Instagram’s Adam Mosseri has a blunt take on AI content: Instagram shouldn’t automatically hide it, but it also shouldn’t pretend it isn’t there. In a recent conversation on Lenny Rachitsky’s podcast, the head of Instagram argued that the platform’s job is not to act like a universal taste filter for everyone, but to make sure people can understand what they’re seeing—and then decide what belongs in their feed.
Mosseri’s position is easy to summarize and harder to implement at scale. He said he doesn’t think Instagram should filter out AI-generated content across the board. Instead, he believes the platform should clearly label AI content so users can tell the difference between something created by a person and something produced by an AI system. “I don’t think we should filter out AI content,” Mosseri said, adding that Instagram should “let you know if content is AI content or not.” The point isn’t just transparency for transparency’s sake; it’s about giving users control over their own experience without forcing one group’s preferences onto everyone else.
But Mosseri didn’t stop at labeling. He also made a more pointed argument about user choice: if you don’t like AI content, you shouldn’t have it in your feed. That line matters because it reframes the debate. The question isn’t only whether AI content should be allowed. It’s whether platforms should build systems that let people shape what they see—especially when the “what” is increasingly ambiguous. As generative tools become easier to use, the boundary between “real” and “synthetic” content gets blurrier, and the consequences of that blur extend beyond aesthetics into trust, context, and even safety.
What Mosseri seems to be advocating is a middle path between two extremes. On one side is blanket removal: a platform decides that AI content is inherently undesirable and blocks it. On the other side is laissez-faire distribution: a platform treats AI content as just another format and does nothing to distinguish it. Mosseri’s approach sits between them. Instagram would not ban AI content, but it would make it legible—so users who want less of it can reduce it, and users who want more of it can lean into it.
That “lean into it” part is where Mosseri’s comments get especially interesting. He suggested that people who enjoy AI content should be able to curate a feed that’s essentially “AI town.” In other words, the platform shouldn’t only protect users from unwanted content; it should also support communities that want to explore AI as a creative medium. This is a subtle but important shift in how platforms can think about moderation and ranking. Instead of treating AI content as a problem to be minimized, Instagram could treat it as a category to be organized—like any other content type that varies in style, intent, and audience.
This framing also reflects a broader reality: AI content isn’t going away. It’s already embedded in the way creators experiment with visuals, editing, and storytelling. Even when AI is used in small ways—enhancing images, generating backgrounds, remixing styles—the result can still look indistinguishable from traditional creation. If platforms respond by simply removing anything that might involve AI, they risk punishing legitimate creativity and confusing users who don’t understand the technical details behind a post. But if platforms do nothing, they risk eroding trust and leaving users unable to make informed choices.
Labeling is the obvious answer, but it’s also the most complicated one. A label sounds simple until you ask what it means in practice. Is it a label for fully synthetic images? For edits that use AI tools? For captions written by AI? For videos with AI-generated elements? For deepfake-like content? For content that is “AI-assisted” rather than “AI-made”? The more categories you create, the harder it becomes to apply them consistently. The more vague the label is, the less useful it becomes.
Mosseri’s emphasis on letting users know suggests Instagram is leaning toward a system that communicates enough information to be meaningful without turning every post into a technical disclosure. The challenge is that “AI content” is not one thing. It can be a stylized illustration, a realistic portrait, a manipulated video, or a marketing image designed to look authentic. Each of those cases carries different implications for deception, consent, and potential harm. A single label may not capture those differences, but it can still provide a baseline of transparency that many users currently lack.
Transparency also changes the social dynamics of content consumption. When users can identify AI content, they can adjust their expectations and interpret the post accordingly. That might mean treating it as art rather than evidence. It might mean being more skeptical of claims attached to the image. It might mean deciding whether to follow certain creators or communities. In that sense, labeling isn’t just about individual preference—it’s about collective literacy. Over time, users learn what kinds of AI content are common, what kinds are likely to be misleading, and what kinds are simply experimental.
