11.3 Disclosure in Practice: Templates That Hold Up

Everything in 11.1 and 11.2 reduces to one editorial decision made hundreds of times: what the disclosure says, and where it sits.

The rule that settles most of those decisions is this. A disclosure must be unavoidable, not available. Content that requires a click, a scroll, an expansion, or a second screen has not been disclosed. This is the standard US regulators apply in advertising, and it is the right standard to apply everywhere, because it is stricter than what most platforms require and it survives being tested by someone who is not on your side.

Three placements, three jobs

Stop thinking of disclosure as a text you write. Think of it as three separate placements, each with a different job.

Most operators do the first, forget the second, and never build the third. The third is the one with a chat-first business model, because chat is where the revenue is.

Account level: one blank line, permanent

This is the blanket statement. On Instagram it corresponds to the AI-generated profile declaration from 11.2. On every platform, it belongs in the bio regardless of whether a setting exists, because the bio is the page a journalist, a brand, or a regulator opens first.

Two requirements: it must be in the first visible lines, and it must use the word. Not "digital creator." Not "virtual artist." Say what it is.

Profile bio line (EN): AI-generated persona. Images and voice are synthetic. or, where a declaration toggle already exists: AI-generated profile · 100% synthetic images and voice

On a Polish edition the same line reads Persona wygenerowana przez AI. Obrazy i głos są syntetyczne. — the wording is fixed once and reused verbatim, so it is identical across every post.

The word choice is not stylistic. "Virtual" and "digital" describe a medium. "AI-generated" describes the thing that triggers obligations on four platforms and in one statute.

Asset level: when text, when the platform label, when both

The decision rule is about ambiguity, not about technology.

Platform label alone is sufficient when the platform provides a visible mechanism, the label appears at first exposure without a tap, and the content is clearly stylized. This covers most output from an illustrated or otherwise non-photoreal persona on YouTube, which shows the label publicly on the watch page.

Your own text alone is what you use when no platform mechanism exists, or where you publish outside the four platforms — a blog, an email, a landing page, a print asset. The obligation does not travel through the platform; it travels with the content.

Both is the correct answer whenever any of these is true:

• The content is paid or sponsored.
• The content is photorealistic, or depicts a person or event that reads as real.
• The content represents the persona as giving advice, recommending a product, or sharing an experience.
• The content is a screenshot, a still, or a clip that will be reuploaded by someone else.

That last one is the one operators miss. A still pulled out of your video and reposted elsewhere carries no label, no metadata, and no context. If the disclosure was only in the video, the still is undisclosed. Put it on the frame where a reupload can keep it.

Caption disclosure (EN): This image is AI-generated. The persona is synthetic and does not depict a real person.

Paid post, caption opener — before the "more" fold: AI-generated persona · Paid partnership with [brand]

On-frame mark, corner or lower third, legible at thumbnail size: AI-GENERATED

The caption version has one hard requirement: it goes in the visible part of the caption, before the truncation point, or in the first seconds of the video. A disclosure that lives after the "more" fold is not a disclosure, and neither is a pinned comment.

Interaction level: the chatbot opener

This is the placement with legal force behind it — Article 50(1) from 11.1 — and it is also the one that costs you the least if you design it well.

The failure mode is a persona that answers in character for forty messages and mentions being an AI only if asked. That is not disclosure. Disclosure happens before the user forms the belief.

The design constraint is real: a blunt "I am an AI" delivered as a system notice breaks the product. So write it in the persona's voice and keep it to one line.

DM opener (EN), before the first substantive reply: Before we start: I'm an AI persona. Everything you see and hear from me is generated. Want to keep going?

Three properties make that work. It arrives before the belief forms. It is unambiguous — "AI persona" is not ambiguous. It does not ask permission to be an AI; it asks whether the user wants to continue, which keeps the conversation moving and keeps the acknowledgment logged.

Log it. Chapter 11.5's register wants the timestamp and the message, because the only defensible answer to "was the user informed" is a record showing that they were.

