Google Is Labelling AI-Generated Ads: What Marketers Need to Change
Jack Amin
Digital Marketing & AI Specialist
Quick Answer
Google is adding transparency signals for ads created or materially edited with its generative AI tools. Marketers should maintain an asset register showing source files, prompts or edit instructions, human approver, claims evidence and usage rights. Review every generated asset for misleading product depiction, bias, prohibited content and brand accuracy. A label does not make an ad compliant, and an unlabelled ad is not automatically human-made.
Transparency Is Becoming Part of the Ad Supply Chain
Google has announced new transparency measures for advertising assets created or materially edited with its generative AI tools. The exact presentation can vary across products and placements, but the direction is clear: AI provenance is moving closer to the media delivery system.
For marketers, this is not a reason to abandon AI-assisted creative. It is a reason to professionalise how assets are generated, reviewed and approved.
The core question changes from “Can we make this image?” to “Can we prove what this image represents, where it came from and who approved it?”
A Label Is Not a Compliance Check
A transparency signal can tell a viewer or platform that AI contributed to an asset. It cannot verify that:
- the product is represented accurately
- a performance claim is substantiated
- a person consented to use of their likeness
- the image is suitable for the audience
- the landing page matches the creative
- the advertiser owns the necessary rights
Those remain the advertiser's responsibilities. Australian Consumer Law applies to the overall impression of advertising, including imagery and implied claims.
Build an Asset Provenance Register
For every production asset, record:
| Field | Why it matters |
|---|---|
| Asset ID and campaign | Traceability |
| Creation method | Human, AI-generated, AI-edited or composite |
| Tool and model | Reproduction and vendor risk |
| Source assets | Rights and consent |
| Prompt or edit brief | Intent and audit trail |
| Claims shown or implied | Evidence review |
| Human approver and date | Accountability |
| Final placements | Context and disclosure check |
Do not place confidential customer information or unlicensed source material into prompts. The register should point to approved records, not become a second repository for sensitive data.
Separate Low-Risk and High-Risk Uses
Low-risk uses may include background extension, resizing, colour variations or abstract decorative imagery. Higher-risk uses include:
- realistic people who could be mistaken for customers or employees
- demonstrations of a product outcome
- before-and-after imagery
- health, finance or safety contexts
- culturally sensitive representation
- regulated products or services
- competitor comparisons
Higher-risk does not always mean prohibited. It means the asset needs stronger evidence and approval.
Review the Product Depiction
Generative systems can alter packaging, interfaces, accessories, proportions and functional details. A polished image may quietly show a feature the product does not have.
Compare the final creative against the actual product or service. For software, check interface elements and workflow claims. For physical products, verify colour, dimensions, included items and safety-relevant details.
If the image is illustrative, make that clear where a reasonable customer could otherwise interpret it as evidence.
Review People, Places and Representation
Ask:
- Could this person be mistaken for a real customer, expert or employee?
- Does the image imply a testimonial or endorsement?
- Are uniforms, locations or cultural details inaccurate?
- Does the generated group rely on stereotypes?
- Are minors, medical settings or vulnerable audiences involved?
Use real, consented photography where authenticity is central to the claim. AI imagery can be excellent for conceptual scenes; it is a poor substitute for evidence.
Connect Creative and Landing-Page Approval
AI can generate dozens of variations faster than a team can inspect them. Approval must therefore happen at the variant level, not only at the campaign concept level.
Check that each variant agrees with:
- offer and price
- product availability
- audience and geography
- disclaimer and qualification
- landing-page message
- brand tone and visual identity
Automatically assembled assets also need sampling after launch because combinations may create an implication no individual component created alone.
Update the Workflow
A defensible production flow is:
- approved brief with audience, claim and risk level
- rights-cleared source assets
- generation or edit within an approved tool
- creator self-check
- independent brand and claims review
- named approval
- upload and platform preview
- post-launch combination sampling
- retained provenance record
Set a threshold for legal or specialist review. Do not make every crop wait for a committee, but do not let a high-risk realistic claim pass through the same lightweight lane as a background texture.
What to Tell Stakeholders
Use precise language:
- “AI-assisted concept, reviewed and approved by [role]”
- “Illustrative image; product appearance may vary” where needed
- “Generated from rights-cleared brand assets” when true
Avoid “100% safe”, “fully compliant” or “human-made” unless you can support the statement.
A 10-Point Preflight
Before launch, confirm:
- source rights recorded
- no confidential prompt data
- product depiction checked
- claims evidence linked
- realistic people assessed
- brand details accurate
- disclosure behaviour previewed
- landing page aligned
- named human approval recorded
- live combinations sampled
The best response to AI labelling is not cosmetic. It is a production system that deserves the transparency it receives.
Official Sources
Frequently Asked Questions
Let's discuss your project
Need an AI-assisted creative workflow with real approval controls?
