Distinguish a disclosed fictional presenter from a fabricated endorser by controlling scripts, experience claims, substantiation, identity, disclosure and audience interpretation.
Short answer: Prove a synthetic spokesperson is not a fake testimonial by showing that the character is clearly disclosed as synthetic, does not claim personal experience it cannot have, does not imitate a real customer without permission, and delivers only substantiated brand claims. Keep the approved script, source evidence, creation records, disclosure render and audience-interpretation test for the exact final asset.
A synthetic face can function as a narrator, a fictional brand character, a dramatized user or an apparent customer. Those are not interchangeable. The risk rises when first-person language, realistic styling, social proof or production choices lead a reasonable viewer to believe a real person used the product and reported an experience.
The governance task is to define the character’s role before scripting and then preserve that role in the final execution. A disclaimer cannot reliably cure a script that invents experience. CDM’s evidence pack connects role, claims, rights, disclosure and audience interpretation so reviewers can assess the complete impression rather than one sentence in isolation.
What the official evidence establishes
The FTC’s endorsements, influencers and reviews guidance links to the Endorsement Guides and the Consumer Reviews and Testimonials Rule and explains that endorsements and reviews must meet consumer-protection standards. Campaign counsel must apply the current rules to the full presentation and target market.
The European Commission’s Article 50 transparency guidance separately addresses transparency for certain synthetic or manipulated content in the EU. A synthetic-media label and testimonial truth are different controls; satisfying one does not automatically satisfy the other.
CDM inference: the lowest-risk role is a clearly fictional presenter delivering substantiated brand copy without personal experience. As the character becomes more human-realistic or uses first-person experience, the evidentiary burden and likelihood of audience confusion increase. This is an editorial risk model, not a legal safe harbour.
The CDM Synthetic Endorser Evidence Pack
Build the pack before production and update it for the final cut. The same script can create a different impression when paired with realistic styling, a customer-like name, comments or testimonial framing.
| Control | Evidence | Release question | Red flag |
|---|---|---|---|
| Role definition | Character brief and prohibited representations | Narrator, fictional mascot, dramatization or endorser? | Role changes between brief and edit |
| Claim ledger | Sentence-level substantiation and qualifiers | Can the brand prove every objective claim? | Avatar says “I used” or implies results |
| Identity and rights | Model source, likeness review and permissions | Could viewers mistake it for a real person? | Celebrity/customer resemblance |
| Disclosure | Copy, placement, timing and accessibility test | Will viewers understand it is synthetic? | Disclosure only in caption or end card |
| Audience interpretation | Structured comprehension test and sample | What did people think the character represented? | Material minority believes it is a real customer |
| Distribution control | Approved versions, partners and edits | Will treatment survive each placement? | Affiliate crops or rewrites the asset |
How to run the Synthetic Endorser Evidence Pack
1. Define the role before writing
State whether the character narrates brand information, represents a fictional persona, dramatizes a scenario or conveys an endorsement. Ban first-person experience and unverifiable identity cues unless counsel approves a documented basis. Name the business owner for role integrity.
2. Create a line-by-line claim ledger
Classify every sentence as objective product claim, opinion, instruction, dramatization or disclosure. Attach substantiation and required qualifiers. Review voiceover, on-screen text, demonstrations and implied comparisons together; the avatar’s tone does not turn a factual claim into harmless fiction.
3. Review identity and resemblance
Document how the face, voice and motion were produced, licenses and consent, and checks for resemblance to employees, customers, creators or public figures. Avoid realistic biographical details that imply a real customer record. Escalate voice or likeness cloning.
4. Design synthetic-media disclosure
Use clear language at a perceptible point in the experience, then test mobile, sound-off, captions, short edits and platform placements. Keep any applicable machine-readable provenance. Disclosure should explain the synthetic nature; it should not be worded as a claim that every legal issue is resolved.
5. Test the net impression
Show the final asset—not just the disclosure—to a representative sample. Ask open questions about who the character is, whether it used the product, whether results are typical and who made the claims. Predefine the confusion threshold and remediation before seeing results.
6. Control versions and partners
Hash or version the approved master, lock material script changes and supply non-removable disclosure guidance. Require partners to use only approved derivatives. Monitor live placements and comments for evidence of confusion, then pause or correct when the character is presented as a real customer.
