Google Ads AI Disclosure Panel: What Sellers Need to Set Up Now
On 2026-07-09 Google shipped a panel called How this ad was made inside My Ad Center. It covers Search, YouTube and Discover. Any user who opens the three-dot menu on an ad can now see whether the creative was generated or edited with AI, right alongside the advertiser name and the usual reasons-you-saw-this explanation.
For cross-border sellers the hard part is not deciding whether to declare. It is knowing what actually happened to the file. Creative gets outsourced, a freelancer runs forty variations through Midjourney, retouches the winner in Photoshop, and hands back a JPEG named hero_final_v3.jpg. When the media buyer ticks the disclosure box in the ad account, that buyer is guessing.
What the panel actually shows
The panel is an information card, not a badge stamped on the ad unit itself. Users have to open the three-dot menu to reach it. Inside, the AI provenance line sits next to advertiser identity and targeting rationale. If the image or video carries an AI generated or AI edited marker, the card says so in plain language.
Shopping ads and Performance Max assets are in scope too. That matters, because PMax already reworks uploaded images on its own. Those transformations happen on Google’s side of the fence and get handled by Google, so you do not control that part of the label.
Nobody knows yet what the panel does to click-through rate. Google has not published data, and there is no credible third-party test with a real control group. Anyone quoting a percentage drop is inventing it. My working assumption is that the effect is small in the near term simply because few users open that menu, but that is a hypothesis, not a measurement, and it should not go into a forecast deck.
Automatic labeling versus self-declaration
This structural split causes most of the operational pain, so it is worth being precise about which side any given asset falls on.
Assets made with Google’s own generative tools get labeled automatically. Image generation inside the Ads interface, Performance Max asset creation, the Product Studio family. Google knows it produced those pixels, the provenance travels with the asset, and the panel populates itself with no human involved.
Anything made with a third-party tool works the other way. Canva Magic Media, Adobe Firefly, Midjourney, Stable Diffusion, an in-house diffusion model. Google cannot detect any of it and relies on the advertiser declaring it at upload time. Google does not independently verify that declaration.
| Dimension | Google generative tools | Third-party AI tools |
|---|---|---|
| Labeling method | Automatic | Advertiser self-declaration |
| Manual step required | None | Yes, at asset upload |
| Independent verification | Internal pipeline, inherently trusted | None performed |
| Risk of a missed label | Very low | High, depends on your process |
| Where the burden sits | Platform side | Advertiser side |
The absence of verification reads like slack in the system. It is not. C2PA metadata embedded by the generating tool, visible watermarks, and competitor complaints all give a retrospective trail. The exposure is not really Google catching you. It is a European regulator asking a question you cannot answer.
Which edits count as AI, and which do not
The table below reflects the public guidance available today. Borderline rows are marked so, and for those you should confirm against the platform policy text for the markets you run in rather than treating this table as authority.
| Creative operation | Counts as AI generated or edited | Note |
|---|---|---|
| Full hero image generated from a text prompt | Yes | Entire frame produced by a model |
| Outpainting to extend a background | Yes | Adds imagery not present in the original |
| Synthetic model wearing the product | Yes | The person is model-generated |
| Swapping in a generated lifestyle backdrop | Yes | The new background is generated content |
| Cropping, resizing, changing canvas ratio | No | Conventional image handling |
| Color grading, exposure, sharpening | No | Conventional post-production |
| Adding a logo, copy layer, or price badge | No | Layout work |
| One-click AI background removal onto white | Confirm against platform policy | Uses a model but adds no new imagery |
| AI upscaling, denoising, restoration | Confirm against platform policy | Enhancement versus generation is unsettled |
| Ad copy written by a language model | Confirm against platform policy | The panel currently centers on visual assets |
| AI voiceover or synthetic presenter in video | Confirm against platform policy | Video guidance is still moving |
When in doubt, declare. Over-declaring carries no penalty that anyone has documented. Under-declaring leaves you explaining yourself. Background removal deserves special attention here, since nearly every mainstream tool quietly swapped in a model for that feature over the past two years. Teams that believe they never touch AI often already do.
