects 'industry standard AI image indicators' embedded by other companies' tools, or when a person discloses that they are uploading AI-generated content themselves -- two distinct triggers, one automated and one self-reported, that the post treats as separate paths to the same label. A later update within the same post, dated 12 September 2024, narrowed where the label appears for some content: material Meta detects as only 'modified or edited by AI tools' moved the label to the post's options menu, while content Meta assesses as fully AI-generated kept the more visible label and continued to disclose whether it was flagged by shared industry signals or by self-disclosure. Meta's own newsroom index, checked separately, confirms this post sits within Meta's ongoing series of policy updates rather than as a one-off release.

The sample and the variance

This is Meta's own account of its labeling mechanics, not an independent audit of detection accuracy or an estimate of how much AI content circulates on its platforms unlabeled. The post itself does not state a detection rate or an error rate, and Meta's own updates to the same post show the policy has already changed twice since the original announcement.

What to try next

As an editorial checklist, a creator or brand publishing AI-assisted short-form content on Facebook or Instagram can treat self-disclosure as the documented, controllable path to correct labeling rather than assuming an embedded technical signal will always be detected.

• Does the AI tool used to produce the content embed the industry-standard signals Meta says it detects, or would self-disclosure be the only way the label gets applied?
• Is the content fully AI-generated or only AI-edited, the distinction the September 2024 update says determines label placement?
• Has Meta published a further update to this same post since 12 September 2024?

A labeling policy that already carries three dated updates in one post is best read as a moving mechanic, not a fixed rule. Checking the post's own latest revision remains part of using it correctly.