Decision rules
No single artefact clears the campaign. Release requires an aligned role, truthful claims, rights, disclosure and acceptable audience understanding.
| Signal | Decision | Evidence to retain |
|---|---|---|
| Script says or implies the synthetic character personally used the product | Remove the experience claim or use a genuine documented endorser | Script diff and revised asset |
| Character resembles an identifiable person without clear rights | Hold and investigate likeness/voice source | Creation record and specialist review |
| Objective claim lacks substantiation | Remove, qualify or substantiate before production | Claim ledger and evidence |
| Disclosure disappears in a short or sound-off version | Redesign the derivative; do not rely on the master | Placement-specific render test |
| Audience test materially misidentifies the character as a customer | Revise role cues, script and disclosure, then retest | Test instrument, sample and responses |
Build the evidence record before launch
Store evidence for the final distributed version and the market in which it runs. A storyboard approval cannot stand in for a materially different edited advertisement.
- Character role, visual/voice generation method and prohibited claims.
- Final script, storyboard, on-screen text and claim substantiation ledger.
- Likeness, voice, training/source or licensing records as applicable.
- Human and machine-readable disclosure treatment and format tests.
- Audience-interpretation method, sample, thresholds and result.
- Approved master and derivatives, partner instructions and live-placement captures.
- Counsel sign-off, monitoring owner, complaint route and withdrawal plan.
Protect model, performer and participant data. Audience testing should use appropriate consent and minimisation. The public campaign record can explain the role and disclosure without exposing confidential production or research information.
Separate fact, CDM inference and the operating decision
Documented fact
US FTC materials govern endorsements, testimonials and consumer reviews, while EU Article 50 guidance addresses certain AI transparency duties; they answer different compliance questions.
CDM inference
A clearly fictional narrator with no experience claims is less likely to be interpreted as a testimonial than a photorealistic first-person customer character, but the final net impression should be tested.
Operating decision
Do not release when the synthetic character implies a nonexistent experience, lacks claim substantiation or rights, loses its disclosure in distribution, or is materially misunderstood in the planned context.
Worked example: a fictional presenter rewritten before launch
A fictional skincare brand creates a photorealistic presenter whose script opens, “I switched two weeks ago and my skin has never looked better.” No real person had that experience. A small end-card says the presenter is AI-generated.
The evidence pack classifies the line as a personal-experience testimonial that the character cannot support. The team changes the role to a disclosed virtual brand presenter, removes first-person use and replaces the result claim with substantiated product information and accurate qualifiers. The disclosure appears early and in captions.
An audience check still finds that some viewers think the presenter is a paid customer because of testimonial-style framing. The team changes the introduction, visual setting and language, then retests. This method reduces ambiguity; campaign counsel must still assess the final asset and jurisdictions.
Limitations and review boundary
Whether an execution is an endorsement, testimonial, dramatization or other advertising depends on facts, jurisdiction and net impression. This article is not legal advice or a determination that a specified disclosure is sufficient. Obtain advertising and AI-transparency counsel.
Audience research has sampling, wording and context limits. Synthetic-media tools and platform labels change. A passing test cannot cure a false claim or unauthorised likeness, and a disclosure cannot make a fabricated experience true.
Related guides
Frequently asked questions
Can a synthetic avatar read ordinary brand copy?It may be possible, but the copy still requires substantiation, rights and appropriate transparency. Avoid styling or first-person language that transforms a narrator into an apparent real customer.
Does saying “AI-generated” make a fake testimonial acceptable?No. Disclosure of synthetic production does not substantiate a claim or create a real experience. Remove invented testimony and review the complete impression.
Can we clone an employee’s voice with consent?Consent and licensing are essential but may not resolve every labour, privacy, publicity, disclosure or consumer issue. Define scope, duration, revocation, security and markets with specialist review.
How should we test audience interpretation?Show the final placement to a relevant sample and ask neutral, open questions about who the speaker is, whether they used the product and who is making the claims. Predefine what response requires revision.
What if a platform automatically labels the asset?Treat the platform label as one layer. Verify its wording, timing, accessibility and persistence, and maintain your own approved treatment where required. Platform behaviour can vary by placement.
Method and publication status
This prepublication draft was researched from the primary materials linked at the claims they support and checked on 12 October 2026. It requires the named specialist review, live-product retesting where indicated, a real 16:9 editorial image and final URL verification before publication. It is operational analysis, not legal, security or compliance advice.
This article is editorial guidance. Apply the principles in proportion to your market, evidence, and responsibilities.