Closing the metadata gap between creative and media
Every judgment above assumes the buyer knows how the file was made. When creative is outsourced, or when creative and media report into different departments, that assumption fails quietly. Three habits cover most of it, and none of them require new software or a budget request.
Start with file naming. Bake the provenance into the filename: sku1234_hero_aigen_mj.jpg for a Midjourney generation, aiedit_ps for something reworked with generative fill, photo for a straight studio shot. It is ugly. It also means the buyer never has to message the designer to find out.
Next, tag the asset library. Dropbox, Google Drive, or a proper DAM like Brandfolder all support custom fields. Add one called AI provenance with three values, human, AI generated, AI edited, plus a free-text field for the specific tool. Make it required for new uploads, then backfill in descending order of spend so live assets get covered first.
Finally, add a column to the delivery checklist. If your agreement with an outside studio lists filename, dimensions and format, add AI usage as a fourth column and put it in the contract annex. A written field beats asking over chat, and it gives you something to point at later. Studios grumble at first, then adapt, because every platform is heading the same direction.
One role that tends to get skipped is review. If assets pass through a pre-launch check, put the disclosure toggle on that checklist next to landing page URLs and UTM parameters. After a few cycles it stops being something anyone forgets.
The EU deadline has already passed
The real clock is not Google’s. Article 50 of the EU AI Act took effect on 2026-08-02, requiring that AI-generated or manipulated content be disclosed to users. Penalties reach 15 million euro or 3 percent of global annual turnover, whichever is higher. For most cross-border sellers that is not a fine, it is the end of the business.
So the sequencing is obvious when resources are tight. Audit EU-facing accounts first, every asset currently serving in Germany, France, Italy and Spain. North America and Southeast Asia can wait a cycle, though not indefinitely, since India and New York State are moving in a similar direction with their own timelines and wording.
Work through it by spend, not by asset count. Pull the top twenty ad groups by cost over the last thirty days, list the image and video assets in each, and verify them one at a time. That set usually accounts for the large majority of budget and is also the most visible to anyone inclined to file a complaint. The long tail can follow.
One last thing about posture. Plenty of teams are treating this as something to minimize, worried the label will hurt performance. The panel is user-facing. It does not enter the auction, and it does not touch Quality Score. What will hurt is a regulator asking which of your creatives were AI-made and nobody in the building being able to answer. If you do one thing this month, add the AI provenance field to your asset library and make it required on upload.
References:
- EU AI Act, Article 50 Transparency obligations, https://artificialintelligenceact.eu/article/50/
- Google My Ad Center, https://myadcenter.google.com/
阅读本文中文版: Google Ads 上线 AI 素材披露面板，跨境卖家该怎么接
Related Articles
TikTok TopReach: Who Should Actually Buy It
TopReach bundles TopView and TopFeed into a single buy. TikTok's own figures claim 59% incremental reach and 3x lower cost per reach, with no sample size disclosed. Here is the spend threshold that makes it defensible, and what to run instead when you fall short.
Meta Brand Memory Is Coming: Prep Your Ad Account Before the AI Learns Your Brand
Meta announced Brand Memory at Cannes Lions in June 2026: AI creative tools that learn your brand voice, visual identity, products, and past creative. It is still in limited testing, which makes right now the window to clean your creative library, consolidate your brand kit, and set guardrails. Includes a six-area prep checklist.
AI Ad Creative Tools Compared 2026: Meta Advantage+ Creative vs Google Asset Studio vs TikTok Symphony
Three platforms take different technical approaches to AI creative: Meta uses conversational signals to auto-generate variants, Google Asset Studio generates video and image assets with Veo/Imagen, and TikTok Symphony repurposes existing content through CapCut integration. Which one fits your workflow depends on your primary platform and where your creative production breaks down.