Instagram priced disclosure, not synthetic content
Instagram says labelled AI-person profiles keep their reach; unlabelled ones may lose recommendation eligibility. The controllable rule is disclosure. Detection remains opaque.
Instagram did not ban AI-generated people. On August 31, it said a profile featuring an AI-generated person can keep its recommendation eligibility by carrying the new AI-generated profile label. If Instagram detects the same kind of profile without that label, the account may become non-recommendable until the creator adds it or wins an appeal.
That is not a synthetic-media prohibition. It is a disclosure rule enforced through distribution. The synthetic person is allowed; the priced action is withholding the label, and the price is reach rather than money.
The label preserves recommendation eligibility
Instagram's August 31 announcement renamed its earlier “AI creator” label to “AI-generated profile”. The platform said creators who proactively add it will not see a change in reach. Unlabelled AI-person profiles that Instagram detects may be removed from recommendations.
The recovery path is equally direct. Add the label, or appeal through Account Status if the decision is wrong. A successful appeal restores eligibility for the profile and its content to appear in recommendations.
This is a useful piece of policy design because the operator can actually perform the requested action. Instagram has not made a judgement about whether synthetic creation is acceptable and hidden the answer inside a general authenticity slogan. It has named a disclosure, attached a distribution consequence and provided a route back.
This is not a rule for every use of AI
The scope is deliberately narrower than “AI content”. Instagram says creators who simply use AI tools in their creative process do not need the profile label. Its current Android help page, which I rechecked on September 11, says the account-level label is for profiles that regularly post an AI-generated person instead of a real human.
That profile label can surface beside the account in search, Feed, Reels, Stories, comments and messages. It is separate from the post-level AI content labels Meta has described for individual images, audio and video. In its content-labelling documentation, Meta says those labels can use self-disclosure or machine-readable signals such as C2PA and IPTC metadata.
Do not turn a profile rule into a blanket instruction to label every retouched photo, polished caption or generated graphic. The trigger is the person the account presents, not the mere presence of an AI tool somewhere in production.
The detector is the opaque side of the bargain
The announcement also says Instagram's systems might not always be perfect. What it does not publish is just as operationally important: the detection signals, decision threshold, error rates, number of affected profiles, enforcement start date or expected appeal time.
That absence does not prove Instagram lacks internal measures. It means a creator cannot audit the classification contract before it acts.
My claim is that the policy contains two systems. The public one is a disclosure contract: decide whether the profile features an AI-generated person, apply the label and preserve reach. The private one is a classification contract: Instagram decides whether an unlabelled account fits that description. You can comply with the first. For the second, you can only keep evidence and prepare to challenge a mistake.
Run the profile like a disclosure control
Steal this before an AI-person account goes live:
- Write one sentence defining why the account does or does not feature an AI-generated person; keep incidental editing outside that decision.
- If it applies, enable the profile label before distribution matters and save a dated capture showing it is active.
- Keep a private provenance record for the character, source assets, generation tools and material edits.
- Check Instagram Account Status after profile or character changes instead of waiting for reach to collapse.
- Assign an appeal owner and retain the labelled-profile capture, recent content and character provenance they would need.
Instagram does not list that provenance record as a requirement. It is my operating rule for making an opaque enforcement decision reviewable without pretending we can inspect the detector.
I did not run an AI-person account or test Instagram's detection system. I read the dated announcement, then rechecked the Android label and Account Status guidance on September 11. That verifies the public workflow and its stated scope, not detection accuracy, enforcement speed or recovery time.
A related archive note, reaches the same boundary from a different mechanism: provenance is an input to review, not a verdict. Here, the label is the controllable input; recommendation eligibility remains Instagram's decision.
Instagram made nondisclosure cost recommendation eligibility—apply the label you can control, and keep evidence for the classification system you cannot inspect.
