On July 9, 2026 Google added a "How this ad was made" section to the My Ad Center panel, the info surface you reach by tapping the three-dot menu on an ad across Search, YouTube, and Discover. It tells you whether the ad was created or edited with generative AI. The mechanism splits cleanly in two, and the split is the whole story. When an advertiser uses Google's own generative ad tools, Google adds the disclosure automatically and embeds an imperceptible SynthID watermark in the output — a signal it can detect later. When an advertiser builds the creative with a third-party tool, Google gives them a control to declare AI use and, in its own words, does not run a check to verify the claim. So one half is provenance Google can prove; the other half is an honor system. That distinction matters because the label is not arriving in a vacuum: it lands three weeks before the EU AI Act's Article 50 transparency obligations become enforceable on August 2, 2026, and on the same industry rails — C2PA Content Credentials and SynthID watermarking — that OpenAI, Google, and a growing list of platforms converged on in May 2026. This guide explains exactly what Google shipped, why the self-report half is weaker than it looks, the provenance layer underneath that is doing the real work, the regulation driving all of it, and what a creator or advertiser should actually change now that "made with AI" is becoming a default disclosure rather than an exception.
On July 9, 2026 Google added a new line to the panel that already tells you why you are seeing an ad and who paid for it. The My Ad Center panel — reachable by tapping the three-dot menu or info icon on an ad across Search, YouTube, and Discover — now includes a "How this ad was made" section that says whether the ad was created or edited with generative AI (Google blog, July 9, 2026). On its face it is a small UI change. Underneath, it is Google formalizing a position on a question every platform is now being forced to answer: when the content you see was made by a machine, are you told?
The important detail is not the panel; it is that the disclosure works two completely different ways depending on how the ad was made, and only one of those ways is backed by proof. Ads built with Google's own generative ad tools get the label added automatically, and Google embeds an imperceptible SynthID watermark in the output so it has a signal it can detect later. Ads built with a third-party tool get a control the advertiser can use to declare AI use — and Google has said plainly that it will not run its own check to confirm that declaration (TechCrunch, July 9, 2026). One half is provenance; the other half is an honor system. This guide is the strategy read on that split: what Google actually shipped, why the self-report half is weaker than it looks, the C2PA-and-SynthID provenance layer doing the real work, the EU regulation driving the timing, and what a creator or advertiser should change now that "made with AI" is on its way to being a default disclosure. It is the Google-surface companion to the Meta-focused AI-generated ads disclosure and UGC-style creatives and to the news write-up at Google adds "How this ad was made" labels.
Strip it to the facts. The disclosure is a section inside My Ad Center, Google's existing per-ad transparency surface, and it appears globally across the three properties where most Google advertising lives: Search, YouTube, and Discover. It reports a single thing — whether generative AI was used to create or edit the ad's content — and you find it the same way you already find "Why this ad" and the advertiser identity, through the three-dot menu or info icon on the ad. Google framed the intent as helping people understand the ads they see while giving advertisers straightforward tools to navigate evolving industry standards. In some markets the ad may also be labeled as AI directly on the creative where local law requires it, but the baseline lives in the panel, not always on the ad face.
The two-path mechanism is the substance. Path one: an advertiser uses Google's generative ad tools — the AI features inside Google Ads that build headlines, images, or video. In that case Google knows the content is AI-made because it made it, adds the "How this ad was made" disclosure automatically, and marks the output with SynthID, its invisible watermark. Path two: an advertiser produces the creative somewhere else — any outside image or video generator — and uploads it. Google cannot see how that file was made, so it gives the advertiser a control to indicate that generative AI was involved, and it does not verify the answer. The honest reading is that Google built a strong disclosure for the content flowing through its own tools and a voluntary one for everything else, because the everything-else case is the one it has no reliable signal for.
A disclosure system is only as good as its worst path, and the third-party path is the worst path by design. It asks the advertiser — the party with the least incentive to volunteer that their ad is synthetic — to flag it, and then declines to check. That is not a criticism of Google so much as a description of the limit: Google can prove provenance for content it generates and cannot prove it for content it did not, so for outside tools the choice is between a voluntary label and no label at all, and a voluntary label is at least a hook the policy and the law can attach to. But it means the panel's "no AI indicated" state carries almost no information for third-party creative. Absence of a label is not evidence of a human hand; it may just be an advertiser who left the control untouched.
This is exactly why the interesting action is not the checkbox but the watermark. Google's automatic path leans on SynthID precisely because a signal embedded in the pixels survives the advertiser's intentions — it does not depend on anyone choosing to be honest. The trajectory is clear: the disclosure that matters long-term is the one that can be detected, not declared, and detection is improving fast. Building an ad strategy around the gap in the self-report path — quietly running third-party AI creative and never touching the control — is betting against that trajectory. It is the same losing bet as evading Meta's detection, which now runs automated identification of third-party AI media, covered in the Meta disclosure guide. The gap is real today and closing on a schedule.
