Instagram's AI disclosure policy is narrower than the panic around it suggests, and it runs on two tracks that people constantly merge. The rule is not "label anything AI touched" — it is "label realistic AI," meaning video that looks real or audio that sounds real, made or altered with AI. On an organic post you disclose that with one switch: the "Add AI label" toggle you flip after tapping Next on a feed post or Reel, or through More, then Label AI on a Story. AI-written captions and scripts are exempt, and plain AI still images don't require the toggle — though Meta may auto-add an "AI info" label when it reads AI signals in the file's metadata. Over the top of your self-disclosure sits Meta's automatic detection, which keys off C2PA Content Credentials and similar provenance data rather than the pixels, and an account-level "AI-generated profile" label that, since August 31, 2026, carries a reach penalty for synthetic-persona accounts that hide it. The second track is ads, and it is stricter: social-issue, electoral, and political ads have a hard, long-standing requirement to disclose AI-created or altered content during the buying flow, Meta auto-labels ad creative it detects as AI, and an AI label never substitutes for the paid-partnership and FTC endorsement disclosure a branded post already owes. This guide walks the whole policy — the realism line, the organic toggle on every surface, the metadata layer, the profile label, and the separate, tougher ad track — so you know exactly what trips the rule before you ship AI content to Instagram at any volume.
The policy is narrower than the noise around it, and getting the boundary right is the whole game. Instagram does not ask you to flag every file an AI tool touched. It asks you to disclose AI content that is realistic — video that looks real, or audio that sounds real, created or significantly altered with AI — because those are the pieces a viewer could mistake for something that actually happened. A photorealistic AI video, a real clip edited with AI to change what it shows, a song with AI vocals, a Reel with a realistic AI voiceover: all cross the line. The question is realism and the potential to mislead, not AI-involvement.
What trips most creators is that the policy runs on two tracks that enforce the same principle with different controls. Organic posts are governed by Instagram's own content rules and a creator-facing toggle. Paid ads are governed by Meta's advertising policies, with a separate and in one key case much stricter obligation. The two share the realism test underneath, but the surface you use, the review you face, and the penalties differ — and treating them as one policy is how a compliant creator still gets an ad rejected. We take the organic track first, since that is where most content lives, then the metadata layer that sits over it, then the account-level profile label, then the ad track that the context around this policy most often overlooks.
For organic content, disclosure is a single switch, and it lives in a slightly different place on each surface. On a feed post or Reel, upload as normal, tap Next, turn on the toggle labeled "Add AI label," then Share. On a Story, share your photo or video, tap More, tap Label AI, flip "Add AI label" on, and share. The label then appears to viewers as post metadata — not buried in a submenu — marking the piece as made with AI across Facebook, Instagram, and Threads. This self-set label is the reliable path precisely because it does not depend on Meta's detector reading your file correctly.
The step-by-step for every surface, including how to appeal a wrong flag, is the companion piece: how to disclose AI-generated content on Instagram. The important property of the toggle is that it is additive and low-cost to use — there is no penalty for labeling a borderline piece, which matters because the realism line has genuine gray areas. When you are unsure whether something could mislead, the asymmetry of the rule rewards disclosing it.
Label it when the content is AI-made or AI-altered and realistic. Instagram's own examples cover AI-generated videos that look real, real videos edited with AI, songs with AI vocals, and Reels with realistic AI voiceovers. Meta specifically requires labeling photorealistic AI video and realistic audio, and warns of penalties for not labeling content that needed it. The through-line is the realism test every major platform has converged on: if a reasonable person could take the synthetic media for a real person, place, or event, disclose it.
The exemptions are where creators over-correct. AI-written captions and scripts do not need the label — the text being machine-drafted doesn't fabricate a realistic depiction of anything. AI-generated or AI-edited still images are not required to carry the toggle either, although Meta may add an "AI info" label to them on its own if it detects AI signals. An obviously stylized illustration, a cartoon, or a plainly graphic quote card generally stays outside the rule, because nobody is being deceived about reality. Flagging a real photograph where you removed a dust speck misrepresents your work in the other direction, so precision cuts both ways.
Self-disclosure is only half the system. Meta runs an automatic detector that can add an "AI info" label even when you skip the toggle, and knowing how it works stops you from assuming an unlabeled upload is safe. The detector does not analyze your pixels — it reads the file's paperwork. It keys off machine-readable provenance signals, chiefly C2PA Content Credentials and similar "AI-generated" metadata that image generators and editing tools write into exported files, and it labels content accordingly.
