Ad Testing
How to Test Your Ads When You Can't Run an A/B Test
Change one thing, then compare it fairly. Use matched before-and-after weeks, take turns by week, pause ads in a few similar areas, or send each version to its own tracked link. Judge every version by the same GA4 key events, run whole weeks, and write the plan down before you start.
- Built-in experiment tools have rules many small accounts can't meet.
- Whole weeks beat a few days, because weekdays and weekends behave differently.
- A UTM tag on each version's link lets GA4 tell them apart.
- A small result is a hint, not proof, so say so when you report it.
On this page
- Why the Built-In Tools Might Not Work for You
- The One Rule: Change One Thing
- Option 1: Before and After, With Matched Weeks
- Option 2: Take Turns by Week
- Option 3: Pause Ads in a Few Similar Areas
- Option 4: Two Landing Pages, Two Tracked Links
- Make GA4 Key Events Your Scorecard
- How Long to Run It
- Write It Down Before You Start
- Frequently Asked Questions
- Sources
Ad platforms make split testing sound like one click. Then you find out your campaign doesn't qualify, or your budget is too small to learn anything. You can still test. You just have to set it up more carefully yourself.
Why the Built-In Tools Might Not Work for You
If your platform's own testing tool works for your campaign, use it. It splits people for you, which is hard to copy by hand. But each tool has rules.
Google Ads custom experiments. Google's help pages list Search, Display, Video and a few other campaign types. Shopping and App campaigns can't use standard custom experiments. To qualify, Google says a campaign must:
- Be active, not paused
- Have its own daily budget, not a shared budget (one budget split across several campaigns)
- Use a standard bid strategy, not a portfolio bid strategy (one bid setting shared across campaigns)
Google also says you can only run one experiment at a time. A 50/50 split doesn't promise equal impressions or spend, and the trial side can take 7 to 14 days to settle. For reliable results, Google points to minimums "such as more than 100 daily conversions." Most small accounts are nowhere near that.
Meta A/B tests. Meta's tool shows each version to a separate group, so nobody sees both. Meta says a test must run 1 to 30 days and recommends at least 7. Your audience should be large enough and not used by another campaign at the same time, and your budget needs to produce enough results to pick a winner.
One honest note: Meta says it doesn't recommend testing by turning ad sets on and off by hand, because audiences can overlap and delivery gets less efficient. The methods below are for when the real tool isn't an option, and they come with caveats you should say out loud.
The One Rule: Change One Thing
If you change the headline, the image and the landing page in the same week, a better result tells you nothing about which change helped.
Pick the single change you care most about. Leave everything else alone, including budget, targeting and bidding. If you're not sure which change to test first, this guide to picking your first test helps you choose.
Option 1: Before and After, With Matched Weeks
The simplest test: run the old version, then the new one, and compare.
- Pick your stretchChoose whole weeks, like four weeks before and four weeks after. Equal lengths, same days of the week.
- Write down the "before" numberNote your key events (the actions you count as wins, like form sends) for the old version.
- Make the changeSwap in the new ad or setting on a Monday, and note the date.
- Skip the settling daysLeave out the first few days while the platform adjusts.
- Compare the same stretch afterLook at the same number for the same number of weeks.
The weak spot is time. Anything else that changed during those weeks can move the numbers too: a holiday, a competitor's sale, a Google update. Check the same weeks last year if you have them. There's more on this in how to tell if a change actually worked.
Option 2: Take Turns by Week
Instead of one long "before" and one long "after," switch back and forth. Run version A one week, version B the next, then A, then B.
This spreads out the bad luck. One slow holiday week won't sink a whole version.
- Switch on the same day and time each week.
- Run at least two weeks of each version.
- Change nothing else between weeks.
You can also use this as an on/off test. Run your ads one week and pause them the next. That shows how many leads come in without ads, which tells you what the ads add. Keep in mind Meta's warning above. Pausing and restarting can disrupt delivery, so expect the first day or two of each "on" week to look a bit off.
Option 3: Pause Ads in a Few Similar Areas
If you serve several cities, counties or regions, you can split by place instead of by time. This is often called a holdout (a group you deliberately leave out so you can compare).
- List the areas you serve and pick pairs that are alike in size and past results.
- Keep ads running in one area from each pair. Pause them in the other.
- Run both for the same weeks.
- Compare leads per area, not total leads.
Most ad platforms let you choose locations in each campaign's settings, so this needs no special tools. It won't work if you only serve one town. Use Option 1 or 2 instead.
