The short version
A/B testing paid ads, fast
- Run two versions, change one thing, and let the data pick the winner. That’s it.
- Test one variable at a time, same budget, same audience, same time, or your data means nothing.
- Give it 3–7 days. Don’t call a winner after 24 hours.
- Optimise for the metric that pays you (CPA, ROAS or CPL), not just clicks.
- Test by funnel stage: stopping power at the top, trust in the middle, the offer at the bottom.
A/B testing in paid ads is how you figure out what actually works, instead of guessing. You run two versions of an ad, change one thing, and see which one performs better. That’s it. It’s simple, it’s essential, and if you’re not doing it, you’re wasting money.
Because even the best-looking ad is just a theory until it’s proven in the wild. A/B testing turns theory into results. Let’s break down exactly how it works, what to test and when, and how to turn test results into revenue.
What A/B testing actually is (and isn’t)
You take a single ad and make two or more versions that are identical except for one variable. Then you let them battle it out under the same conditions to see which drives more clicks, conversions or revenue. This isn’t throwing spaghetti at the wall, it’s controlled experimentation that tells you, definitively, what your audience prefers.
Without it, you’re just hoping your gut is right, and hope is not a strategy. Even tiny tweaks, a different CTA, headline or image, can lift your click-through rate and lower your cost-per-acquisition. That’s how you make your budget work harder, not bigger. If you’re watching ROAS but not testing consistently, you’re leaving easy wins on the table. Here’s how to know your real numbers with the ROAS calculator if you haven’t already.
What to test, by funnel stage
Not all tests are equal. What you test should depend on where someone is in your funnel, because what works on a stranger is very different from what works on someone deciding whether to buy today.
| Stage | Test this | The goal |
|---|---|---|
| Top (TOFU) | Hooks, video vs image, brand vs problem messaging, emojis | Stop the scroll, cheap qualified clicks |
| Middle (MOFU) | CTA tone, social proof type, content offer, copy length | Build desire and trust, more leads |
| Bottom (BOFU) | Offer format, urgency framing, CTA button, whole landing pages | Close the sale, cut CPA, lift ROAS |
Top of funnel: grab attention fast
At this stage you’re interrupting people mid-scroll. They don’t know you yet, so test for stopping power: hook variations (a curious question vs a bold statement), video vs static image, static vs animated carousel, brand-focused vs problem-focused messaging, and whether emojis help or hurt in your industry. The goal is maximum scroll-stopping power and cheap, qualified clicks. For the creative side of this, our guides to video creatives that convert and the best hooks for ad creatives are the place to start.
Middle of funnel: build desire and trust
Here, people know who you are. Now convince them you’re the solution, so test for resonance and proof: CTA tone (“Learn more” vs “Get the guide”), social proof type (a testimonial vs a logo bar), content offer (guide vs quiz vs checklist), and copy length (short and punchy vs story-driven). Copy length alone can meaningfully move your cost per lead when you get it right. The goal is more engagement, downloads and leads by sharpening the message.
Bottom of funnel: drive the sale
This is where the money is made, so your tests should push people over the line: offer format (percentage off vs free shipping vs limited-time), urgency framing (a countdown timer vs “only 3 left”), CTA button colour and placement (sounds trivial until a small change lifts conversions), and entire landing-page variations, headline, layout, testimonials, mobile UX. Landing-page tests often drive the biggest revenue jumps. The goal: get the sale, cut CPA, lift ROAS.
How to run a profitable A/B test
1. Test one variable at a time
Change one element. That’s it. Test a new headline and a new image and a new CTA together and you’ll have no idea what actually moved the needle.
2. Keep budgets and audiences consistent
Same budget for each variation, same audience, running at the same time. You need a clean experiment, or the data means nothing. If your budget itself is the question, here’s how to set the right budget for your campaigns.
3. Run it long enough to get real data
Don’t declare a winner after 24 hours. Most tests need at least three to seven days to reach anything close to statistical significance, depending on your spend and audience size.
4. Choose the right metric
Don’t judge on CTR alone. If your goal is purchases, optimise for CPA or ROAS. If you’re building an email list, cost per lead is your number. Optimise for the metric that actually pays you.
5. Pick a winner, then scale it
When a test ends, take the winner and build on it. Make new variations based on what worked. Testing isn’t one-and-done, it’s a feedback loop that keeps compounding.
Platform-specific A/B testing tips
Meta ads (Facebook and Instagram). Use the A/B Test tool inside Ads Manager, it splits traffic 50/50 automatically and it’s great for creative, audience and placement tests. Deep dive: our Meta ads service and approach.
Google Ads. Use ad variations for search campaigns to test headlines, descriptions and display paths, and prioritise conversion data over CTR. More on how we run Google Ads.
TikTok ads. Creative is king, so test video hooks and visual pacing, use Spark Ads for a native feel, and yes, test different music and captions, it genuinely moves the numbers.
And always test your retargeting too. Different audiences respond to different messaging, and your retargeting strategy deserves its own experiments.
The accounts that win aren’t the ones with the cleverest single ad, they’re the ones testing constantly and actually writing down what they learn. Half the “genius” creative you admire is just the survivor of fifty quiet tests nobody saw. Build the habit and your account gets smarter every month.
Common A/B testing fails (don’t be that guy)
- Testing too many things at once — no useful data.
- Shutting tests down too early — false positives.
- Testing without enough traffic — inconclusive results.
- Not documenting results — no learning, just guessing again later.
- Relying only on CTR — misleading if the conversions aren’t happening.
If your ads pull plenty of clicks but no sales, this one’s for you.
Final thought
A/B testing isn’t about being “data-driven”, it’s about being profit-driven. It’s how you stop wasting budget, sharpen your best ideas, and keep your campaigns evolving as your audience does. You wouldn’t launch a product without market research, so don’t launch a campaign without testing.
Paid ads, tested and tracked
No time to run tests and a business?
We run the experiments, decode the results and scale what works, so your budget compounds instead of leaks. Book a call and we’ll find the wins hiding in your account.
A/B testing FAQs
What is A/B testing in paid ads?
It’s running two versions of an ad that are identical except for one element, under the same conditions, to see which performs better. It replaces guesswork with evidence about what your audience actually responds to.
How long should you run an A/B test?
Usually three to seven days, longer if your spend or audience is small. Calling a winner after 24 hours is how you get false positives and scale the wrong ad.
What should you A/B test in ads?
Test by funnel stage: hooks, formats and messaging at the top; CTAs, social proof and copy length in the middle; offers, urgency and landing pages at the bottom. Change one variable at a time so you know what moved the result.
How many things should you test at once?
One. Change a single variable per test. If you swap the headline, image and CTA together and results improve, you have no idea which change did it, so you can’t repeat it.
What metric should you optimise for?
The one tied to your goal, not just clicks. For sales, optimise for CPA or ROAS; for leads, cost per lead. CTR is useful early, but it’s misleading if the conversions aren’t following.