Testing ad creatives with small variations

A test is two ads that differ in one way, shown under the same conditions, judged on one number you picked in advance. Most "tests" are not that. They are two ads that differ in everything, shown at different times, judged on whichever number looks best afterward. The second kind feels productive and teaches nothing. This article is about the first kind, and about how to run it when making a version is cheap enough to do it often.

Why small variations

If version A has a different hook, a different presenter and a different length than version B, and B wins, what did you learn? That B is better. You did not learn why, so you cannot carry the lesson to the next product. Change one thing and the result is a lesson you can reuse: this hook beats that hook, for this product, with this audience.

Small variations also keep the test honest. When the two versions share everything but the hook, you are not tempted to explain the result with the thing you happened to like.

What to vary, in order

Not every change is worth testing. Vary the things that move the result most, and vary them in this order.

1. The hook

The first sentence decides whether the rest is heard. It is the single highest-leverage change, and it is also the cheapest: the words change, nothing else does. With Clipelle, "Review script first" pauses after the script is written and shows three hooks; you pick one and the rest of the script stays the same. Make one project per hook with everything else identical. For how to write hooks that are worth testing, see hooks for short product ads.

2. The offer

With an offer line or without one, or two different offers. The Offer field is at most 80 characters and it is the only way a deal gets into the spoken call to action, so the two versions will differ in exactly one sentence at the end.

3. The length

Clipelle makes 10, 20, 30, 45 or 60-second ads. A 20-second version of a winning 30-second script is a legitimate test: fewer words, same message. Keep the hook fixed when you test length, or you will not know which change did the work.

4. The tone

Elegant, playful, direct or warm. The tone changes the wording of the whole script, so it is a bigger change than a hook, but still one variable.

5. The presenter and the voice

A different presenter look, a different voice, a different accent or speaking speed. Test this after the message is settled. A better presenter reading the wrong message is still the wrong message.

6. The direction

The Direction field (up to 200 characters) tells the script who the ad is for and what to emphasize. "For parents buying a first school bag" versus "for commuters who cycle" produces two scripts with different problems and benefits. This is a bigger test than a hook test; it is really a test of audience.

7. The look

Caption style, accent color, font, logo. Test these last, if at all. They matter for consistency across your ads more than for any single result.

What to keep fixed

Within a test pair, keep identical: the product photo, the product name and description, the language, the presenter quality (Standard unless you have a reason), the format and, unless length is the variable, the length. In Clipelle you get this for free by duplicating your choices and changing one field. Give the projects names that say what changed: "Bottle, hook A, 30 s" and "Bottle, hook B, 30 s."

A test plan you can run this week

Say you sell an insulated water bottle. Here is a plan with three rounds.

Round one: three hooks. Three 30-second Standard ads, same everything, different first sentence:

• "Your coffee is cold by the second meeting."
• "This bottle has outlived three phones."
• "Stop buying a new bottle every summer."

Three 30-second Standard ads are 3 × 30 = 90 credits. Run them under the same conditions and keep the one that does best on your chosen number.

Round two: with and without an offer. Take the winning hook. Make one version with the Offer line "Free shipping on your first order" and one without. Two more ads, 60 credits.

Round three: length. Take the winner of round two and make a 20-second version. One more ad, 20 credits. If the shorter one holds up, you have a cheaper ad to run from then on.

Six ads, 170 credits, three lessons you can write down: which hook, whether the offer helps, whether 20 seconds is enough. The project page shows the real amount spent on each one next to its estimate, so your log can carry exact costs. For how those numbers fit into the whole bill, read what a video ad costs in time and money.

How to judge a pair

Before you run anything, write down one metric and a minimum amount of exposure. The metric should be as close to the sale as you can measure reliably: clicks to the product page, add-to-carts, orders. The exposure should be enough that a handful of random clicks cannot flip the result. How much is enough depends on your numbers, so decide it from your own history, not from a rule of thumb you read somewhere.

Then follow three rules:

1. Run both versions at the same time, with the same budget and the same audience. A version shown on a weekday against a version shown at the weekend is not a test.
2. Judge only on the metric you wrote down. If you look at five numbers after the fact, one of them will favor the version you liked.
3. Treat a small difference as a tie. A tie is a result too: it tells you this variable does not matter much for this product, so move on to the next one.

Keep a log

A test you cannot look up is a test you will repeat. Keep a plain table:

Project	What changed	Hook or value	Credits	Result
Bottle, hook A, 30 s	Hook	"Your coffee is cold by the second meeting."	30	...
Bottle, hook B, 30 s	Hook	"This bottle has outlived three phones."	30	...

Fill in the result with the metric you chose and nothing else. After a few products, patterns show up: maybe problem-first hooks win for you, maybe offers only help on higher prices. Those patterns are the real return on testing.

Mistakes that waste the test

• Changing two things. The most common mistake, and the easiest to avoid when a new version is one changed field.
• Comparing across lengths to judge a hook. A 10-second ad and a 60-second ad differ in far more than the first sentence.
• Testing the presenter first. Settle the message, then the messenger.
• Calling a winner early. The first day is noise. Wait for the exposure you wrote down.
• Testing a product that has not earned it. If the first ad gets no response at all, the next test is a different problem statement or a different audience in the Direction field, not a caption color.

When the product itself is the test

For dropshipping, the question is often not "which hook" but "does anyone want this at all." The discipline is the same: one variable, same conditions, one metric, a decision rule written down in advance. The dropshipping page explains how Clipelle fits into that, and test a dropshipping product with a video before you order stock walks through the whole sequence.