Which ad platform makes you declare AI on an ordinary commercial ad?

01Do you have to tell Meta your ad was made with AI?

Not for an ordinary commercial ad. Meta's help page puts the duty in one lane: "For ads about social issues, elections or politics, advertisers are already required to disclose if the image, video or audio are digitally created or altered, including through the use of third-party AI tools." A skincare brand running a generated hero shot has no Meta disclosure to make.

I am reading these policy pages as the person who uploads the files, not as a lawyer, and all of them are linked below. Most write-ups mix two systems. Meta's AI Info label is applied by Meta, on its own detection, and it goes on detected AI content without any advertiser action from mid-2026, on a date Meta's page gives as June 1, 2026, read as paraphrase rather than verbatim. Meta's disclosure duty is a self-declaration in the ad creation flow, and exists only in the political lane.

One correction to carry out of that table. There is no Meta advertising standard called Unrealistic Outcomes; the URL returns 404 and the Wayback Machine shows it doing so as recently as February 2026. What rejects an overstated generated before-and-after is Unacceptable Business Practices, which prohibits ads that "Use deceptive or exaggerated claims about the success of a product or service" and the same about health-related benefits. The claim gets you rejected. What a viewer is told about a person on camera is the FTC's question rather than the platform's.

You must not misrepresent Output as human generated when it is not.

Meta, Ad Creative Generative AI Terms

02TikTok is the one that asks

Its advertising policy, in the section on edited media and AI-generated content, says significantly edited media and AIGC is allowed if you "Apply the AIGC label , or by adding a clear disclaimer, caption, watermark, or sticker of your own." The consequence sits in the same section: "If we identify AI-generated content that has not been disclosed, your ad will be rejected or restricted." It binds every advertiser, and TikTok tells the uncertain ones to label anyway.

Three things trigger it: fully AI-generated image, video or audio; the primary subject doing something they did not do; and voice cloning putting words in somebody's mouth. TikTok also bars creative using a public figure's likeness without permission. One caveat, stated plainly because I would want it stated for me: TikTok's domain does not answer from this machine, so every TikTok line here comes from an archive capture.

03Which edits count as significant?

Every platform draws a line between touching the file and changing what it depicts, and all four draw it in nearly the same place. Resizing, cropping, color correction and repair sit on the safe side everywhere. A whole generated image, or a real person made to say something they did not, sits on the other side everywhere. Two edits fall in the gap. Those are the ones worth knowing.

Layer four is the one that catches working teams out. Generating a background is routine, it happens on most of what we deliver, and two of these four platforms classify it in opposite directions. Layer five starts a legal duty rather than a platform one, and the state law behind it has a $1,000 first-violation penalty that has nothing to do with whether an ad gets approved.

Type your disclosure line and read it at feed size

Type in it

The line you want on screen

How long it holds

Type size

Plate behind the type

Words-

Dwell floor-

Your hold-

Contrast-

SaaS - studio demo. The string is one of Google's own published examples for election ads. Try your own against a dark frame. The legal standards say difficult to miss and unavoidable; the readability floors underneath that are craft, and the sister journal measures them.

04What does Google Ads require of a US advertiser?

Nothing on an ordinary ad, and two specific things elsewhere. Verified election advertisers must disclose synthetic content that inauthentically depicts real or realistic-looking people or events, wherever verification is required. The United States is one. Mechanically it is a checkbox in the campaign's altered or synthetic content setting. Separately, Google's help page tells advertisers to add labels to creatives, or use the AI label setting, for EU, Indian and New York requirements.

That second sentence is the one nobody expects: a platform help page naming a US state law before most agencies have. The same page carries the line to put in front of a client who thinks generation changes anything, that all advertising content is subject to Ads Policy regardless of how it was made. Google writes the election disclosure for you in mobile feeds, Shorts on mobile and in-stream video. Everywhere else, you build it.

META    ads about social issues, elections or politics only
        > ad creation flow > the disclosure question
        Meta then surfaces it on the ad and in the Ad Library.
        Everything else: no advertiser switch exists. The AI Info
        label runs on Meta's own detection from mid-2026, on a
        date Meta's page gives as June 1, 2026, as paraphrase.

TIKTOK  every advertiser, political or not
        > apply the AIGC label, OR add your own clear disclaimer,
          caption, watermark or sticker
        Symphony avatar exports already carry an AI-generated label.

GOOGLE  verified election advertisers
        > campaign settings > "Altered or synthetic content"
        Everyone else, where EU, Indian or New York requirements
        apply: the AI label setting, or labels on the creatives.

YOUTUBE uploads to a channel, not bought inventory
        > video details > Attributes > "AI use"
        Photorealistic content may also get a label in the player.

05YouTube's rule is written for creators, and nobody says whether it reaches ads

This is the cleanest example of the confusion the whole subject runs on. The YouTube altered-content disclosure lives in Studio, under a video's Attributes, written throughout as an instruction to somebody uploading to their own channel. The penalty language is about creators and the Partner Program. Nothing on the page addresses inventory bought through Google Ads, and no page I could find extends it there.

So the answer to a client asking whether a YouTube ad needs the AI attribute is that the two systems are separate and nobody has said they connect. Less satisfying than a rule, more useful than a guess. If the brand also posts the same cut on its own channel, the Studio attribute applies to that upload on its own terms, and the craft reasons a generated cut reads as generated matter more than the attribute does.

06What to settle before you upload

Whoever uploads the file needs one thing from whoever built it: a note saying which edits were generative, shot by shot. On our deliveries that note travels with the cut, because the person clicking upload is usually not the person who chose the background. Everything below takes ten minutes once that note exists, and an afternoon of guesswork when it does not. Where the file also has to clear a hard format spec, like Amazon Sponsored Brands video, run both lists in one pass. If the buy reaches the EU or the UK, that is a different rulebook again, and the sister journal covers it at its own sources.

The platform question has a small answer and a large one. The small one is a toggle on TikTok and a checkbox on Google. The large one is that the file is starting to answer for you: California makes the provenance get embedded at the vendor, and from January 2027 the biggest platforms have to surface it rather than strip it. Whatever you declare at upload is going to be read against the metadata, not against your memory of the brief.