AI images can run in ads. Keep the receipts.

“Can we run this AI-made image in the client’s paid campaign?” The Artificial Intelligence (AI) tool made the picture in seconds, so this is the question left over. Three smaller ones answer it: who owns the image, who pays if someone objects, and where it came from.

Can an AI image run in a paid ad at all?

Often, yes. But “can we” is three questions wearing one coat. A useful picture is the rental counter. You can usually drive off with the car. Whether you hold the title, whether the insurance pays out, and what the receipt says are separate matters, and each lives in different paperwork.

Who owns an AI image?

Possibly nobody, which matters the day a competitor copies the ad. In its January 2025 report on copyrightability, the U.S. Copyright Office concluded that prompts alone do not provide sufficient human control to make users of an AI system the authors of the output¹. Its wider finding is that copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements¹. A federal appeals court reached the same place in March 2025. In Thaler v. Perlmutter, the D.C. Circuit held that the Copyright Act requires all eligible work to be authored in the first instance by a human being².

People still own what they add. The Office says copyright protects the original expression of a human author even if the work also includes AI-generated material¹, and the court noted that the rule still rewards people who pursue rights in works that they make with the assistance of artificial intelligence². Retouching an image, or building it into a layout your designers made, is the kind of human contribution that can count. Whether a given edit is enough is a question the Office says depend[s] on a case-by-case determination¹. That is the U.S. view. The same report surveys other countries, and a global client’s campaign runs in many of them.

Who pays if someone says it copies their work?

Ownership covers your claim on the image. Cover answers the other side: a third party says the output copies theirs. Some vendors now put that risk in the contract. Microsoft says it will defend the customer and pay the amount of any adverse judgments or settlements that result from the lawsuit³ when a commercial customer is sued for copyright infringement over its Copilots or their output, provided the customer used the content filters and other safety systems built into the product. Google Cloud offers a generated output indemnity too, and it only applies if you didn’t try to intentionally create or use generated output to infringe the rights of others⁴. Both are conditional, and both are contract terms, so they follow the plan you bought and the settings you used. Read the clause in your own agreement before you rely on a vendor’s blog post.

Will the ad platform label it?

Then the platform weighs in. Meta says that when its in-house generative tools make a significant edit to the image or video⁵, it applies an AI label. It also detects ads created or edited using third-party AI tools through industry-standard signals⁵ and labels those “AI info”. If the tool put a photorealistic human in the ad, the label sits beside the Sponsored tag instead of behind the three-dot menu. How labeling rules differ by law and by platform is in our brief on labeling AI content⁶, and the provenance data a platform can read is the subject of our brief on content credentials⁷.

What's Real?

• •Pure AI output may carry no copyright. The Copyright Office says protection does not extend to purely AI-generated material¹, and the D.C. Circuit has held that the Copyright Act requires a human author in the first instance².
• •Human work on top still counts. Protection reaches the original expression of a human author, even if the work also includes AI-generated material¹. How much human input is enough is undecided, and the Office calls it a case-by-case question.
• •Indemnity exists, with strings. Microsoft ties its commitment to using the content filters and other safety systems³ in the product, and Google’s applies only if you did not try to infringe on purpose.
• •Platforms are building labels in. Meta applies an “AI info” label to ads created or edited using third-party AI tools⁵ when it detects the signals, so a quiet AI image may not stay quiet.

The “so what” — for anyone serving tech clients

• •Put the rights check in the brief. Before anyone generates anything, record which tool and which plan will make the image, and whether its terms cover commercial use and indemnity.
• •Add human work you can point to. Retouch it, composite it, set it in a layout your designers built. That is the part the law can protect, and it keeps the final asset yours.
• •Keep a receipt file. Save the tool, the version, the prompt and the date with the final image, plus any content credentials, so “where did this come from?” has a one-minute answer. Brand fit is a separate check; the layered approach to brand safety⁸ applies here too.
• •Tell the client what is covered. Say plainly which parts of the asset are protected, which are indemnified by the vendor, and which depend on a person’s edits.