By Nick Quirk
September 13, 2026
9 min read
A/B testing your Google Ads means running two versions of one variable (an ad, a landing page, a bid strategy) against each other and letting the data decide which one earns more revenue per dollar spent. Done properly, it turns opinion arguments into evidence. Done badly, it burns budget and produces conclusions that fall apart the following month.
This guide covers what is worth testing, what is a waste of time, how long a test needs to run, and the part most articles skip: how to run a clean test inside an account that Google’s automation is quietly reshaping in the background.
What A/B Testing in Google Ads Actually Is
An A/B test (also called a split test) shows variation A to one slice of traffic and variation B to another, with everything else held constant. The only thing that should differ between the two versions is the single element you are measuring. Change three things at once and you learn nothing useful about any of them.
Google Ads has a native tool for this called Experiments, found under the Campaigns menu. It lets you clone a campaign, alter one setting, and split traffic 50/50 (or any ratio you choose) between the original and the variant. The platform reports both arms side by side and flags when a difference is statistically meaningful, which is more reliable than eyeballing two campaigns you built separately.
Why Bother Testing at All
Because small percentage gains compound. A campaign spending $5,000 a month at a 3% conversion rate produces a certain number of leads; push that to 3.9% through better ad copy and a faster landing page and you have added 30% more leads without adding a cent to the budget.
There is a second, less obvious benefit. Testing gives you a defensible reason for every change in the account, which matters when a client or a finance director asks why performance moved. Harvard Business Review’s refresher on A/B testing makes the point well: the discipline is less about the winner and more about learning what your audience actually responds to.
What to Test, in Rough Priority Order
Not every variable deserves the same attention. Test the things that sit closest to the money first, and leave cosmetic tweaks for when the big levers are settled.
1. Landing pages
This is usually the highest-use test and the most neglected one. Try a short form against a long form, a single-offer page against a multi-service page, or a page with a phone number in the hero against one with a booking widget. Page speed belongs in this bucket too, since a slow template quietly suppresses every ad above it. If your site runs on WordPress, the fastest WordPress themes in 2026 are worth reviewing before you blame the ads.
2. Offers and value propositions
“Free quote” versus “$99 diagnostic, credited to your job” is a genuine test of intent quality, not a wording change. Offer tests tend to produce the largest swings in conversion rate, often 20% to 60%, because they change what the visitor is being asked to do.
3. Ad copy and headline angles
Test angles, not adjectives. Price-led against speed-led, risk-reversal against social proof, question headlines against statement headlines. Our breakdown of ad copy that drives clicks covers the angles worth putting head to head.
4. Bidding strategy
Maximise Conversions against Target CPA, or Target ROAS at two different targets, are ideal Experiments because the platform can split traffic cleanly. Compare the manual and automated bidding approaches on your own data rather than on a blog’s assertion.
We cover this topic in more depth in How to Audit Your Google Ads Account in 30 Minutes.
For a closer look at this topic, see our guide: Tracking Offline Conversions and Phone Calls from Google Ads.
For a closer look at this topic, see our guide: Local Services Ads: A Guide for Home Service Businesses.
5. Audience, match type and structure
Broad match with a smart bidding signal against phrase match. A tight ad group structure against a consolidated one. Adding an observation audience layer against none. These tests take longer to read but reshape the whole account when they land.
6. Assets and extensions
Sitelink sets, callouts, image assets and promotion extensions all move click-through rate by a few points. Worth testing, but only after the four items above are settled.
The Gap Nobody Covers: Testing Inside Automated Campaigns
Classic split testing assumes you control what gets shown. In modern Google Ads you often do not. Responsive search ads shuffle headlines and descriptions per auction, and Performance Max decides channel mix on your behalf. That breaks the neat A/B model most guides still teach.
Three workarounds hold up in practice:
- Test at campaign level, not asset level. Two campaigns, identical except for the variable, with traffic split by Experiments. This is the only truly clean method left for automated formats.
- Pin strategically for copy tests. Pin one headline position across both RSAs so the tested message always appears, leaving the rest to rotate. You lose a little asset diversity and gain a readable result.
