Most ad accounts change when someone finds time. A good one changes every week on purpose, with a small number of new ideas, a fair test and a short written verdict. The process is simple. The hard part is being honest with yourself about the results.
Here’s the loop we use, step by step. The examples use Meta, but the loop is the same on TikTok and Google.
Benchmark context: Motion’s “winner” definition measures spend allocation relative to an account’s median, with a minimum spend floor. It does not establish profit, incremental sales or the results a new client will achieve. Use the findings to choose tests, then judge your account against its own economics.
Step 1: Decide which kind of test this is
There are two kinds of creative test, and mixing them is the most common way to learn nothing.
| Concept test | Iteration test | |
|---|---|---|
| The question | What should we say, and to whom? | What’s the best way to say it? |
| What changes | Person, reason to buy, format, visual world | Hook, first three seconds, headline, length, CTA |
| When to run it | Every week | Only on a concept that has already shown signal |
| The trap | Iterating on a weak idea because it was expensive to make | Shipping versions so similar that Meta treats them as one ad |
Meta makes the same distinction. It describes iteration as two ads “with identical visuals, but different text CTAs” and warns that it “should not replace true creative diversification” (Meta, Demystifying Creative Diversification).
If a concept loses, you want to know whether the idea lost or the execution lost. Keep the two tests apart and you’ll know.
Step 2: Write the brief and the pass mark before launch
Every ad gets one sentence of hypothesis and one pass mark, agreed before it goes live. For example:
- Hypothesis: New parents will respond to “sleep through the night” more than to the cooling fabric.
- Variable: Person and reason to buy. Format stays UGC-style video.
- Pass mark: Cost per purchase at or below target after a fair spend.
Writing the pass mark first stops the most human mistake in testing: deciding what success meant after seeing the numbers.
Step 3: Choose a structure
There are three common ways to run creative tests on Meta. None is universally right.
- Meta’s built-in creative test. Inside an existing campaign, you can test 2 to 7 copies of an ad with a share of budget. Meta suggests no more than 20% of the existing budget, the campaign must use the Highest volume bid strategy, and results appear in Experiments. Meta notes that “a confidence level is not included” (Meta, Set up a creative test). Learnings stay in the campaign, and winners keep running.
- A separate testing campaign. One concept per ad set, equal budgets, winners moved into the main scaling campaign afterwards. Many media buyers prefer this because each concept gets a fair share of spend.
- No separate test at all. New ads launch straight into the live campaigns and are judged daily against the target. This suits high-volume accounts with strong cost controls.
Pick one and stay consistent for a few weeks, so results are comparable.
Step 4: Use spend as the clock, not days or impressions
An ad that has spent a tenth of your target cost per purchase hasn’t had a chance to prove anything. Judge each test against how much it has spent relative to your target, not how long it has run or how many impressions it got. Impressions are a poor clock because CPMs vary: two thousand impressions at a cheap CPM and two thousand at an expensive one are very different tests.
A practical rhythm many buyers use:
- Around one times target CPA spent with no conversions: watch it. Look at hook rate and click-through rate to see whether it’s earning attention.
- Two to three times target CPA spent with no conversions and weak early signals: switch it off.
- At or below target over several days with a handful of conversions: promote it, and brief iterations.
Tune these to your account. The principle matters more than the multiples.
Avoid significant edits while a test runs. Meta’s learning phase usually ends “after about 50 results in the week after the ad set’s last significant edit,” and edits reset it (Meta, About the learning phase).
Step 5: Read the result the same way every time
Use the same order every week:
- Did it stop the scroll? Hook rate on video is 3-second plays divided by impressions (Motion).
- Did it hold attention? How much of the video did people watch?
- Did it get the click? Outbound click-through rate.
- Did it make the sale? Cost per result against target.
The first three explain the result. Only the last one decides it. An ad with a beautiful hook rate and no sales still lost.
Step 6: Expect most ads to lose
This is normal, and it’s the reason for the weekly rhythm. In Motion’s 2026 benchmark across 578,750 Meta creatives, about 5% of ads became winners by its definition (at least 10 times the account’s median spend, with a $500 floor). That’s roughly one in twenty. Motion’s own planning heuristic: “if you’re testing 20 ads per month, expect ~1-2 winners,” and if you test five a month, “expect to go multiple months between winners” (Motion, 10x benchmark).
So make genuinely different ads, steadily, and give winners a chance to appear. Volume on its own won’t do it.
Step 7: Write it down
Finish every week with a one-screen read-out. Ours looks like this:
| Idea | Format | Verdict | Why |
|---|---|---|---|
| A. New-parent angle | UGC-style video | Scaling | Beat the pass mark, strong hold |
| B. Us vs the old way | Static | Holding | Strong clicks, people drop off on the product page |
| C. Founder story | Founder-led video | Retired | Low hook rate after a fair spend |
Then two lines: what we learned, and what next week’s run will test. Add a third line when it applies: anything that isn’t a creative problem, such as a slow product page or a checkout issue.
Over a few months, this log becomes the most valuable asset in the account. It tells you which people, reasons and formats work for your brand, so each week’s ideas start from evidence.
The same loop on TikTok and Google
Only the tools change. TikTok Ads Manager has its own split test, and Google Ads has experiments, with asset groups in Performance Max and Demand Gen doing the job of ad sets. The rules stay the same: one variable per test, the pass mark written down before launch, spend as the clock, and a short written verdict.
The weekly rhythm
- Monday: read last week’s results, write the read-out, brief the new concepts
- Tuesday to Wednesday: produce the ads
- Thursday: launch, without touching last week’s tests
- Following week: read on spend, decide, repeat
Checklist
- Label every test as a concept test or an iteration test
- One-line hypothesis and pass mark written before launch
- One structure, used consistently
- Spend relative to target CPA as the clock
- No significant edits mid-test
- Same reading order: stop, hold, click, sale
- One-screen read-out every week, including non-creative issues
Sources
- Meta for Business, Demystifying Creative Diversification
- Meta Business Help Center, Set up a creative test in Meta Ads Manager
- Meta Business Help Center, About the learning phase
- Motion, What counts as a winning ad: the 10x benchmark
- Motion, Thumbstop rate / hook rate