To see where this is going, separate two things people blur together: a label and provenance. A label is a claim displayed to a viewer — "this ad was made with AI." Provenance is verifiable evidence of how a piece of content was actually produced. Google's automatic labels are backed by provenance; its third-party control is a label without it. And the provenance layer is not a Google-only project — it is an industry standard the major players converged on in mid-2026. C2PA Content Credentials are a cryptographically signed metadata manifest, ratified as an ISO standard, that records which device or model produced a file and every edit applied since, so anyone can check the chain. SynthID is Google DeepMind's complementary approach: an imperceptible watermark woven into the pixels of an image, the waveform of audio, or the token patterns of text, with over 100 billion pieces of media watermarked since 2023.
The consolidation is what makes this durable rather than a one-off feature. In May 2026 OpenAI and Google aligned on a dual-layer model — C2PA metadata for the manifest, SynthID-style watermarking for the embedded signal — and companies including ElevenLabs and Nvidia announced adoption in the same window; Google said C2PA verification and SynthID detection are coming to Search and Chrome (C2PA Viewer, 2026). Google's ad disclosure is the advertising-facing expression of that same stack. The "How this ad was made" panel is the human-readable tip; SynthID and C2PA are the machine-readable foundation the regulators are actually asking for. If you only watch the label, you miss that the ground under AI content is shifting toward always-on, checkable provenance — the practical consequence of which is that "was this made with AI" is becoming a question with a technical answer, not just a self-reported one.
The timing is not a coincidence, and the driver is European. The EU AI Act's Article 50 sets transparency obligations for AI systems: generative outputs must be marked in a machine-readable format and detectable as artificially generated, people interacting with an AI must be told, and deepfakes and AI-generated content must be labeled. Those obligations become enforceable on August 2, 2026 — three weeks after Google's announcement — and non-compliance can draw penalties reaching €15 million or 3% of global annual turnover (EU AI Act, Article 50). The Commission's June 2026 Code of Practice on transparency of AI-generated content points squarely at C2PA Content Credentials as the machine-verifiable layer, which is why the whole industry is standardizing on the same rails at once.
Google is not new to this pressure. Its 2023 rule requiring election advertisers to disclose synthetic or digitally altered content was the first version of the same idea, and the July 2026 expansion generalizes it from political ads to advertising broadly. Read the sequence and the direction is unmistakable: disclosure started as a narrow, high-stakes carve-out and is becoming a default expectation across categories and platforms. A global advertiser cannot easily run one labeling regime for the EU and none elsewhere, so the compliant behavior tends to become the universal behavior. The safe planning assumption for any creator or brand is not "will I have to disclose AI use" but "how do I disclose it without the label costing me anything." The wider regulatory backdrop for synthetic personas specifically is in regulating humanlike AI, and the trust dimension in the AI influencer manipulation trend.
Google's panel is one node in a fast-forming pattern. Meta applies an "AI info" label to detected or Meta-made AI ad media and, since June 1, 2026, runs automated detection of third-party AI creative. YouTube requires creators to disclose realistic altered or synthetic content, TikTok auto-labels AI content it detects and reads C2PA credentials on upload, and the platforms are wiring the same C2PA and SynthID signals into their pipelines. The through-line is that disclosure is migrating from a thing a conscientious creator chooses to do into infrastructure the platforms and the law enforce by default. When four of the largest distribution surfaces all move the same direction in the same year, that is not a trend to monitor — it is the new baseline to design around.
The useful reframe is that this is less about restriction and more about a repricing. When AI content was undisclosed, the value of a piece of content and the fact that a machine made it were separate; a disclosed feed collapses them, so the label becomes part of how the content is received. That changes the calculus in a specific way: cheap, generic AI creative gets no worse legally for being labeled, but it gets more obvious, and an audience that already pattern-matches generated content will pattern-match it faster with a marker attached. Meanwhile genuinely good work carries the label without penalty, because the disclosure is not the problem — thin, voiceless output is. This is the same generic-versus-original line platforms are drawing on organic content, described in AI content authenticity strategy, now extended to paid. Disclosure does not devalue AI content; it devalues lazy AI content specifically.
Three practical moves follow, and none of them is "stop using AI." First, know your own provenance. The single thing a disclosure regime demands is that you can answer, per asset, how it was made — which elements are generative, which are captured or human-authored, and what tool produced each. Advertisers who cannot answer that will either over-declare, under-declare, or guess, and all three are avoidable if you track it from the point of generation. Second, declare where asked and stop treating the control as optional. Detection is improving on a published schedule; the window where quietly-uploaded third-party AI creative goes unlabeled is closing, and being caught hiding it costs more reputationally than the small marker ever costs in performance.
Third, make disclosed AI content good enough that the label is a footnote, not a confession. This is the part most within your control and the part that actually decides outcomes. A labeled ad that carries a real voice, a specific claim, and a recognizable identity reads as a brand using a tool; a labeled ad that reads like the model's default output reads as slop with a warning sticker. The difference is the human layer — a point of view the generation started from and a review step before it ships. The step-by-step of setting the disclosure controls themselves, on the platforms that have them, is in how to disclose AI-generated ads; the broader move of ad creation into the ad platforms is in AI ad generation moves inside the ad platforms.
