Testimonial Video Ads With No Customers Yet: 4 Honest Ways

The advice you will hear most often is wrong in a way that costs founders money: that you cannot run testimonial video ads until you have customers willing to film. You can. The mistake is assuming the format depends on a paying customer's face. It does not. It depends on a four-beat narrative arc, and that arc is agnostic about who delivers it. The data buyers I trust look at hold rate and cost per result, not at whether the speaker swiped a card. So the real constraint at launch is not the customer. It is whether you can produce the arc without inventing one, because inventing one is now a legal liability with a specific name attached to it.

That name is the FTC's revised Endorsement Guides, finalized in 2023. The update did something narrow and consequential: it explicitly addresses fake reviews and AI-generated testimonials, and in 2024 the Commission's companion rule on fake reviews made it formally unlawful to create, buy, or sell reviews that misrepresent the reviewer's experience, including reviews generated by AI for someone who never used the product. Penalties run into civil fines per violation. So the framing problem is no longer abstract. A fabricated quote attributed to "a happy customer named Sarah" is not a gray area; it is the exact thing the rule was written to stop. Everything below stays on the right side of that line and still produces ads that test well.

Why testimonial video ads win before you have a single customer

Two mechanics drive performance, and neither one requires a verified purchase. The first is native feel. A person speaking into a phone camera matches the organic content the feed is engineered to surface, which buys you a half-second of attention before the viewer's ad-detection reflex engages. That half-second is most of the game; on a cold audience, three-second hold rates for native-looking creative routinely beat studio-polished spots by a wide margin, and hold rate is the single metric most predictive of downstream cost per acquisition. The second mechanic is problem-first qualification. Open on a specific pain and you filter the audience for you, so the clicks you pay for skew toward people who recognize themselves.

Both levers come from structure, not from the speaker's billing status. A founder saying "I rebuilt the same revenue report every Monday for a year" carries the identical narrative weight as a customer saying it, and at launch the founder version often qualifies the audience harder because the conviction is real. The same structural logic underpins social proof done well in video ads and the UGC script formulas the platforms keep recommending. Read those for the adjacent formats; the point here is narrower. You are buying the arc, and you can source the arc four legitimate ways.

Four legitimate sources, ordered by the proof they carry

1. Real beta-user quotes, rendered as on-screen text

You do not need a beta user on camera. You need their actual words and their permission. The strongest line you can run this week is probably already sitting in an export you have not searched. Pull it, attribute it accurately ("Early user, B2B agency"), and render it as a styled on-screen quote with a voiceover reading it aloud. You are not staging a filmed endorsement. You are quoting feedback you genuinely received, which is precisely what the Endorsement Guides permit so long as the attribution is true and the result is representative.

The procedure for finding these lines matters, because most founders skip it and write fiction instead. You almost certainly have a corpus of real sentences in churn surveys, Intercom or Slack exports, App Store reviews, and Stripe email replies after a charge. Run that corpus through a model with a tight extraction prompt: "Here are 40 raw messages from beta users. Return only verbatim sentences where the user describes a specific outcome, a time saved, or a problem solved. Do not paraphrase, do not combine sentences, and flag any quote that names a feature." Then a second pass: "For each quote, draft a one-line permission request I can email the author." Send the requests, get a written yes, and you have testimonial copy you can defend to a regulator. The words are real, the consent is documented, and the arc is intact.

2. Founder narration, the "I built this because" arc

Founder narration is a testimonial about the problem rather than a claim about the product, and it is honest by construction because you are the source. The arc is the same: name the pain you personally hit, describe the workaround everyone tolerated, then show the thing you shipped. No customer required, no consent to chase. There is a real craft to making the founder story land without sounding rehearsed, and getting it right before you scale spend is cheaper than discovering it failed at $2,000 in. The advantage over a polished claim ad is conviction the audience can hear, and conviction reads as authenticity even when production is rough.

3. Demonstration-led proof, where the screen is the witness

When you hold no quotes at all, replace the assertion with a demonstration. Show the result on screen instead of having a stranger praise it. A screen recording of the product doing the thing carries more proof than any spoken endorsement because the viewer evaluates it directly and you make no claim you cannot perform live. For a tool, that is the input going in and the output coming out in real time. Demonstrations also sidestep the entire endorsement question, since there is no endorser, only evidence.

4. Aggregate framing, the most abusable and therefore the most disciplined

You can frame category and audience honestly without inventing scale. "Built for performance marketers" and "made for indie founders shipping paid social" are true statements about who the product is for, and they imply momentum without asserting a number. What the rule forbids is the count you cannot substantiate: "trusted by 4,000 founders" when the figure is invented, screenshots of reviews that never happened, or any implied customer base you do not have. Aggregate framing describes the target, not the tally. If a number appears on screen, you must be able to produce the database row behind it.

