A useful three-version healthcare AI UGC test changes one creative variable at three deliberate levels while keeping the approved audience, claim, evidence, presenter role, disclosure, offer, destination, and delivery setup as stable as possible. Write the decision rule before launch. If every version changes several things, the result can rank assets but cannot explain what to make next.
The number three is a practical design choice, not a guarantee of statistical certainty. Data volume, allocation, platform behavior, timing, and outcome quality still determine what the test can support.
What hypothesis would change production?
Choose a question that changes the next brief. Good candidates include the opening hook, order of proof, pace, on-screen wording, or call to action. Do not test a vague idea such as "authenticity."
Example hypothesis:
Identifying the visitor's question in the first sentence will lead to more qualified service-page visits than opening with the practice, without changing the approved answer.
If the result supports the hypothesis, future scripts can lead with the question. If it does not, the team knows what to retain. A hypothesis with no production consequence is reporting theatre.
Design three levels of the same variable
For an opening test, the versions might use these three deliberately different levels of the same variable:
- **Version A as the control:** "Our clinic offers an assessment for..."
- **Version B as the moderate change:** "Wondering what happens at your first assessment?"
- **Version C as the stronger change:** "Your first assessment begins with one decision..."
The exact lines must match the approved service truth. The point is that all three vary the opening frame, not the claim itself.
Do not let the third version quietly add urgency, a patient testimonial, a different presenter, or an unsupported outcome. A stronger creative expression is not permission to strengthen the healthcare claim.
Complete the test card
Use one page that gives production, review, launch, and analysis teams the same controlled specification.
**Audience decision:** What question is the viewer trying to resolve?
**Approved claim:** What exact statement can all versions make?
**Test variable:** Which single creative factor changes?
**Constants:** Presenter, visual setting, length range, disclosure, evidence, offer, destination, placement, audience configuration, optimization, and review status.
**Primary metric:** The one measured behavior closest to the hypothesis.
**Quality check:** The downstream enquiry or disposition that prevents a cheap action from being mistaken for value.
**Minimum interpretation condition:** What delivery balance, data quality, and runtime are needed before making a decision?
**Stop conditions:** Claim concern, disclosure failure, wrong audience, broken destination, material operational change, or another specified risk.
**Prewritten actions:** What the team will do if A, B, C, or no version provides a clear signal.
Google Ads experiment guidance encourages defining a clear hypothesis and success metrics and testing variables in a controlled way. Platform experiments can help with allocation, but the interface does not make a weak question or unreliable conversion event valid.
Choose a metric that matches the variable
If the test changes the opening, early engaged viewing or landing-page continuation may reveal whether the opening earned attention. If it changes the call to action, an approved contact action may be closer. If it changes proof order, watch behavior and downstream quality together.
Avoid choosing the cheapest platform metric merely because it will reach a decision faster. A hook can increase views while attracting less relevant visitors. Pair the primary metric with a quality or safety check.
Define each metric. YouTube, for example, distinguishes Shorts views from engaged views under its current measurement system. Paid platforms also update definitions. Record the version and date of the metric used.
There is no universal sample size or percentage that makes every creative test conclusive. Set interpretation requirements with the analyst who understands the platform, account volume, and business outcome.
Keep delivery comparable
Launch versions in the same experiment or tightly controlled context where possible. Use the same eligible audience, placements, geography, bid or optimization setup, schedule, and destination. Document any platform allocation differences.
Do not compare a control that ran during a stable period with a variation launched after service availability changed. Do not change the landing page halfway through without resetting the interpretation. If the platform reallocates aggressively toward an early performer, note that exposure was not even.
Retain rendered files, scripts, settings, IDs, and screenshots. "Hook two" is not enough to reconstruct what changed.
Prewrite the four possible decisions
Before launch, write what each plausible outcome means for the next production decision.
**Version A leads on the primary and quality measures.** Keep the current opening. Do not manufacture a change merely because a test was run.
**Version B leads.** Use the moderate question-led opening in the next controlled production set and verify it again across another suitable concept.
**Version C leads without quality loss.** Consider the stronger opening pattern, while preserving claim and disclosure review.
**No clear version emerges.** Keep the control or choose the simplest acceptable version, and test a different variable. Do not declare all three equivalent unless the evidence supports that conclusion.
Prewriting removes the temptation to invent a story around whichever chart looks most interesting.
A full fictional test
A dermatology practice has an approved, disclosed synthetic-presenter concept explaining how to ask about a consultation. All versions use the same presenter, setting, duration, caption, service-page destination, audience, and spend. Only the first spoken line changes from practice-led to question-led to direct decision-led.
Version B produces more measured service-page continuations than A, while qualified-enquiry reason codes remain similar. Version C earns more starts but more wrong-service enquiries. The next decision is not simply "B won." It is to retain the question-led opening and avoid the stronger wording that appears to broaden audience interpretation. These are fictional outcomes used to show the logic, not performance expectations.
If the qualified-enquiry data is incomplete, the team can make only the narrower creative observation and should not claim lead-quality superiority.
Close the test with a learning record
Record the hypothesis, exact versions, delivery context, metric definitions, exclusions, outcome, uncertainty, and next production instruction. Link the claim and disclosure approvals to each render.
Healthcare advertising teams need fewer vague "winning creatives" and more decisions that survive the next briefing conversation. A three-version test earns its cost when everyone can state what changed, what stayed fixed, what the evidence supports, and what to make next.