Instagram Limits Recommendation Reach for Unlabeled AI-Generated Character Accounts
Instagram announced that it will limit the reach of accounts featuring AI-generated characters that do not self-identify: posts will no longer be recommended to non-followers on Reels and Explore. The label has been changed from "AI Creator" to "AI-Generated Profile" to indicate that the character is not a real person.
The company stated that users are frustrated when they initially believe a character is real, only to later discover it is AI-generated. Unlabeled accounts will be notified and can check their account status to see if they have lost recommendation eligibility; adding the label or appealing can restore eligibility. Accounts that create individual content using AI tools but are still run by real people do not need to add the profile-level label.
Accounts that proactively label themselves claim their recommendation reach remains unchanged. The system may misjudge, and the official allows for appeals. This measure targets entire character accounts and does not address the filtering of other generated content in the information stream. Spanish virtual model Aitana Lopez and others have been used as case studies by the media.
Adam Mosseri, the head of the platform, has made multiple adjustments to recommendations and labeling during his tenure. Brands that buy ads featuring unlabeled characters will face reduced recommendation levels. Whether traffic can be maintained after labeling depends on whether the algorithm treats the label as a quality signal.
In market mechanisms, the cost is borne by brands that want to buy the trust premium of real people, while the sale involves the distribution of unlabeled virtual characters. This event is driven by platform rules, shifting funds from the hidden character accounts' thousands of exposures to labeled or real accounts; benefiting are virtual characters willing to label themselves and still receive business orders, while those relying on impersonating real people for recommendations face pressure.
Source: Public Information
ABAB AI Insight
The platform is penalizing not the generated images, but the generation of characters without disclosure. Limiting recommendations is cheaper than deleting accounts: the accounts still exist, just not visible on the Explore page. The label changed from "Creator" to "Generated Profile," shifting identity from professional to existential. Individual content can be unlabeled, but an entire account with a fake identity must be labeled, with the dividing line drawn at the profile picture.
The capital path is the thousands of exposures in influencer e-commerce. Money flows from brand budgets into accounts with recommendation eligibility. If unlabeled characters drop out of Reels, business order quotes will need to be renegotiated. The cost of virtual characters is computational power, while the cost of real influencers is their face and availability; the platform uses labels to separate and price these two types of supply. Appeals serve as a valve for misjudgments and a step down for insufficient enforcement precision.
The benchmark is that advertisements must disclose they are ads, and election content must disclose funding sources. Social recommendations are at a stage where "generated content is cheap, but generated identities still want to enjoy the trust of real people for free": the dregs of content have already flooded, and the identity layer becomes the next barrier. Not filtering generated videos in the information stream indicates that products still require those seconds, just not that kind of deceptive profile.
Structurally, this represents a transfer of pricing power. The mechanism is: when recommendations determine business order prices, disclosure becomes a permit for entry. Unlabeled characters can still post, but lose the platform's subsidized distribution. The trust premium has been pushed from the algorithm's shadows to the label bar; those unwilling to label must bear the cost of customer acquisition themselves.
ABAB News · Law of Cognition
The platform can choose not to delete accounts, only to turn off non-follower recommendations.
Generated content can remain unlabeled, but generated identities often must be labeled.
Once the trust premium is pushed to the label bar, those who do not label must acquire customers themselves.