Still, labeling alone doesn’t solve the core problem Mosseri points to: if you don’t like AI content, you shouldn’t have it in your feed. That implies some form of user control beyond passive disclosure. Users need a way to tune their feed so AI content appears less often—or not at all—without requiring them to block every account that posts it. The difference between “labeling” and “filtering” is crucial here. Labeling tells you what something is. Filtering changes what you see. Mosseri’s comments suggest Instagram wants to avoid a one-size-fits-all filter while enabling personalized control.
This is where Instagram’s ranking and recommendation systems come into play. Instagram doesn’t just show posts in chronological order; it curates feeds based on predicted interest. If AI content is labeled, Instagram can incorporate that information into ranking decisions. For example, if a user indicates they prefer fewer AI posts, the system can downrank labeled AI content for that user. Conversely, if a user engages heavily with AI-generated art, the system can upweight it. That would align with Mosseri’s “AI town” idea: the feed becomes a reflection of the user’s interests, not a uniform mix imposed by the platform.
But personalization introduces its own risks. If users can filter out AI content entirely, will that create echo chambers where certain types of information never reach them? Possibly. Yet the same is true for any content preference. The key difference is that AI content is uniquely capable of mimicking authenticity, which makes the stakes higher. If someone filters out AI content, they might miss creative work. If someone filters in AI content, they might be exposed to more synthetic material that could be misleading depending on context. The solution is not necessarily to prevent either behavior, but to ensure users understand what they’re choosing.
Mosseri’s stance also implicitly acknowledges that banning AI content is not a clean policy lever. AI generation is widespread and often integrated into normal workflows. Many creators use AI tools for legitimate purposes: style transfer, background generation, concept art, and accessibility features. Others use AI to produce spam, impersonation, or deceptive media. A blanket ban would catch both. A blanket allowance would also catch both. The practical policy question becomes: how do you separate benign from harmful uses without requiring perfect detection?
Detection is notoriously difficult. AI-generated content can be subtle, and the tools evolve quickly. Even if a platform can detect some AI content reliably, adversaries can adapt. That’s why Mosseri’s emphasis on labeling is significant. Labeling can be based on multiple signals, including metadata from tools, creator disclosures, and detection models. It doesn’t have to rely solely on perfect identification. Even partial accuracy can be useful if the label is applied consistently and users can control their feed accordingly.
There’s also a legal and ethical dimension. Platforms are increasingly expected to provide transparency about synthetic media, especially when it could mislead users. While laws vary by region, the direction of travel is clear: regulators and policymakers want clearer disclosure, not silent distribution. Mosseri’s comments fit that trajectory. If Instagram can communicate AI involvement to users, it reduces the chance that synthetic content is treated as indistinguishable from human creation.
At the same time, Mosseri’s approach avoids the trap of turning Instagram into an arbiter of truth. Labeling doesn’t guarantee that a post is false or true; it only clarifies how it was produced. That distinction matters. A labeled AI image could still be used to express genuine ideas or artistic interpretations. A labeled AI video could still be a parody or a creative reenactment. The label is a tool for context, not a verdict.
This is where Instagram’s unique culture comes into focus. Instagram is not just a news platform; it’s a visual social network where aesthetics, identity, and community norms are central. People follow accounts for styles, themes, and personalities. AI content already exists in that ecosystem, sometimes as art, sometimes as satire, sometimes as marketing. If Instagram were to ban AI content, it would disrupt communities and potentially push creators elsewhere. If Instagram were to allow AI content without disclosure, it would undermine trust and make it harder for users to interpret what they’re seeing.
Mosseri’s “transparency plus choice” approach tries to preserve the creative ecosystem while addressing the trust gap. It also suggests Instagram is thinking about AI content as a spectrum rather than a binary. Some AI content is obviously synthetic. Some is less obvious. Some is mixed with human creation. A mature policy would reflect that spectrum by providing users with information and controls rather than a single yes/no rule.
The “AI town” idea also hints at a future where Instagram could offer more structured experiences around AI. Imagine curated feeds or discovery modes that highlight AI art, AI music visuals, or AI-assisted photography techniques. That’s not just a moderation strategy; it’s a product strategy. It turns a contentious issue into a category that can be explored. It also gives creators a clearer path to reach audiences who want that kind of content. In that scenario, labeling becomes the bridge between creators and consumers: it helps the