Open any ad on Instagram or Facebook in 2026, tap the three dots, and a line reading "AI info" now sits in the "About this ad" section on a growing share of them. Meta added that label in stages through the year and now flags AI-generated material on its own, without waiting for the advertiser to declare it. For a photorealistic AI-generated person, the label can move out of the menu and sit next to the "Sponsored" tag in the feed itself.
Labeling replaced takedowns
Meta stopped removing deepfake videos in July 2026 unless they break some other rule, so a manipulated clip stays up with a label attached and only comes down when it violates a Community Standard like voter interference, or when the Take It Down Act applies and still forces non-consensual intimate imagery to be pulled within 48 hours. For anything the deepfake rules do not reach, Meta leaves the clip on the feed and lets the label do the work of telling viewers what they are looking at.
The rollout came in two stages
The rollout ran in two stages. Organic posts started carrying a "Made with AI" label in May 2026. That label came partly from creators disclosing it themselves and partly from Meta reading industry-shared signals in the file. The advertising side went automatic in July. Background Generation, Image Generation and Add Animation inside Ads Manager now put an "AI info" label on the ad with no advertiser input. Undisclosed AI content sat at around 14% of ad rejections by 2026, the third most common reason on the platform. A first violation kills the ad and adds a policy strike. A second within 90 days triggers a 24-hour account hold. A third can push the account into suspension.
What trips the AI label
What trips the label is which part of the ad was generated. Background Generation, Image Generation and Add Animation inside Ads Manager trigger it directly. Third-party tools like Photoshop generative fill, DALL-E, Firefly and Canva trigger it too, because Meta reads the C2PA Content Credentials those tools bake into the file and labels the ad the moment it is uploaded. Cropping, brightness and manual color correction do not trip it. Generating the main subject of an image, a person, a product render or a whole scene, generally does.
Content Credentials is the standard doing the detection. It is a block of signed metadata that records how a file was made and edited, written in by tools from Adobe, OpenAI and other vendors as they export the file. It stays inside the file through re-uploads. Meta reads the credential at upload and labels the ad on that, rather than analyzing the pixels frame by frame.
Where the label shows up
The label lives in a few places. On most ads it sits inside "About this ad," reached through the three-dot menu next to the "Why am I seeing this ad?" details. On an ad whose main subject is a photorealistic AI-generated person, it moves up beside the "Sponsored" tag where the feed shows it without a tap. Meta also applies it after the fact. An ad approved and already running can be flagged once the system reads AI in the creative, which pulls the campaign mid-flight rather than catching it at review.
The AI info panel shows one of two lines. Either the ad's content was created or edited with Meta generative AI features, or Meta detected the use of third-party generative AI tools. The first is anything built inside Ads Manager. The second is a file that arrived already carrying a generative signal. Both open from the same panel, and both link out to Meta's Ad Library.
The disclosure is public. It appears in Meta's Ad Library, the searchable archive of active ads, on the same card as the spend range and the regions the ad reached. It stays there for the life of the campaign, readable by anyone who looks the advertiser up.
Organic posts get their own tag
Organic posts use a separate tag, "Made with AI." Creators switch it on with a toggle when they publish a reel, story or post. Meta applies it automatically when shared industry signals flag photorealistic AI video, audio or images in the upload. On organic content the tag is disclosure and not a strike, and Meta has said it does not cut reach.
How to disclose it yourself
Disclosing by hand is a single control. At the ad level in Ads Manager, under the creative, a toggle marks the ad as containing AI-generated content, and it carries into every placement in the ad set. An advertiser who used a third-party tool flips it there. An advertiser who used Meta's own generation features does not need to, since the label is already attached by the time the creative is built.
What it means in production
For a brand on Meta the change lands in production. An ad whose hero image came out of Firefly, or whose background was filled in with Image Generation, ships with the label attached once it reaches Ads Manager. A crop, a brightness tweak or a manual color pass does not add it. The generated subject does.
The practical read for a media team is short. Anything built with Background Generation, Image Generation or Add Animation is labeled. So is anything run through Firefly, DALL-E or Canva that keeps its Content Credentials. So is a swapped background or a generated face. A stock photo cropped and color-graded by hand is not. Sorting assets into those two buckets before upload is what keeps a campaign out of the rejection pile.
Meta is not the only one
Meta is not the only one. Google switched on a "How this ad was made" panel across Search, YouTube and Discover on 9 July 2026, labeling ads built with its own AI automatically and leaving third-party tools like Midjourney or Runway for the advertiser to declare by hand. TikTok has read the same C2PA credentials to auto-label AI clips since 2024, and treats the label as disclosure rather than a reach penalty, so tagged content still earns through the Creator Fund. All three read the metadata the same way, so a file flagged on one platform tends to get flagged on the next.
Sources: Meta Business Help Center, Meta Transparency Center, Google Blog, TikTok Newsroom.
