On this page
Choose the decision you want to make, record the differences and define success before looking at the results.
Write a question narrow enough to answer
“Which ad is best?” hides too many differences. “Does a question opening work better than a result-first opening for this demonstration?” gives you a clearer comparison. If you want to isolate the caption, keep the footage and timing fixed.
You can also compare complete treatments, but name that honestly. If the face, words, music and pacing all change, the result describes the whole edit, not a single hook phrase.
Keep the surrounding conditions comparable
Use the same product action, offer and destination where possible. Record audience, placement, schedule and budget differences rather than treating them as invisible. Where an advertising platform offers a suitable experiment tool, consult its current documentation and use the configuration that matches your question.
Organic posts are usually an observational comparison because delivery conditions differ. Posting two videos on different days can suggest what to explore, but it does not isolate the opening by itself.
| Keep stable where possible | Record if it changes |
|---|---|
| Product demonstration and offer | Any feature, price or destination change |
| Closing instruction | A different conversion action |
| Caption style and audio | Extra visual or sound differences |
| Audience and placement | Different delivery context |
| Measurement definition | Different denominators or reporting windows |
Separate attention from the business result
An opening can hold attention while attracting people who will never use the product. Choose a primary metric related to the decision and keep supporting metrics separate. Use the same definition and denominator for each version.
For example, an early-view rate can describe attention only if both versions use the same platform definition. Click-through rate should use a consistent click type and impression denominator. Purchase conversion requires a consistent attribution window and a reliable record of the purchase. Do not compare unrelated dashboard numbers as though they were one funnel.
Use a small experiment record
Fill this in before running the comparison. Include the decision you will make if the evidence is inconclusive, such as keeping the baseline and collecting more data.
| Field | Example entry |
|---|---|
| Question | Question opening versus result-first opening |
| Versions | A and B, saved export filenames |
| Primary outcome | One consistently defined metric |
| Guardrail | No misleading claim or broken destination |
| Delivery notes | Audience, placement and time window |
| Decision rule | Set a planned review point; do not chase hourly noise |
| Limitations | Uneven delivery, small sample or other changes |
Review the result without inventing certainty
Inspect the exports again before interpreting numbers. A missing caption or playback issue can make the intended comparison invalid. Check whether delivery was sufficiently comparable and whether there is enough evidence for the decision.
There is no universal sample count that proves every hook test. Required data depends on the baseline, effect size and experiment design. If the result is noisy, label it inconclusive. Document the next question rather than turning a small difference into a rule for all future videos.
Questions & answers
Should I choose the video with the most views?
Only if views are the relevant outcome and the delivery is comparable. For a purchase goal, inspect the rest of the path too.
What if the result is tied?
Keep the uncertainty explicit. You can retain the simpler baseline, collect more suitable data or test a more meaningful difference.
Find the opening for your next edit.
See the public samples, then unlock the library when you’re ready to work with the originals.
Explore the library