What changed
On July 9, 2026, Google announced that it is adding a "How this ad was made" panel to My Ad Center, the settings hub reachable via the three-dot menu or info icon on ads across Search, YouTube, and Discover. The panel tells a viewer whether generative AI was used to create or edit the ad they're looking at. Until now, that obligation applied only to election ads, a policy Google introduced in 2023, per Search Engine Land. Every commercial ad on Google's largest surfaces is now in scope.
The mechanism has two tiers. When an advertiser builds an ad with Google's own generative tools, disclosure is automatic — Google embeds SynthID and other machine-readable signals and populates the panel itself, per Search Engine Land. When an advertiser brings in creative made elsewhere — Midjourney, Sora, an in-house model — the advertiser flips a new self-declaration control. Google has confirmed it will not independently verify that declaration. It names the EU, India, and New York as markets where local law already forces a visible on-ad label regardless of the panel; everywhere else, the honor system governs.
The case for a voluntary label
There's a real problem here worth taking seriously before dismissing the fix. Generative AI has made it trivial to fabricate product photography, celebrity-style endorsements, and lifestyle scenes that never happened, at a cost approaching zero. Google's own policies already ban misleading ads, but "misleading" is a high bar — a synthetic product shot that's merely flattering, not false, can slip through untouched. Consumers reasonably want to know whether the shoes in an ad photo exist in the real world or were generated overnight. A disclosure panel that costs advertisers nothing to check and costs Google little to build is a low-friction way to surface that distinction to anyone who bothers to click the info icon, without banning AI-made ads or subjecting every campaign to prior review.
Why self-declaration without audit falls short
The weakness is structural, not incidental. A disclosure regime built entirely on advertiser self-interest works only for advertisers who were already inclined to be straightforward. The advertiser most likely to use AI-generated content deceptively — a fake testimonial, a doctored "before and after" — is also the advertiser least likely to tick the box admitting it. Google's own SynthID watermarking proves the company has the technical means to detect AI provenance when it wants to; declining to apply that capability to third-party uploads makes the disclosure symbolic for exactly the ads where it matters most.
Regulators are not waiting to find out whether the honor system holds. Article 50 of the EU AI Act becomes directly applicable on August 2, 2026 — barely three weeks after Google's announcement — and requires providers of generative AI systems to mark synthetic audio, image, video, and text output in a machine-readable format, with deployers obligated to flag deepfakes and AI-manipulated content to viewers. Non-compliance carries fines up to €15 million or 3% of global annual turnover, whichever is higher, confirmed by TrueScreen's analysis of the regulation. That is not a self-declaration regime — it is a marking mandate backed by penalties sized to bite even a company Google's size, and it lands on Google's own EU ad business within a month of this rollout.
The proportionate middle ground
Google's timing is not a coincidence; a voluntary, globally-deployed disclosure feature announced weeks before a binding EU deadline reads as pre-positioning to show good-faith compliance infrastructure before regulators start asking questions. That instinct — building the tooling before the law forces it — is the right one, and it's a better outcome than either extreme: an outright ban on AI-assisted advertising creative would sacrifice real efficiency gains (Google's own framing of cheaper product photography and localization is genuine), while a fully permissive market with no disclosure at all would leave consumers unable to distinguish synthetic claims from real ones.
But a proportionate policy shouldn't stop at "advertisers may tell the truth if they choose to." The more defensible middle path is what the EU is already forcing and Google has only partially built: machine-readable provenance (C2PA content credentials and SynthID-style watermarking, which Google's help documentation already references) applied consistently regardless of which tool produced the ad, with platform-level spot-checks rather than blanket government pre-clearance. That preserves the low-friction, innovation-friendly posture regulators should favor, while closing the exact gap — no verification of third-party claims — that Google itself has now publicly admitted exists. Voluntary disclosure is a reasonable first move. It should not be mistaken for the finished policy.