Google Ads AI Labeling in 2026: What the Policy Actually Says

Quick answer: Google does not require you to label an AI-generated ecommerce ad. Its policy note of 9 July 2026 says Google permits labels inside image and video creatives and is rolling out an optional AI label setting across five products. The only Google Ads disclosure that is actually mandatory covers election ads, where a checkbox in campaign settings has been required for some time. What can create a real obligation is law, and Google names three sources: the European Union, India and New York. Google's own page then disclaims its tool: "Use of the AI label setting in Google's advertising products doesn't guarantee compliance with specific regulations." Read that sentence twice before you take a vendor blog's word for what you now have to do.

What the July 2026 update actually says

The document is short. Titled "Updates to AI labeling requirements (July 2026)" and posted on 9 July 2026, it opens with a permission rather than a duty: "Starting in July of 2026, Google permits adding text or visual labels directly within image and video ad creatives that were generated or modified using AI to help advertisers comply with emerging AI transparency regulations."

Three things follow from that wording, and all three are operationally useful.

First, the label is yours to add or not add. You can burn it into the creative, or you can use Google's AI label setting, which the note says launches gradually through July in Google Ads, Display & Video 360, Campaign Manager 360, Merchant Center and Ads Editor.

Second, there is a quiet carve-out that saves you a rejection, and the policy page grants it to advertisers directly rather than only to Google's own tools. The note says "Advertisers can add these labels directly to their creatives or use the AI label setting", and then, of those labels: "These labels will not be in violation of Google policies prohibiting text overlays and watermarks." Merchant Center repeats it for labels you apply "using your own design tools". If you have ever had an image asset knocked back for overlay text, that matters: a burned-in "Made with AI" strip is not supposed to trip that policy.

Third, the note explicitly preserves the election rule. Election advertisers must continue disclosing synthetic or digitally altered content through the "Altered or synthetic content" section in campaign settings. The new AI label does not replace it.

Where the disclosure is actually visible

This is the part most coverage skips, and it changes how much the label costs you in practice. The Google Ads and Display & Video 360 help pages describe two different surfaces:

So outside those three markets, labelling an asset does not paint a badge across your creative. It files the disclosure into the ad's own information panel, a surface a viewer has to go looking for. Inside them, it paints an overlay on the ad. If you were hesitating because you pictured a permanent watermark on every impression, that fear is only correct for a subset of your targeting.

The labels Google applies for you, and cannot un-apply

Both the Merchant Center and Display & Video 360 pages carry the same sentence: Google "may label your assets in some cases, such as in regions where Google itself is legally required to do so, or when we get signals to do so from other platforms." Assets generated through Google's fully automated creative features may also be labelled on your behalf. In all of those cases the help pages state that the labels cannot be overwritten.

"Signals from other platforms" is the clause worth planning around. It means provenance data travelling with your file can decide the label for you, the same mechanism that produces automatic labels on TikTok and Meta. Google does not publish which signals it reads or how it weighs them, so you cannot predict the outcome from your side. Decide what you want disclosed and set it deliberately, rather than discovering what Google decided after the campaign is live.

Google never defines AI-generated, and that is the real gap

The advertising policy does not draw a line between generating an ad with AI and editing one with AI. It offers no test, no examples and no threshold. That is conspicuous, because Google itself publishes a workable test two doors down and so do its competitors.

The third row is not a footnote if you buy Shorts or Demand Gen. Those ads run off an uploaded YouTube video, so the upload rulebook and the ads rulebook both attach to the same file, with different triggers and different consequences. AI UGC ads on YouTube Shorts and Demand Gen takes that pair apart alongside the format specs.

Borrow Google's own election language when you need a working rule, because it is the only line Google has drawn. That policy exempts "image resizing, cropping, color or brightening corrections, defect correction (for example, 'red eye' removal), or background edits that do not create realistic depictions of actual events". Everything past that, on a synthetic presenter ad, is generation.

The other thing missing from the AI labeling note is a penalty. There is no enforcement paragraph, because there is no general duty to enforce. The Google Ads policy that still applies to a deceptive AI ad is Misrepresentation, whose Manipulated media section reads, in full: "Manipulating media to deceive, defraud, or mislead others is not allowed." That page does not mention AI, deepfakes or synthetic content anywhere. Its harshest enforcement is reserved for unacceptable business practices and coordinated deceptive practices, where accounts can be suspended immediately without warning.

The three jurisdictions Google names do not bind you equally

Google's one-line justification is that "AI regulations in the European Union, India, and New York require that ads with certain AI-generated or edited assets include disclosures and/or labels." True, but those three laws point at three different parties, and only one of them points at you.

India is why Google auto-labels rather than asking you to. The obligation there sits on the platform, which is exactly the situation Google's "regions where Google itself is legally required to do so" sentence describes.