- Read asset ratings as directional only. “Best”, “Good” and “Low” asset labels are useful hints, not verdicts, because impressions are never evenly distributed. Swap out low performers, but do not call it a test.
For campaign types that resist clean splits entirely, before/after testing with a stable date range is acceptable, provided you account for seasonality. If you are running the newer formats, our guide to Performance Max campaigns explains where the reporting limits sit.
How Long Should a Test Run?
Two to four weeks is the working answer for most small and mid-size accounts, with a floor of one full conversion cycle. If your average lead takes 12 days to close, a seven-day test is measuring noise.
Sample size matters more than calendar time. As a rough guide, you need roughly 100 conversions per variation to detect a 20% difference with any confidence, and several hundred to detect a 5% difference. A campaign generating 15 conversions a month simply cannot run a valid ad copy test; it can only run offer and landing page tests where the expected swing is large.
Always run tests in whole weeks. Tuesday behaves nothing like Saturday in most industries, and a test that starts mid-week skews the comparison.
A Workable Testing Process
- Write the hypothesis first. “Adding pricing to headline 1 will raise conversion rate because it filters out budget shoppers.” If you cannot phrase it this way, you are guessing.
- Pick one primary metric. Cost per conversion or ROAS, not click-through rate. CTR is a secondary signal that often moves in the opposite direction to profit.
- Build the variant in Experiments and split traffic 50/50.
- Leave it alone. No budget changes, no keyword edits, no bid tweaks mid-flight.
- Read the result at the planned end date, not the moment one arm looks ahead.
- Apply the winner, then document it and start the next test on a different variable.
Google’s own documentation in Google Ads Help walks through the mechanics of creating and applying an experiment if you have not used the tool before.
Common Testing Mistakes
- Calling it early. A variant leading on day three wins by chance roughly half the time.
- Changing several variables at once. You get a result with no explanation attached.
- Testing trivia. Button colour rarely matters in paid search; the offer nearly always does.
- Ignoring wasted spend. A test running over dirty traffic measures the wrong thing, so clean the account with negative keywords first.
- Never retesting. A winner from 2025 may lose in 2026 as competitors and auction dynamics shift.
Free tools cover most needs here: Google Ads Experiments for the campaigns themselves, Google Analytics 4 for downstream behaviour, and any online significance calculator for checking whether a difference is real. Google Optimize was retired in 2023, so on-site page testing now runs through third-party tools or your CMS.
Frequently Asked Questions
What are common A/B testing mistakes?
The three most damaging are stopping a test before roughly 100 conversions per variation, changing more than one variable at a time, and judging winners on click-through rate instead of cost per conversion. A fourth, subtler mistake is running tests during an unusual period such as a holiday week or a major promotion, which makes the result impossible to generalise.
Should I turn on A/B testing on Facebook?
Yes, if the campaign spends enough to produce at least 50 to 100 conversions per variation, usually meaning a minimum of about $20 to $30 per day per ad set. Meta’s built-in A/B Test tool splits audiences without overlap, which is more accurate than duplicating ad sets manually, and the same discipline of one variable at a time applies.
What is the main purpose of A/B testing?
The main purpose is to isolate the effect of a single change so you can attribute a performance difference to that change rather than to chance or timing. In paid search, that usually means finding which ad, landing page or bidding setup produces cheaper conversions, then rolling the winner out account-wide.
Does Google do A/B testing?
Yes. Google runs thousands of live experiments a year on its own search results, and it also gives advertisers a native split testing tool called Experiments inside Google Ads. Google’s teaching material on the subject, including the A/B Testing course it produced with Udacity, is still a solid primer on the statistics behind it.
How many tests should I run at once?
One per campaign is the safe limit, since overlapping experiments contaminate each other’s traffic. Larger accounts can run several simultaneous tests as long as each one lives in a separate campaign with its own budget.
Want a Second Opinion on Your Testing Plan?
If you are newer to the platform, start with our walkthrough on launching your first Google Ads campaign, then come back to testing once you have a month of clean data. If you would rather have someone build and read the experiments for you, there are real advantages to working with a local agency, and you can get in touch with SEO Quirk to talk through your account.