The agency-specific case for AI disclosure
Most agencies treated the EU AI Act as a client-side problem — something the in-house legal team at the brand would handle. That reading hasn't aged well. Article 50 attaches to the "deployer" of the AI system, defined as "a natural or legal person… using an AI system under its authority." If your account team prompts Midjourney to generate a hero image for a client landing page, the agency is the deployer in that transaction. The client is the publisher. Both touch the asset; both can be reached.
That joint-and-several reading isn't hypothetical. The first wave of regulator guidance from the European Commission's AI Office has explicitly named "contractors and creative agencies producing content on behalf of platforms" as in-scope deployers. The practical consequence: your MSAs need a clean disclosure clause, your account managers need a per-client SOP, and your delivered assets need labels that survive whatever happens downstream. The generator handles the asset-and-SOP layer; your legal team handles the MSA. Together it's the smallest viable compliance footprint for an agency running multiple client books in 2026.
The four channels where agencies actually owe disclosure
Disclosure obligations don't map cleanly to a single content type — agencies produce in mixed surfaces, and each surface has its own placement rules. The generator's presets cover the four where agency exposure is highest:
- Editorial & SEO content for client blogs. Cover image from Midjourney or Firefly, AI-assisted drafting, sometimes AI voice for the audio version. Per-post badge above the byline plus an image overlay; JSON-LD schema in the post's
<head>. The white-label preset strips any agency-identifying language so the post reads as the client's own. - Paid social & performance ads. Meta's "AI Info" flag, TikTok's "AI-Generated Content" toggle, and Google Ads'
aiContentDisclosurefield are platform requirements that don't replace on-creative labeling. The generator emits the platform-specific JSON your ad-ops or media-buying tool can ingest, plus the on-creative overlay sized for 1:1, 4:5, and 9:16. - Email campaigns & lifecycle marketing. Subject-line A/B tests written by GPT, AI-generated hero images in the email body, AI-personalized copy variants. The disclosure goes in the email footer with a one-line schema-equivalent for analytics. Less visible than blog placement, but auditable.
- Influencer & endorsement deliverables. When the agency briefs an influencer and any of the deliverable was AI-produced on the agency's behalf, the FTC's 2024 Endorsement Guide update requires that to be disclosed alongside the standard #ad. The generator's combined preset emits both labels in a single line.
Per-client preset libraries: how agencies actually scale this
The slowest version of compliance is the version where every account manager opens the generator from scratch for every brief. The fast version is a per-client preset library. The pattern that has worked for the multi-client teams we've seen:
- One preset per client × channel. Acme DTC blog. Acme DTC paid social. Acme DTC email. Each preset locks the brand voice line, the responsible-party language, and the JSON-LD
publisherfield so account managers don't edit them. - Schema in the CMS partial, not the asset. The visible badge ships per-asset. The JSON-LD block lives in the client's post template head. New posts inherit it automatically; deliverable count goes up but compliance overhead stays flat.
- Preset name in the PM tool deliverable record. Asana, Monday, Notion, Wrike — whichever you run, the deliverable card stores the preset slug used. That's your audit trail when a regulator or a client's in-house counsel asks for one.
Built that way, the marginal cost of disclosure on a new asset is one click in the PM tool plus a paste in the CMS. Below the threshold where account managers route around it.
Compliance is also business development
The agencies winning enterprise pitches in 2026 are the ones whose RFP responses include an AI compliance posture. Procurement teams have started asking. The agencies without an answer are losing the round.
Enterprise procurement now routinely includes AI-handling questions in RFPs — not just whether you use AI, but what disclosures, what audit trail, what indemnification posture. A documented per-client disclosure SOP backed by a generator and a preset library reads as "mature" in a way that "we'll figure it out per-engagement" does not. That's a real revenue lever, particularly in regulated verticals where the alternative answer kills the relationship at the legal-review step.
Regulated-industry clients: pharma, finance, legal
If your client book includes any of these, the EU AI Act sits on top of an existing compliance stack rather than replacing it. The generator's presets stack the layers in the right order so the asset doesn't bounce back from client-side review:
- Pharma & medical-device clients. EU MDR plus EMA promotional review on top of Article 50. The preset emits the AI label plus a placeholder for the EU-MDR-required device identifier and a flag the medical legal-regulatory (MLR) reviewer expects.