That design has a defining consequence: the detector can tag a real photo that only passed through an AI-powered editing step while a fully synthetic image slips through once its metadata is stripped. It is unreliable in both directions, which is the entire reason the self-disclosure toggle is the part you actually control. The full operating manual for that detector — why real photos get flagged, how to appeal through Account Status without scrubbing provenance, and how the two labels interact — is in Instagram AI-content detection and labeling, and the cross-platform picture of how these label systems work is in AI content labeling in 2026.
Separate from any single post sits an account-level label. If a profile is built around an AI-generated person — a virtual influencer or other synthetic persona — Meta expects it to carry the "AI-generated profile" label, which you set under Edit profile. This is not the per-post "AI info" tag; it declares that the identity behind the whole account is synthetic, and it is independent of whether each individual post is realistic AI.
The part with teeth arrived on August 31, 2026, when Instagram renamed the former "AI creator" label to "AI-generated profile" and said accounts featuring an AI-generated person that skip it may see reduced reach, while accounts that apply it are not penalized for being AI. So for a synthetic-persona account, disclosure now protects distribution and concealment costs it. Crucially, the rule is narrow: a real creator who merely edits photos, polishes captions, or makes graphics with AI does not need the profile label and is not affected. The full account-level playbook is Instagram's AI-profile disclosure rules, with the step-by-step at how to label an AI-generated Instagram profile.
Paid content does not run on the organic toggle, and here the policy diverges sharply. The firmest, longest-standing obligation is for ads about social issues, elections, or politics: advertisers must disclose when such an ad has been digitally created or altered in ways that are consequential to its claim — a photorealistic image or video, realistic audio, a person who doesn't exist, an event that didn't happen, or a real person depicted saying or doing something they didn't. Meta announced this requirement in November 2023 and it took effect in 2024; the disclosure is made during the ad-buying flow, Meta then marks the ad and logs it in its ad library, and it rejects ads that fail to disclose, with repeated failures risking penalties against the advertiser. Inconsequential edits like resizing, cropping, or sharpening are exempt unless they are material to the claim.
For ads more broadly, Meta applies an AI label to creative it detects as AI-generated, surfacing it in the ad's information panel so viewers can see how the ad was made. Meta's posture here is disclosure and context, not prohibition — AI ad creative is allowed, it just has to be legible. Beyond platform policy, statutory rules are now layering on top of all of this: several US states have passed AI-ad disclosure laws, and the mechanics of complying with them are in how to comply with state AI-ad disclosure laws, while the EU's labeling obligation is covered in the EU AI content-labeling law playbook. The step-by-step for disclosing AI ad creative itself is how to disclose AI-generated ads.
One trap sits at the seam of the two tracks: creators assume that flipping the AI label on a branded post covers their disclosure duty. It does not. Instagram's AI label and the paid-partnership disclosure are answers to two different questions — "was this made with AI?" and "was I paid to post this?" — and a sponsored AI post owes both. The paid-partnership tag, and the FTC's endorsement standards it maps to, exist independently of anything about AI; the FTC also bars fake reviews and testimonials, including AI-generated ones attributed to people who don't exist. For regulated niches like finance or health, an AI disclosure can even create its own compliance exposure, which is why branded AI content is the case most worth routing past a compliance check. The broader influencer-trust framing is in AI creator sponsorship transparency.
The enforcement is distribution-based, not a fine. On the organic side, Meta can add its own "AI info" label to content it detects as AI, and it warns of penalties for failing to label realistic content that needed it; a synthetic-persona account that hides the "AI-generated profile" label can stop being recommended to new people, with the owner notified and able to add the label or appeal. On the ad side, undisclosed social-issue, electoral, or political AI creative gets rejected outright, and repeat failures can draw penalties against the advertiser account. In every case the cost is the reach or the spend you lose when a post is throttled or an ad never serves — which, for content built specifically to perform, is the loss that matters.
Disclosure is also not a universal permission slip. Labeling content does not make otherwise-prohibited material allowed; AI ads and posts still have to clear Meta's underlying content standards, so the toggle answers "is this real?" and does nothing for "should this exist?" Those stay separate questions, the same way they do on TikTok's AI disclosure policy.
The policy is trivial for one post a week and becomes a process problem the moment you produce realistic AI at volume — which is the premise of generating content with AI in the first place. At scale, two questions have to be answered for every piece: does this trip the realistic-AI line that needs the organic label, and is this shipping as an organic post or as paid ad creative that falls under the stricter track? Miss either and the failures accumulate into throttled posts and rejected ads you only notice weeks later in the analytics.