Advertising: the standard is stricter and the penalties are per violation

Paid content sits under advertising law, which has no artistic exception, no platform label that satisfies it, and a per-violation penalty structure.

The FTC's endorsement rules require disclosures to be clear and conspicuous. Operators read that as "present." It means "unavoidable." The patterns that fail, in order of how often they appear:

• A hashtag buried in a block of twenty hashtags.
• #ad placed after the "more" fold on a long caption.
• The disclosure in a pinned comment, which requires the reader to tap.
• The disclosure only in the endorser's bio, on a different screen.
• The disclosure flashed in the final two seconds of a video.
• The disclosure written in a language the audience does not read.

Contrast with what passes: Ad or Paid partnership in the first line of the caption, plus the platform's own paid-partnership tag, plus a spoken line or a persistent on-screen mark in video. Redundant by design, because each viewer encounters content differently.

The penalty anchor. The FTC's maximum civil penalty runs to $53,088 per violation, and each undisclosed post is a violation. [AS OF 2026-10] The enforcement history is what makes the number real: $9.4 million against Google and iHeartMedia in 2019 for undisclosed radio endorsements that never aired as the script described, and $1.26 million against Kim Kardashian in 2022 for promoting a crypto asset without disclosing the payment. The tea brand Teami settled undisclosed-endorsement and health-claim charges in the same period. [AS OF 2026-10]

Three lessons from those cases, all of which apply to a persona account:

1. The penalties land on the brand and the advertiser, not only the creator.
2. The undisclosed element is usually the payment, not the product.
3. The agency does not need intent. It needs the post.

If your persona takes sponsorship, the disclosure belongs in the contract as a deliverable requirement, not in your judgment on the day. Chapter 9.6 covers brand-deal structure.

Synthetic performer laws: a second, newer layer

Beyond advertising rules, several US states have created specific rights around digital replicas of performers. California's rules cover digital replicas of deceased personalities and require contractual representation for replicas of living performers. New York's law addresses the use of a performer's digital replica, and Tennessee's ELVIS Act covers voice. [AS OF 2026-10]

What that means for you in practice, and where to look:

• If your persona is built from a real performer's likeness, these laws apply in addition to publicity rights. Chapter 11.4.
• If you sell services replacing a performer's work with a generated one, the contract you sign is the regulated artifact, not the video.
• If your persona is an invented composite with no real referent, these laws are not your problem — which is one more argument for inventing rather than deriving.

The text of these statutes moves. Verify the current version before relying on any summary, including this one.

What you leave with

Five fragments and one decision rule. The fragments:

1. AI-generated persona. Images and voice are synthetic. — bio, first line.
2. This image is AI-generated. The persona is synthetic and does not depict a real person. — caption, before the fold.
3. AI-generated persona · Paid partnership with [brand] — paid posts, first line.
4. AI-GENERATED — on-frame mark, legible at thumbnail size.
5. Before we start: I'm an AI persona. Everything you see and hear from me is generated. Want to keep going? — DM opener.

The rule: unavoidable, not available.

These are English templates. Every localized edition translates them into the target language and keeps the platform tag verbatim — Paid partnership is a platform label, not a phrase to translate.

Where this goes wrong

Treating disclosure as one text used everywhere. Three placements, three jobs. A bio line does not cover a reuploaded still.

Placing the text after the truncation point. This is the most common failure and the least defensible, because the disclosure exists and does not function.

Writing vague copy to soften it. "Digital creator" and "virtual artist" describe a medium and disclose nothing. Vague disclosure is worse than none, because it shows you knew.

Trusting the platform label to carry the legal duty. From 11.2: the label satisfies the platform, your own disclosure satisfies the law.

Forgetting the chatbot. Chat-first businesses earn most of their revenue in a surface where they have no disclosure at all. Fix the opener before you scale the automation.

Copying the templates without localizing them. A disclosure in English on a Hindi-language account is not clear and conspicuous to that audience.