- Finance & fintech clients. MiFID II marketing-communication labeling in the EU, US FINRA Rule 2210 for broker-dealers. The generator's finance preset emits both an AI disclosure and the "not investment advice" standard line that ad and content review wants.
- Legal-services clients. Multiple US state bars have published 2024–2026 guidance on AI in attorney advertising. ABA Center for Professional Responsibility has the rolling tracker. The legal-vertical preset reads against the strictest active state rule so the asset clears in any jurisdiction.
Compliance vs. theatre: what bad agency disclosure looks like
| Pattern | What it does | Status |
|---|---|---|
| Agency-wide AI policy on agency website | Doesn't reach the client's readers; not asset-level | Non-compliant |
| Single boilerplate clause in MSA appendix | Contractual but not perceptible; doesn't satisfy Article 50 | Non-compliant |
| Disclosure in client's site footer only | Reader doesn't see it at first exposure to the AI element | Non-compliant |
| Per-asset badge for blogs only, not paid ads | Covers one channel; ad creative is unprotected | Partial |
| Per-client preset across blog, ads, email, social | Asset-level on every channel; survives audit | Compliant |
| Per-client preset + JSON-LD + PM-tool audit log | Reader, scraper, regulator, and client procurement all see it | Best practice |
Workflow for a typical agency content factory
The realistic publishing rhythm for a mid-sized agency is 200–800 client assets a month across 8–15 brands. At that volume, every gram of friction matters. The pattern that holds up:
- Onboarding step. When you sign a new client, build their preset library in the generator that day — before the first deliverable. Five presets (blog, paid social, email, landing page, white-label) take about 20 minutes. Save the slugs into the client's onboarding doc.
- Brief step. The brief template includes a "disclosure preset" field. The strategist or AM picks the preset slug when they create the brief, not later.
- Production step. The producer pastes the preset output. The visible badge goes in the asset; the JSON-LD goes in the CMS template once and stays there.
- QA step. The reviewer's checklist has "disclosure preset matches brief" as a tick item. Five seconds. The audit trail is the brief itself.
Built that way, your agency carries roughly 0.5% additional production overhead for full Article 50 compliance — well below the threshold where teams route around it. For broader context on agency AI governance, the IAB publishes a rolling agency-AI-policy template that pairs naturally with this generator's output.
Frequently asked questions
Is the agency or the client liable for missing AI disclosure under the EU AI Act?
Both. Article 50 treats the agency as the "deployer" and the client as the publisher; joint and several liability is the default. If your MSA is silent on AI, ambiguity flows back to the agency. The generator handles the SOP half; your legal team handles the contract clause.
Do paid ads need AI disclosure too, or only organic content?
Paid ads are explicitly in scope. Meta, TikTok and Google now layer their own platform-level AI flags on top — meeting one rule doesn't satisfy the other. The generator emits both the platform-flag JSON and the on-creative overlay.
How does the FTC's 2024 Endorsement Guide update interact with AI disclosure for influencer briefs?
The 2024 update added AI-generated endorsements as separately enforceable. If your agency briefs an influencer and any of the deliverable was AI-produced, the disclosure trail names that. The combined preset emits the AI label and the FTC #ad disclosure as one block.
How do agencies actually scale disclosure across hundreds of monthly assets?
Per-client preset library · JSON-LD baked into the CMS partial · preset slug logged in the PM tool deliverable record. That's the audit trail. Marginal cost per asset drops to one click.
What about regulated-industry clients — pharma, finance, legal?
The EU AI Act sits on top of EU MDR / FDA, MiFID II / FINRA, and state-bar advertising rules. The vertical presets stack the layers in one block so client compliance review doesn't bounce the asset back.
We white-label content for resellers. Whose name goes on the disclosure?
The disclosure names the AI element, not the agency. The white-label preset removes any agency-identifying language so the publishing brand carries the label. Your reseller MSA should still assign disclosure responsibility explicitly.