Meta is telling your customers your ad was made by AI

Half of American consumers say they would rather spend their money with a brand that doesn't use generative AI in its advertising. That's Gartner, published March 2026, off a survey of 1,539 US adults.

You could ignore that number for a while, because your customers had no reliable way to tell. Meta closed the gap on June 1.

Since then, Meta has run automated detection across ad media to spot images and video created or edited with generative AI. Not only its own tools. It reads the provenance data that Photoshop, DALL-E, and a growing list of other editors write into the file, and when it finds a match it attaches a label reading "AI info" to your ad. Nobody asks your permission. There is no checkbox to decline.

On July 7 Meta moved that label into a new "About this ad" panel, one tap behind the three-dot menu on every promoted post. For most ads it stays there. For an ad whose main subject is a photorealistic AI-generated person, the label can move up beside the word "Sponsored," in the feed, where your customer reads it without tapping anything.

What triggers the label?

Two paths.

The first is Meta's own creative tools inside Ads Manager. Background Generation, Image Generation, and Add Animation all mark their output. If you clicked the button that invented a new background behind your product, Meta knows, because Meta built the button.

The second is third-party software. Meta names Photoshop and DALL-E, and says "or others." The detection runs on C2PA content credentials, an industry provenance standard that writes a small record into the file describing how the image was made. Adobe embeds it. Once generative fill touches your photo, that record travels with the JPEG into Ads Manager.

A landscaping company in Brownsburg hit this in June. The owner photographed a finished patio job, then used generative fill to erase the neighbor's shed from the corner of the frame and stretch the lawn a little further left. A real project for a real customer, with one cosmetic cleanup. The credential went into the file anyway. The ad ran labeled.

The label measures one thing: whether a certain kind of software touched the pixels. It says nothing about whether the ad is honest.

What does the label cost you?

I don't have a clean number, and I'm not going to invent one. Nobody has published reliable before-and-after click-through data on labeled versus unlabeled ads. Anyone quoting you a precise percentage this month is guessing.

What we do have is the Gartner survey, and the numbers sitting next to that 50% are the ones I'd watch. Sixty-eight percent of the people surveyed said they frequently wonder whether the content in front of them is real. Sixty-one percent said they frequently question whether the information they use to make everyday decisions is reliable.

Those people are already suspicious. Now they have a reason.

Disclosing that a tool is powered by AI is not the same as promoting that a tool is powered by AI.

That's Kate Muhl, the Gartner analyst who ran the study. She meant it as advice about product design. It reads just as well as a warning about ad creative.

The exception nobody mentions

Meta's guidance says limited edits that don't significantly change an image or video may not get a label. Cropping, straightening, color correction, exposure. The ordinary work of making a photo look decent.

So this is not a rule against editing your photos. Adjust your levels. Fix the white balance on that fluorescent-lit interior shot. What triggers the label is generation, the software inventing pixels that were never in front of the lens.

One caveat on the tool list. Photoshop writes content credentials and Meta names it. Canva's position on the standard is less clear, and I have not seen Meta or Canva confirm what happens to a Canva AI image inside Ads Manager. If your creative comes out of Canva's AI features, treat the label as possible rather than settled, and go check.

What to do before your next ad run

Four things, in order of how much they'll save you.

1. Check one live ad the way a customer would. Open your own promoted post, tap the three dots, open "About this ad." Ten seconds. You'll know whether you have a problem.
2. Ask where your creative came from. If a freelancer or a nephew makes your ads, ask which files went through an AI tool. Most people won't volunteer it, because most people don't know the metadata follows the file.
3. Photograph the real thing. Your phone, your actual inventory, your actual staff, whatever light comes through the front window. For a shop selling a physical product or a finished job, this is faster than prompting anyway, and it's the one version that can't get labeled.
4. Keep generation off the hero image. If you're going to use AI somewhere, put it where nobody scrutinizes. The main image is the one that can push the label up next to "Sponsored."

A used-car lot off Crawfordsville Road in Speedway does the third one better than most agencies I've worked with. The owner walks the lot at 7am for the light, shoots forty photos on a four-year-old iPhone, and posts them unretouched with the gravel and the power lines still in frame. His ads look like a guy standing in a parking lot, because that's what they are. No label, and nothing to explain.

---

Meta built this detection because enough advertisers were passing off generated images as photographs that the platform needed a way to sort them. Your shop now sits inside a system designed for somebody else's problem.

The fix is the least sophisticated thing available. Go outside and take the picture.

Sources

#	Source	What it backed up
1	Meta adds updated disclosure tags for AI-generated ads	The July 7, 2026 update, the "About this ad" panel, and Meta's statement naming Photoshop and DALL-E.
2	About AI info on ads created or edited with third-party generative AI tools	Automated detection of third-party AI tools beginning June 1, 2026, and that no advertiser action is required.
3	Labeling AI content, Meta Transparency Center	The "AI info" label across Facebook and Instagram, and Meta's note that its labeling methodology is still evolving.
4	Gartner survey finds 50% of consumers prefer brands that avoid using GenAI	The 50% figure, the 1,539-person US sample, and the authenticity numbers.
5	Half of consumers prefer brands that don't use generative AI, Gartner finds	Kate Muhl's quote on disclosure versus promotion.