The EU and New York are the interesting pair, and the popular claim that their triggers are inverted does not survive the primary texts. Article 50(4) bites on a deep fake, and paragraph 113 of the Commission's Article 50 Guidelines sets out fourcumulative criteria, not three: "(i) resemblance of (ii) existing (iii) persons, objects, places, entities or events (iv) that would falsely appear to a person to be authentic or truthful". Criterion (ii) is met where the simulated subject resembles "someone or something that exists, can plausibly exist or could have plausibly existed in reality", so plausibility is enough and actual existence is not required. Criterion (iii) is the one that decides synthetic UGC: the Guidelines define "persons" as realistic human beings "including digital replicas of real persons, realistic AI-generated human avatars or personas".

An invented, photorealistic presenter is therefore inside the EU deep fake definition rather than at the weak end of it, and paragraph 114 lists "fully AI-generated actors" among the uses of AI likely to make content falsely appear authentic. New York's statute defines a synthetic performer as an asset intended to create the impression of a performance by "a human performer who is not recognizable as any identifiable natural performer", and requires the advertiser to "conspicuously disclose" it. That makes New York the narrower rule: it catches the invented presenter and stops at the recognisable real person, while the EU catches both. Two live regimes that overlap on the exact creative you are most likely to run, and the AI label setting is one checkbox that Google says guarantees compliance with neither.

Note also that Article 50(2), the machine-readable marking duty, falls on the model provider rather than on you, and providers whose systems were on the market before 2 August 2026 have until 2 December 2026 to comply.

What this means if you generate AI UGC

One detail from our own model registry is worth stating, because it decides which statute you are reasoning about. Every video model we run requires a starting image except the two Sora models, so in almost every render you supply the first frame. Sora is the exception, and OpenAI retires both Sora 2 and Sora 2 Pro on 24 September 2026, which makes it a bad foundation for a workflow anyway.

The practical consequence: for nearly every ad you make, you already know whether the person on screen is an invented AI actor or a photograph of a real human, because you chose the input. That is the exact fact New York's synthetic performer test and the EU's deep fake test both turn on, and it is a question about your source image, not about the model. Our per-model cost benchmarks and the lip-sync model comparison both start from the same registry.

A checklist that survives all three regimes

• Set the AI label deliberately on every image and video asset in Asset Studio, rather than leaving it undesignated and letting an automatic label decide for you.
• If you target the EU, India or New York, expect a visible overlay on the ad. Design the creative so an overlay does not cover your hook text or product.
• Never generate an ad presenter who resembles a real, identifiable person without their permission. That is the one case both the EU deep fake test and Google's Misrepresentation policy hit at once.
• Do not treat an invented presenter as the safe option for EU targeting. The Commission's Guidelines put realistic AI-generated personas inside the deep fake definition, so Article 50(4) wants a human-readable disclosure on that ad, and Google's label setting is not a substitute for one.
• Add your own visible disclosure inside the creative when the same file will also run on TikTok or Meta, whose rules are stricter and differently worded. Google says a self-added label will not breach its text overlay policies.
• Burn any visible label in before the final export. We tested this on our own pipeline and a C2PA provenance manifest did not survive even a lossless remux, so a burned-in label is the disclosure most likely to survive a trip through an editor.
• Keep the election checkbox in mind if you ever touch political or issue advertising. It is a separate setting and it is mandatory.
• Record which ads used AI and how. Both statutes above turn on knowledge, and a log is the cheapest evidence you will ever produce.

None of this is legal advice, and Google says as much about its own tool: "Seek legal guidance and take measures as needed to ensure your ads and assets follow all local legal obligations and policies." What you can take from the policy page itself is narrower and more useful than the headlines suggest. Google built you a labelling mechanism. It did not hand you a duty.

Sources

• Google Advertising Policies Help, "Updates to AI labeling requirements (July 2026)" (posted 9 July 2026)
• Google Ads Help, "Use AI content label settings and disclosures"
• Display & Video 360 Help, "Use AI content label settings and disclosures"
• Google Merchant Center Help, "Use AI content label settings and disclosures"
• Google Advertising Policies Help, Misrepresentation (Manipulated media)
• Google Advertising Policies Help, Political content ("Altered or synthetic content")
• YouTube Help, "Disclosing use of altered or synthetic content"
• EU AI Act, Article 50, and the European Commission's FAQ on transparency obligations under Article 50
• European Commission, Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50, C(2026) 5054 final, 20 July 2026, paragraphs 113 and 114 (read as the published PDF)
• New York State Senate, S8420A (General Business Law 396-b, synthetic performers)
• Freshfields, "India targets deepfakes and AI-generated content" (law-firm analysis of the MeitY 2026 amendments, used here for the Indian dates and duty-holder because the gazette notification is the primary text)

All eleven were read directly in August 2026. Policy pages change without notice; the dates above are how you check whether this one has gone stale.

The same creative rarely faces the same rules twice. TikTok's AI disclosure rules put the labelling burden squarely on you and reject undisclosed ads, while Meta labels detected AI itself. If you are more worried about the ad looking synthetic than about declaring that it is, what makes AI UGC look fake is the more useful read. UGC Vids AI generates the ads themselves, from $49/month, free for 3 days.