Kompozy is a full AI content generation and multi-platform publishing engine — not a repurposing tool — and the reason it helps here is that the format you choose already answers the first question. Persona Shorts and Persona HeyGen put a synthetic person on screen; Persona Photos are face-locked AI images; Marketing Shorts and Persona VFX produce the kind of realistic avatar-led creative that typically runs as a paid ad. Those are the label-worthy, ad-track-sensitive outputs. A text post, a quote graphic, an AI-written blog, or a clip of your own real footage is not something anyone is deceived by. Because you generated the asset, you never have to defer the realism call to Meta's unreliable metadata reader — the determination travels with the format.
Where that pays off is the second question, the one specific to this policy: track routing. Kompozy's per-post review pipeline means nothing publishes straight from the model — every piece lands in a review queue first, and that gate is where you confirm the organic AI-label decision and, just as importantly, flag which assets are destined to become paid Instagram or Meta ad creative so they get the stricter ad-track treatment and the branded-content disclosure they also owe. From one Persona Brief, Autopilot fans output across the eight social platforms plus blog and email, and the single review step is what keeps the Instagram toggle from being set and then forgotten on the TikTok and YouTube cross-posts. The honest boundary: Kompozy does not flip Instagram's "Add AI label" toggle for you, run your paid ad buys, or make your Ads Manager disclosures — those live in your Instagram and advertiser accounts and stay yours. What it removes is the reason disclosure drifts at scale: the uncertainty about what each asset is, and which of the two tracks it belongs on.
Instagram asks you to disclose AI content that is realistic — video that looks real or audio that sounds real, created or altered with AI — not every file AI touched. On organic posts you disclose with the "Add AI label" toggle (tap Next on a feed post or Reel, then flip it; on a Story, tap More, then Label AI). Meta also auto-labels files that carry AI provenance metadata. AI-written captions and scripts are exempt, and plain AI still images don't require the toggle, though Meta may add an "AI info" label itself.
When the content is AI-made or AI-altered and realistic: photorealistic AI video, a real video edited with AI to look real, a song with AI vocals, or a Reel with a realistic AI voiceover. You do not need it for AI-written captions or scripts, and still images don't require the toggle — though Meta may auto-label them. The test is whether a reasonable person could mistake the media for a real person, place, or event, not whether AI was involved at all.
Yes — the ad track is separate and stricter. Social-issue, electoral, and political ads have a hard requirement to disclose AI-created or digitally altered content during the ad-buying flow; Meta rejects ones that don't and can penalize repeat offenders. For ads generally, Meta auto-labels creative it detects as AI in the ad's information panel. And an AI label never replaces the paid-partnership and FTC endorsement disclosure a branded or sponsored post already owes — those are different obligations that stack.
By reading the file's metadata, not the image. Meta keys off machine-readable provenance signals — C2PA Content Credentials and similar "AI-generated" fields that generators and editors write into exported files — and applies an "AI info" label when it finds them, even if you skipped the toggle. Because it reads metadata, it can tag a lightly edited real photo and miss a fully synthetic image once the metadata is stripped. That unreliability is exactly why self-disclosing realistic AI yourself is the dependable path.
For organic posts, Meta may add its own label when it detects AI and can apply penalties for not labeling realistic content that needed it; an undisclosed synthetic-persona account can stop being recommended to new people as of August 31, 2026. For ads, Meta rejects undisclosed social-issue, electoral, or political AI creative and can penalize repeated failures. The practical cost is lost distribution or a dead ad, not a fine — which for content built to perform is the whole point.
Kompozy is a full AI content generation and multi-platform publishing engine, and some of what it makes — HeyGen avatar video, face-locked persona images — is the realistic AI the policy targets, while text posts, quote graphics, and carousels sit clean outside it. Because you generate the asset in the engine, you already know which track it belongs to. Kompozy's per-post review gate is where you route each piece to the right disclosure decision before it ships organically; the Instagram toggle and your Ads Manager disclosure stay yours.
Instagram's AI disclosure policy asks you to label realistic AI — video that looks real or audio that sounds real, made or altered with AI — not every AI-touched file. On organic posts you flip the "Add AI label" toggle (tap Next, then the switch; on Stories, More, then Label AI). Meta also auto-labels files carrying AI metadata. Ads follow a stricter track: social-issue, electoral, and political ads must disclose AI-created or altered content during the buying flow, and an AI label never replaces paid-partnership disclosure.
Get started → · ← All guides · Compare Kompozy vs other tools