Notes7 min read

AI made the ad. A real customer said the words.

Not fake ones. Since October 21, 2024, a US FTC rule bans testimonials that misrepresent that the person exists or used the product, and AI-generated ones are named. Use AI to produce content from real, consenting customers, and label realistic AI media.

From the course · lesson 3.12

AI video tools can now produce a smiling, natural-looking "customer" in a few minutes. They'll say whatever you type. For a new product with no reviews yet, that's tempting.

Don't use AI-generated testimonials. In the US they can break a federal rule, most big platforms require labels on realistic synthetic people, and many buyers dislike it even when it's disclosed. Take Kudos, a made-up testimonial app. One fake customer would undercut the whole product. Kudos exists to show proof that's real.

The good news is that AI is still useful here. It just belongs on the production side, never on the endorsement side.

The FTC rule on AI-generated testimonials (United States)

The Federal Trade Commission (FTC) is the US agency that polices deceptive advertising. In August 2024 it voted 5–0 to approve a rule on consumer reviews and testimonials. The rule took effect on October 21, 2024, and lives at 16 CFR Part 465.

A testimonial, in the FTC's words, is an advertising message people are likely to believe reflects a real consumer's (or celebrity's) opinions or experience. The core ban says a business can't write, create or sell a review or testimonial that materially misrepresents any of these:

  • that the person exists,
  • that they actually used the product, or
  • what their experience with it was.

A business also can't buy or spread one that it knew, or should have known, was fake in those ways. The FTC's announcement names "AI-generated fake reviews" as an example of reviews by someone who doesn't exist. The agency can seek civil penalties from people who violate the rule knowingly.

The same rule also bans paying for reviews that must say something positive (or negative), undisclosed reviews from company insiders, scaring people into removing bad reviews, and buying fake followers or views.

What the FTC says about AI avatars

The FTC's own Q&A on the rule says there is "no blanket prohibition" on AI-generated avatars in marketing. A clearly fictional brand character is not, by itself, a fake review. It becomes a problem when the avatar is used to deliver a testimonial that's fake.

The FTC's older Endorsement Guides (16 CFR Part 255) add three points that matter for AI production:

  • If an ad presents people as actual customers, it should use actual customers in both the audio and video, or clearly disclose that they aren't.
  • An endorsement can't be reworded or taken out of context in a way that distorts the person's opinion or experience.
  • If an ad says the endorser uses the product, they must have really used it when they gave the endorsement.

So "an AI actor reading a real customer's quote" isn't automatically fine. If viewers would take that actor for the customer, you need a clear disclosure, or better, use the customer.

Platform labeling rules

The platforms have their own rules on top of the law. These cover posts you publish. Paid ads have separate ad policies, so check those too.

PlatformWhat it asks of you
YouTubeDisclose content that's meaningfully altered or synthetic and looks realistic, such as a real person appearing to say something they didn't. Using AI for scripts, titles or thumbnails, beauty filters, or cloning your own voice for a voiceover doesn't need disclosure. Creators who keep skipping it can face penalties.
TikTokLabel AI-generated content that contains realistic images, audio or video. TikTok may label it automatically, for example when the file carries Content Credentials (a provenance standard from the group known as C2PA). It doesn't allow AI likenesses of adult private people without their permission.
Meta (Facebook, Instagram, Threads)Use Meta's disclosure tool when you post photorealistic video or realistic-sounding audio that was digitally created or altered. Meta may apply penalties if you don't. Labeled posts show "AI info."

The EU AI Act (European Union)

Article 50 of the EU AI Act sets transparency rules that apply from August 2, 2026. For marketers, the key part is about deep fakes: AI-made or altered images, audio or video that resemble real people, places or events and would falsely seem authentic. Those must be disclosed. Content that is clearly artistic, satirical or fictional gets a lighter duty to note it in a way that doesn't spoil the work. Whether the Act applies to you depends on where your AI is used and who sees it, which is a question for a lawyer.

What buyers think

Even legal, labeled AI people carry a cost. In a Bentley-Gallup survey of 3,270 US adults in May 2026, 49% viewed businesses' use of AI to make ads negatively and 19% positively. Sixty-two percent said AI-created people or voices in ads are unacceptable even when disclosed. Adults aged 18 to 29 were the most negative, at 66%. Yet 75% found it acceptable to use AI, with disclosure, for brainstorming and early drafts.

For a product built on trust, that settles it. Real faces for trust, AI for the craft around them.

What AI may do, and must never do

AI can help withAI must never
Captions, trims, color and sound cleanup on a real customer's videoGenerate a person and present them as a customer
Reframing one clip for 9:16, 1:1 and 16:9Write a quote the customer didn't give or approve
Quote cards from the customer's approved words"Improve" wording so it says more than they meant
B-roll, product shots, motion graphics, explainersUse a customer's face or voice beyond what they agreed to
Translating or dubbing, with consent and a labelPost realistic AI faces or voices without a label
Testimonial clipsStandby
One real customer clip becomes a master edit, the cost quote is approved, then the batch is reframed for every format.

The Kudos example

Here's how this can fit together in Kudos:

  1. Collect. Kudos asks a real client for feedback. They can type or talk.
  2. Draft and approve. AI tidies their rough words into a short quote. The client sees the final wording and approves it. If they change it, their version wins.
  3. Record consent. Kudos stores who said it, the approved text, the date, and where it may be used, including whether AI editing or dubbing is allowed.
  4. Produce. Agents turn the approved quote and clip into captioned videos, quote cards and a short case study, after you approve a cost quote for any paid generation.
  5. Label. Anything with realistic AI changes gets the platform label before it's posted.
  6. Block. A hook, a rule that runs automatically before anything is published, stops any asset that shows a generated person as a customer or has no consent record. You can't forget it on a busy day, because it doesn't rely on you remembering.

The pitch writes itself: real testimonials, AI-produced. The compliance work is also the feature.

Try it

Pick one real piece of praise you've received, from a client email, a message or a call.

  1. Ask the person if you can use it, and where.
  2. Draft a short quote from it, then send them the exact wording to approve.
  3. Write a one-line consent record: name, approved text, date, allowed uses, and whether AI editing is OK.
  4. Make one quote card from it. Check it against the table above before you post.

That's the whole loop at small scale. In the course you build this into your own project and automate the checks. Real proof like this is also what helps a vibe-coded app find its first customers. If you write a press release with an imaginary customer quote, keep that quote internal, as Write the press release before you build explains. You can see where it fits on the course page.

Sources

Learn this inside Claude Code.

Join the waitlist for Chief of Agents. You'll get the free beginner's handbook and skills right away, and hear first when the course opens.

One email to confirm, then the handbook and skills. Unsubscribe anytime.

Keep reading

  1. 6 min read

    From designer to builder: directing a crew of agents at work

    What working with AI agents looks like day to day after ten years in design: the work of direction, and what stayed the same.

  2. 6 min read

    Building with AI got cheap. Judgment didn't.

    When AI can build an app in an afternoon, the hard part is judgment: what to build, what good looks like, and who will pay.

  3. 6 min read

    AI closed the skill gap. It widened the taste gap.

    Why does AI output look generic? It drifts to the average. Here's why, plus a side-by-side drill that trains your eye to fix it.

All notes →

© 2026 Chief of Agents · Notes · PrivacyIdeate. Build. Sell. You lead, the agents work.