One fully written testimonial ad, beta-quote version

Rather than a fill-in-the-blanks template, here is a complete transcript you can dissect for the mechanics. It is a beta-quote ad for a hypothetical analytics tool, cut to roughly 28 seconds, with the on-screen treatment noted. If the opening beat is where you stall, a tested scroll-stopping hook formula will earn that first second.

Voiceover and on-screen quote, 0 to 4 seconds: "An early user emailed me this last month." On screen, a styled quote card reads: "I was rebuilding the same revenue report every single Monday." Attribution under the card: "Beta user, e-commerce."

4 to 9 seconds: "Forty minutes a week, and the numbers were already stale by the afternoon." The screen shows a cluttered spreadsheet, cursor flailing.

9 to 16 seconds: "She connected her Stripe account, and the report just built itself." Screen recording: the actual connect flow, then a finished dashboard rendering.

16 to 23 seconds: "Now it posts to her team's channel at eight every morning. Her words: she hasn't opened a spreadsheet since." Quote card returns with that exact line, same attribution.

23 to 28 seconds: "If Monday reporting is eating your week, the first one is free for fourteen days." End card with the offer.

Every claim here traces to a real message and documented permission. The voiceover narrator is yours, the quotes are hers, and the demo is the product doing what it actually does. Swap the beta quote for first-person founder lines and the same transcript becomes a founder-narration ad with zero consent overhead. That interchangeability is the whole point: the arc is fixed, the source is a choice.

The avatar question, answered precisely

Generated faces are cheap and getting cheaper, which makes one distinction the difference between a presenter and a violation. An AI avatar reading your founder script is a stand-in narrator for words you own; the avatar is a delivery mechanism, like a teleprompter with a face. An AI avatar invented to portray "a satisfied customer" who never existed is a fabricated testimonial, and that is squarely what the 2024 fake-review rule prohibits. The rendering does not decide which one you made. The script does. If the words belong to a real person or to you, an avatar can voice them. If the words belong to a fictional customer, no avatar makes that legal. Knowing when AI avatars actually work and when they backfire is the practical companion to staying compliant.

Producing enough variants to learn anything

A single testimonial ad tells you almost nothing; the variance between hooks on the same product is larger than most founders expect, and the winner is frequently the one you would have bet against. Running a handful of variants per round is what gives the auction something to optimize toward, and there is genuine nuance in how many ads are worth running at once. The volume requirement is the actual reason founders are tempted to fabricate testimonials, because filming ten real ones is expensive. It is also why you do not have to: you can produce ten honest variants from the four sources above for the cost of writing ten scripts.

A production loop that scales without a film crew:

1. Write three to five scripts off the transcript above, varying only the hook and the result line so the test isolates what changes attention.
2. Render each as a captioned ad in its source format: founder narration, beta-quote-on-screen, or demo-led, instead of reshooting a person.
3. Cut each script to 9:16 for TikTok, Reels, and Shorts and to 1:1 or 16:9 for the Meta and LinkedIn feed. Same script across aspect ratios counts as distinct creative in the auction.
4. Ship the batch, let a few days of spend surface the winner, then write three fresh hooks off whatever won and repeat.

Caption everything, and treat it as data rather than dogma: as of 2026, the large majority of paid-social video impressions still play with sound off by default, so an unread quote is a wasted quote. A clean treatment is a high-contrast quote card with the speaker line and attribution legible at thumb distance, the kind of styled on-screen quote shown in the transcript above. The mechanics of burned-in captions that survive muted autoplay are worth getting right once and then reusing.

Two questions specific to this bind

Is it illegal to use an AI avatar in a testimonial ad?

It depends entirely on whose words the avatar speaks. Using an avatar to narrate your own founder script, or to read a real beta user's permitted quote, is a presentation choice and is permitted. Using an avatar to impersonate a customer who never existed is a fabricated endorsement and falls under the FTC's 2024 fake-review rule, which makes generating such reviews unlawful and carries civil penalties. The avatar is neutral. The attribution is what is judged.

Can I use beta-user quotes without their permission?

You should not, and the cost of getting permission is one email. Even when a message was sent to you directly, running it as advertising attaches the person's words to a commercial claim, and the Endorsement Guides expect that endorsements be genuine and authorized. A one-line written yes, kept on file, converts a private message into defensible ad copy. It also protects you if the user later objects, which removes the only real downside of using their voice.

If writing and rendering that batch of variants is the part slowing you down, that is the narrow job Aitachyon handles: you supply the words from your own quotes or founder script, it produces the captioned, multi-aspect cuts you test. The output stays honest because the input is yours; the tool renders, it does not invent the customer.