Google adds information about AI use in ads
My Ad Center will show whether an ad was created or edited with AI. The signal may come from Google's tools, an advertiser disclosure or a local requirement, and each route provides a different degree of certainty.
On July 9, 2026, Google announced a My Ad Center section that says whether an ad was created or edited with artificial intelligence. The official Google Ads post places the feature in the three-dot menu or information icon on ads across Search, YouTube and Discover, with global access. The change does not necessarily put a label over every image. In its general form, a user must open the panel and enter “How this ad was made.”
The announcement describes three routes for producing the information. When an advertiser uses Google's generative advertising tools, Google automatically adds the disclosure to the panel. When the creative is made elsewhere, the advertiser receives a control for declaring AI use. Where local requirements apply, a label may also appear directly on the ad, either automatically or after the advertiser uses that control.
This architecture corrects a common problem in headlines about transparency. “Google will label ads made with AI” sounds like a universal detector, but the source does not promise that Google will identify every synthetic asset made externally. It promises disclosure based on several signals. A label that is present provides information; an absent label does not prove by itself that no AI was used.
Disclosure, provenance and enforcement are different layers
A disclosure is the message a person sees. Provenance is technical or documentary information about a file's origin and transformations. Enforcement consists of rules, checks and consequences intended to make a declaration accurate. A system may be strong in one layer while remaining weak or silent in another.
Google says it already embeds imperceptible signals such as SynthID in output from its generative tools. That provenance marker can support automatic disclosure inside an ecosystem that recognizes it. For a creative made elsewhere, the company offers a manual declaration. The release does not describe a universal detector for those files or claim that every omission will be discovered. Adding that promise would go beyond the source.
The policy against misleading ads operates on a different axis. In the same announcement, Google says deceptive advertising remains prohibited whether AI was used or not. A synthetic image can honestly represent a concept, while a real photograph can accompany a false claim. The label answers how an asset was made or edited. It does not certify that a product exists, performs as advertised or comes with a legitimate offer.
Five questions for auditing a label
The first question is coverage: which ads, formats, surfaces, regions and types of editing are included. Google names Search, YouTube and Discover, but it also says the visible form depends on local requirements. Readers should not convert “globally accessible” into “identical in every country.”
The second is the trigger: which evidence activates the notice. It may be Google's internal record of its own tool, an advertiser selecting a control or a regional requirement. An interface should distinguish those origins because they do not provide the same degree of verification. When a panel shows one common phrase, users should not assign it more certainty than its mechanism supports.
The third is placement. A warning on the creative attracts attention during exposure; an explanation behind a menu serves the person who chooses to investigate. The official announcement confirms that “How this ad was made” lives in My Ad Center and that a direct label depends on local circumstances. Potential visibility and actual exposure are therefore different metrics.
The fourth is enforcement: what happens after an omission or false declaration. On the launch page, Google points to its general ban on deception and advertiser verification, but it does not publish a detection rate for external AI, an audit sample or a specific penalty for this control. The supported conclusion is not “Google checks nothing.” It is “this source does not allow external verification to be measured.”
The fifth is auditability: whether a user, researcher or advertiser can preserve the evidence. A strong disclosure identifies the creative, advertiser, basis for the label, date and correction history. The July announcement explains the user-facing view, not a complete public log of labeling decisions. That limit identifies the next information to request.
How to use the panel without confusing the signal
My Ad Center began as a control for the advertising experience. In its original 2022 introduction, Google said the menu could adjust brands and topics, limit sensitive categories and control personalization. The new section adds creative provenance to that context. It does not turn the panel into a forensic analysis.
When an AI disclosure appears, use it as a starting point. Check who paid for the ad, visit the transparency center for other creatives from the advertiser, and verify decisive features on the seller's page. For a physical product, look for independent photographs, dimensions, materials and return conditions. The useful question is not only whether an image is synthetic, but which claim it asks you to believe and which action it tries to prompt.
When no disclosure appears, do not conclude that the creative is a photograph or that generative editing was absent. The signal may not have been activated, the change may sit outside the stated scope, or the asset may come through another route. Apply the same controls to the content: advertiser identity, offer consistency, product evidence and purchase conditions.
A protocol for advertisers
A company producing ads can avoid relying on the memory of the person uploading a campaign. For every asset, record the originating tool, input files, edits, responsible owner, associated commercial claim and target jurisdiction. Then assign the required disclosure and retain a copy of the served ad or panel.
The inventory also needs to distinguish assistance from substantive generation. Cropping, adjusting color, composing a nonexistent scene or creating a synthetic person do not affect perception in the same way. Google says the section covers content created or edited with AI, but regional duties may define their own thresholds. The team should document the transformation before classifying it and escalate a legal uncertainty instead of inventing one global rule.
Before launch, another person compares the final asset with the record and with the advertising claim. Afterward, the team should open My Ad Center from the served creative to confirm that the information reaches the user. That end-to-end check catches failures that a selected control inside the campaign console cannot reveal.
Evidence still needed to judge the system
The announcement establishes scope, placement and input mechanisms, but it does not provide outcomes. Assessing effectiveness will require figures with denominators: the share of AI ads labeled through each route, omissions detected, corrections, time to remediation and the number of people opening the panel. Each metric must separate first-party tools, external declarations and local obligations; combining them would hide where the system works.
The transferable skill is to read any label across five axes: coverage, trigger, placement, enforcement and auditability. Applied to Google, they show a real improvement in information and its documentary limit. The feature opens a door for asking how an ad was made; it does not replace verification of what that ad claims.
That distinction prevents a useful signal from becoming a guarantee Google has not offered.
This article was produced with artificial intelligence under human editorial oversight.