A disclosure regime rewards two things a creator rarely has together: knowing precisely how every asset was made, and being able to produce disclosed AI content that is still good enough that the label costs nothing. Kompozy helps with both, and its angle here is specifically provenance and consistency rather than volume. Because Kompozy is the engine that generates the content — it does not hand you someone else's file to upload blind — every output has a known origin: which model drafted the copy, which produced the image, which rendered the video. When a platform's "made with AI" control asks the question, you are not guessing; you have a truthful, consistent answer to give across all nine social platforms and the blog and email destinations it publishes to, instead of a different guess per surface. In a world where the honor-system half of Google's system depends on advertisers actually knowing what they made, being the one who knows is the whole advantage.
The second half is that Kompozy is built to keep disclosed AI content on the original side of the generic-versus-original line, so the label is a footnote. A Persona Brief encodes how you actually sound — your positions, your examples, your do-not-say list — so drafts start in your voice rather than the model default, and AI-tell filters strip the template cadence before anything ships. Persona video runs on an avatar built from your own declared identity rather than an anonymous synthetic stranger, and HyperFrames renders pixel-exact, on-brand carousels and graphics instead of the interchangeable look labeled slop wears. Crucially, the publishing flows run through a per-post review pipeline and quality gates, not fire-and-forget automation — a person approves each piece, so labeled AI creative goes out looking like your brand and reading like your voice. That is the compliant posture the transparency shift is heading toward: not hiding that AI was involved, but disclosing it on content good enough that disclosure is a non-event.
Google added a "How this ad was made" section to My Ad Center on July 9, 2026, telling viewers across Search, YouTube, and Discover whether an ad was made with generative AI. The system has two halves — automatic, SynthID-backed disclosure for ads built with Google's own tools, and a voluntary, unverified control for third-party creative — and the split reveals where this is all heading: toward provenance you can detect rather than labels you have to trust. It arrives on the eve of the EU AI Act's Article 50 becoming enforceable and on the same C2PA-and-SynthID rails the industry standardized on in mid-2026, which is why Meta, YouTube, and TikTok are all moving the same direction. For creators the takeaway is not to fear the label. It is to know how your content is made so you can disclose truthfully, declare where asked because detection is closing the gap, and make your disclosed AI content distinctive enough that "made with AI" is a footnote rather than a confession.
It is a section Google added to the My Ad Center panel on July 9, 2026 that tells you whether an ad was created or edited with generative AI. You reach it by tapping the three-dot menu or info icon on an ad across Google Search, YouTube, and Discover. When an ad was built with Google's own generative ad tools, the disclosure is added automatically; when it was made with a third-party tool, the advertiser has a control to declare AI use themselves.
Only for ads made with its own tools. When advertisers use Google's generative ad features, Google adds the label automatically and embeds an imperceptible SynthID watermark in the output, so it has a provenance signal it can detect. For ads built with third-party tools, Google provides a control for the advertiser to declare AI use but says it will not run its own check to verify that claim — so the third-party half of the system relies on advertiser honesty.
Across Search, YouTube, and Discover, inside the My Ad Center panel reached from an ad's three-dot menu or info icon — the same place Google already surfaces why you are seeing an ad and who paid for it. In some markets the ad may also be labeled as AI directly on the creative if local law requires it, but the baseline disclosure lives in the My Ad Center panel rather than always on the ad face.
The timing tracks the regulation. The EU AI Act's Article 50 transparency obligations — which require AI-generated and manipulated content, including deepfakes, to be labeled and machine-readable — become enforceable on August 2, 2026, with non-compliance penalties that can reach €15 million or 3% of global annual turnover. Google's move also builds on its 2023 requirement to disclose synthetic or altered content in election ads and lands on the same C2PA and SynthID rails the wider industry converged on in mid-2026.
A label is a claim shown to viewers; provenance is verifiable evidence of how content was made. C2PA Content Credentials are a signed metadata manifest — now an ISO standard — that records which tool or model produced a file and every edit since, and SynthID is Google's invisible watermark embedded in the pixels, audio, or text of its AI outputs. A self-reported label can be wrong or absent; a provenance signal can be checked. Google's automatic labels rest on SynthID; its third-party control is a label without that backing.
Assume disclosure is becoming the default and get ahead of it. Know exactly how each asset was made — which parts are generative and which are not — so you can answer a "made with AI" control truthfully rather than guessing; declare AI use where a platform or law asks, since detection is improving and evasion is a shrinking bet; and make disclosed AI content good enough that the label is not a liability. Keep a real voice and a human review step so labeled AI creative still reads as yours, not as generic output.
On July 9, 2026 Google added a "How this ad was made" section to the My Ad Center panel across Search, YouTube, and Discover, telling viewers whether an ad was created or edited with generative AI. Ads built with Google's own AI tools get the label automatically and carry an imperceptible SynthID watermark; for third-party tools, advertisers self-declare AI use and Google does not verify the claim. It is part of a wider transparency shift — the EU AI Act's Article 50 (enforceable August 2, 2026), C2PA Content Credentials, and SynthID watermarking — pushing "made with AI" toward a default disclosure rather than an exception.
Get started → · ← All guides · Compare Kompozy vs other tools