Option 4: Two Landing Pages, Two Tracked Links
This one is for testing where the ad sends people, not the ad itself. Make two versions of the page. Then give each one its own tracked link.
A UTM tag is a short label added to the end of a link that tells GA4 where a click came from. Google says to always include three: utm_source, utm_medium and utm_campaign. Add utm_content to tell two versions apart. For example:
yoursite.com/quote?utm_source=facebook&utm_medium=paid_social&utm_campaign=spring_quote&utm_content=page_a
The second link is the same, except it ends in page_b and points to the other page.
You can build these with Google's free Campaign URL Builder. Google notes that values are case sensitive, so "Facebook" and "facebook" show up as two different sources. Pick lowercase and stick with it.
Then rotate the links by week (like Option 2), or run two otherwise identical ads at once, one per link. If they run at once, the platform will likely favor one ad. That's fine. Compare the rate (key events per visit), not the totals.
Make GA4 Key Events Your Scorecard
No GA4 on your site, or no key events yet? Start with Conversion Tracking for Beginners. Without a reliable count of results, no test can tell you anything.
Each ad platform counts results its own way. Using GA4 as the referee gives every version the same scorecard.
- Make sure your wins are marked as key events in GA4. Google defines a key event as an action that's especially important to your business.
- Open Reports > Acquisition > Traffic acquisition.
- Switch the table's main dimension to Session campaign, then add Session manual ad content as a secondary dimension (click the + next to the first column). Each
utm_contentvalue, like page_a and page_b, gets its own row. You can also compare by Landing page. - Note the sessions and key events for each version.
Also count what GA4 can't see. Phone calls and emails often never reach GA4. A simple tally of leads by week keeps you honest. If your numbers look strange, run a quick GA4 sanity check first.
How Long to Run It
Longer than feels comfortable. A few rules of thumb:
- Whole weeks only. Weekdays and weekends behave differently, so always compare full weeks.
- At least two weeks per version. One week is one week's luck.
- Skip the settling days. Google says a Smart Bidding learning period can last up to 7 days after big bidding changes, like a new bid strategy or new conversion goals. Even after a simple ad change, skip the first few days.
- Watch the count, not just the rate. If each version has five leads, a gap of two is noise. Keep going, or call it a hint.
If your sales take a while, run longer still. Meta's own advice is to extend a test when customers usually take more than a week to act.
Write It Down Before You Start
This is the step people skip, and the one that keeps you honest. Before you change anything, write a short plan:
| What to write | Example |
|---|---|
| The one change | New headline: "Same-day quotes" |
| The method | Take turns by week, 3 weeks each |
| The scorecard | Quote form sends (a GA4 key event) per visit |
| Start and end dates | Six full weeks, Monday to Sunday |
| What counts as a win | Clearly more form sends per visit, in most weeks |
| What else is happening | Holiday slowdown in week 1 |
Here's a made-up example of a clear write-up.
Vague
"The new headline worked way better."
Clear
"Over three weeks each, the new headline got 14 quote requests from 410 visits. The old one got 9 from 395. It won in two of three weeks. It's a small sample, so we'll keep it and check again next month."
Keep the notes somewhere you'll find them in six months.
Where Moonlit Tools Fits
Moonlit Tools' testing plans use the same idea: one change, a saved baseline, a success measure and check-ins at 30, 60 and 90 days, then a recorded decision. Moonlit Tools doesn't run ads or split traffic. Proving your work with Moonlit Tools.
Frequently Asked Questions
Is a before-and-after test good enough?
It's good enough to learn from, as long as you compare matched weeks and change one thing. It can't prove the change caused the result, because other things change over time too. Say "this lines up with" rather than "this caused."
Can I test two ads at once without the A/B tool?
You can run two ads side by side, but the platform will usually favor one. The people who see each ad also won't be split evenly. Compare rates through UTM links and treat it as a rough read.
What if I only get a few leads a month?
Test bigger changes and run longer. A small tweak won't show up in a handful of leads. A new offer or a very different landing page has a better chance.
Sources
- About custom experiments, Google Ads Help
- Set up a custom experiment, Google Ads Help
- About A/B testing, Meta Business Help Center
- What are best practices for A/B tests for Meta ads?, Meta Business Help Center
- About Smart Bidding learning, Google Ads Help
- URL builders: Collect campaign data with custom URLs, Google Analytics Help
- Conversions vs. key events in Google Analytics, Google Analytics Help
See What Google and AI Tools Can Read
Moonlit Tools reads your site the way search engines and AI assistants do, then tells you what to fix first and shows you why. Scan your own website free. No card needed.
Scan My Site Free Or apply for 90 days of Launch free