The Ad Machine

Five platforms at enterprise scale

• •Adobe Firefly: 9 billion+ images generated since March 2023 launch. Firefly Services API in production at Marriott, IBM, Gatorade, and others — enabling automated enterprise creative pipelines that generate thousands of asset variations with minimal human review. Positioned as "commercially safe" generative AI due to licensed training data. IP ownership of outputs depends on human authorship documentation, not platform terms.
• •Meta Advantage+: 4M+ advertisers using AI campaigns that give Meta autonomous control over creative generation, audience targeting, placement, and bid optimization. Advantage+ generates image backgrounds, text overlays, and creative variations without advertiser approval of each combination. Reports 22% lower cost per acquisition — at the cost of creative governance control.
• •Google Performance Max: replaced Smart Shopping and Local campaigns — not as an option but as the default. AI controls bidding, audience, placement across Search, Display, YouTube, Discover, Gmail, and Maps, and creative asset combinations. Enterprise media teams built campaign governance around human-controlled ad groups; Performance Max removes that structure.
• •Persado: AI generates emotion-optimized copy at sentence level, trained on behavioral response data across hundreds of millions of consumer interactions. Deployed at Goldman Sachs (450% CTR lift in one tested campaign), JPMorgan Chase, Humana, Bayer, Prudential. EU AI Act Article 5 applies to emotion-optimization tools for EU campaigns.
• •Agency AI platforms: WPP Open ($250M+ investment), Publicis.ai, Omnicom OMNI, IPG AI. Agency AI creative generation runs on proprietary platforms under agency MSA terms — most enterprise marketing agreements predate these platforms and do not address data rights or IP ownership for AI-generated agency creative.

What AI actually does in enterprise marketing creative — and where governance breaks down

• •Autonomous creative combination: Performance Max and Advantage+ assemble ads from approved asset libraries without human review of each combination. The approval workflow governs individual assets; it does not govern combinations. Compliance-sensitive claims, competitive positioning, and brand standard violations can appear in AI-assembled combinations that passed individual asset review.
• •Emotion-optimization at sentence level: Persado and Jacquard (formerly Phrasee) generate copy variants optimized for specific psychological trigger patterns by audience segment. Most enterprise marketing teams deployed these tools on the basis of A/B performance data without a legal review of the optimization mechanism against EU AI Act Article 5 standards.
• •Automated creative at pipeline scale: Firefly Services API generates thousands of localized, personalized, or variant creative assets in automated pipelines. Human authorship is present at the pipeline design level but may not be present at the individual asset level — creating copyright protection gaps for assets that could not exist without AI generation.
• •AI-generated copy for claims-bearing content: Jasper, Copy.ai, and comparable tools are generating advertising copy including comparative claims, performance claims, and benefit statements. FTC substantiation requirements apply to these claims regardless of whether a human or AI wrote them. Most enterprise review workflows treat AI-generated copy as equivalent to human-written copy without AI-specific substantiation checkpoints.

The governance gap: four dimensions most enterprise CMOs have not closed

• •IP ownership: The USPTO's February 2023 guidance established that AI-generated creative without meaningful human authorship is not eligible for copyright protection. Most enterprises have not audited which active campaign assets have defensible copyright versus no protection. Competitors can legally reproduce AI-only creative.
• •Data rights: Platform agreements for Adobe Firefly, Meta, Google, and Persado include rights for vendors to use creative inputs, outputs, and performance data for model training. Most enterprises signed these agreements before AI creative generation was a platform feature. Default terms in some accounts grant training rights without active configuration to opt out.
• •EU AI Act Article 5: Prohibitions on AI systems using "subliminal techniques" or "manipulative techniques that exploit psychological vulnerabilities" took effect August 2, 2024. Emotion-optimization tools deployed in EU campaigns require qualified legal review against this standard. Most enterprise marketing legal teams have not conducted tool-by-tool Article 5 assessment.
• •FTC compliance: AI-generated testimonials, endorsements, and comparative claims carry identical substantiation requirements to human-written ones. AI copy generators produce claims-bearing content at a volume and velocity that outpaces the review workflow. Most enterprise advertising approval workflows have no AI-specific checkpoint for substantiation review of AI-generated claims.

Four governance postures.

Five material risks.

Regulatory posture and platform data rights summary

• •USPTO / Copyright: AI-generated creative without meaningful human authorship is not eligible for copyright (February 2023 guidance; consistent Copyright Office decisions through 2025). Human selection, curation, and arrangement creates authorship — but documentation must exist at the time of creation.
• •EU AI Act Article 5: Prohibited practices include AI systems using subliminal or manipulative psychological techniques in commercial contexts. In force August 2, 2024. European AI Office guidance on advertising-specific applications in progress. Enforcement posture: active. No grace period for compliance.
• •FTC: 2023 policy statement on AI impersonation, deceptive endorsement rulemaking, and ongoing dark patterns enforcement all have direct advertising applications. Substantiation requirement applies to AI-generated claims. Disclosure requirement applies to AI-generated testimonials and endorsements.
• •Adobe Firefly: Enterprise tier includes opt-out for model training; default configuration in some account types grants training rights. Firefly Services API terms require review for enterprise pipeline deployments. "Commercially safe" claim addresses training data origin, not output copyright.
• •Meta / Google: Advertising terms grant broad rights to use creative and performance data for platform optimization. These rights predate AI creative generation as a feature. Opt-out provisions are limited. Most enterprise advertisers have not reviewed current terms against AI creative use cases.
• •Agency platforms (WPP Open, Publicis.ai, Omnicom OMNI): Data rights and IP terms for client creative within proprietary agency AI platforms are governed by individual agency MSAs. Most MSAs predate these platforms. Require explicit addenda for AI creative data rights, model training, and IP ownership.