How do you set up an A/B test on Google Ads?

The three testing tools (experiments, ad variations, page tests), what is worth testing, the volume and duration required, and how to read the results without getting it wrong

The essentials

  • Three tools: campaign experiments (splitting traffic between the original and a variant: bidding, targeting, structure), ad variations (text and assets), and landing page tests (by final URL or external tool).
  • What to test first: bidding strategy and target, landing pages, hooks and pinned assets, match types, audiences.
  • Volume and duration: at least 100 conversions per arm or 4 full weeks; below that, no reliable conclusion.
  • How to read it: on the objective metric (cost per conversion, ROAS), with the confidence interval displayed; one test at a time per campaign.

On Google Ads you can test almost anything: a bidding strategy, a landing page, a headline, a match type, an audience. But testing badly (two changes at once, one week, fifty clicks) produces false decisions. This article presents the platform's three testing tools, what is worth testing, the volume and duration required, and how to read the results. Tests on social networks follow their own logic; see Meta Ads and LinkedIn Ads mistakes.

The three testing tools

ToolWhat it testsHowLimits
Campaign experiments (Custom experiments)Bidding strategy, target CPA or ROAS, budget, targeting, match types, structure, landing pagesGoogle creates a copy of the campaign, you edit the copy, traffic is split (50/50 recommended) for a set duration; the results appear with a confidence intervalSearch and Shopping; Performance Max has its own limited experiments (PMax uplift test, final URL test)
Ad variationsText: replace, add or edit headlines, descriptions, paths and URLs, across several campaigns at onceFind and replace or asset update; traffic split; results per variationText only; responsive ads already test their own combinations
Landing page testsTwo pages for the same groupCampaign experiment with a different final URL, or an external tool that splits traffic on the page (site-side A/B test)Ads traffic alone may not be enough; combine it with SEO traffic in a site-side testing tool

Tests on the Google Business Profile listing follow a different logic, with no native tool; see A/B testing on Google Business Profile.

What is worth testing, by order of gain

  1. The bidding strategy and its target: moving from "maximise conversions" to a target CPA, or changing the target by 20%; this is the test that shifts cost per conversion the most.
  2. The landing page: a dedicated page against the home page or a category page; a short form against a long one; the effect on conversion rate is often 20 to 50%.
  3. The hooks: benefit against problem, price shown or not, proof (reviews, a figure) in the headline; pinned against free assets.
  4. Match types: broad match with Smart Bidding against phrase or exact match, on a mature campaign.
  5. Audiences: bid adjustments or targeting on in-market and remarketing audiences.
  6. Scheduling and locations: extended delivery at the weekend, extension to a neighbouring town.

Do not test what the platform already tests: the order of headlines in a responsive ad, ad rotation under automated bidding. The assets to optimise are detailed in improving the Quality Score.

Creating a campaign experiment

  1. In Google Ads: Campaigns, Experiments, "New experiment", "Custom experiment".
  2. Choose the base campaign and name the experiment after the hypothesis ("target CPA €80 against €100").
  3. Change only the element being tested in the copy.
  4. Set the traffic split at 50% and the duration at 4 weeks minimum (6 for long cycles); choose the split by search (the default) or by cookie if the test concerns the page.
  5. Define the objectives: the decision metric (cost per conversion or ROAS) and a control metric (conversion rate, cost).
  6. Launch, change nothing for the duration, read the results with the confidence interval, then "Apply" (merge into the original) or "End".

Volume, duration and reading the results

RuleWhy
One change per testOtherwise the effect cannot be attributed
At least 100 conversions per arm, or 4 full weeksBelow that, chance dominates; Google displays "not enough data"
Full weeksSmooths out variations from Monday to Sunday
No changes during the testEvery change restarts learning and distorts the comparison
Decide on the objective metricA better click-through rate with a degraded cost per conversion is a loss
Require a confidence interval that excludes zero (asterisk in the report)Without significance, a 10% gap is noise
One test at a time per campaignTwo simultaneous experiments contaminate each other
Record the contextPromotion, season, competitor: to be logged in order to interpret the result

Ad tests

Responsive ads automatically assemble headlines and descriptions and favour the combinations that perform; the "Assets" report shows the performance of each headline (low, good, best). To test a precise hypothesis (a price in the headline, a hook), pin the asset in position 1 in one variation and compare it with the "Ad variations" tool or with two ads in the same ad group over 4 weeks. Then replace the "low" assets with new ones; this is a monthly cycle. The visuals of Display, Performance Max and video campaigns are tested by asset groups; see high-performing visuals and videos for ads.

The frequent mistakes

  • Concluding after a week or 30 clicks.
  • Testing two things at once (bidding and page) and attributing the effect to one of them.
  • Comparing a campaign over one month with the previous month: seasonality and competition change; use a split-traffic experiment.
  • Judging on click-through rate rather than on cost per conversion.
  • Applying the winning variant without checking significance.
  • Not testing at all because "it works": CPCs rise every year; what works today will cost more tomorrow.
Our advice: launch a single experiment this month: your main campaign with a target CPA 20% lower, on 50% of the traffic, over four weeks. This is the test that most often answers the question that matters, "can I pay less for the same volume?", and it requires neither creative nor a page.

How GreenRed helps

Rather than juggling several tools, GreenRed's SEA tracking and optimisation brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.

Frequently asked questions

How long should an A/B test on Google Ads run?

Four full weeks at least, six for long purchase cycles, or until 100 conversions per arm. Full weeks smooth out weekly variations; below that volume, Google displays "not enough data" and any conclusion is down to chance.

Can you test Performance Max?

Partially: Performance Max experiments let you test adding a PMax campaign alongside existing campaigns (uplift) and compare final URLs. Fine-grained bidding and targeting tests are not available as they are in Search; you then compare periods carefully, or asset groups.

Should you test responsive ads?

Responsive ads already test their combinations of headlines and descriptions; the "Assets" report shows which ones perform. Test precise hypotheses by pinning an asset in a variation, and replace the assets rated "low" every month. Two to three ads per ad group are enough.

How do you test a landing page with Google Ads?

Create a campaign experiment with a different final URL on the copy, at 50% of the traffic, split by cookie so that the same user always sees the same page. If Ads traffic is not enough, use a site-side A/B testing tool that splits all traffic (Ads and SEO) between the two versions.

Optimise your Google Ads campaigns

GreenRed connects your ad spend to the revenue generated so you can manage your ROI channel by channel.

Manage my campaigns

Related articles