Google Ads experiments & A/B testing

Stop guessing. Test Google Ads changes properly.

Magnify combines diagnosis with its own ecommerce experiment engine: SKU-level title tests plus concurrent matched-pair randomized catalogue holdouts that verify what Google actually published before calling a result valid. Start with the free 92-point Google Ads audit.

Run my free 92-Point Google Ads Audit →See what the audit checks

✓ Read-only first analysis✓ Nothing changes without approval✓ 92 checks · 23 audit sections

Why this matters

The biggest A/B testing mistake happens before the experiment starts.

Advertisers often test a bidding strategy, campaign structure or PMax setting when the real problem is broken measurement, weak product data, wasted search demand or insufficient volume. A controlled experiment cannot rescue a bad hypothesis. Magnify starts by diagnosing the account so the test is aimed at the right uncertainty.

What you get from this workflow

Turn audit findings into specific, falsifiable experiment hypotheses.
Use native Google Ads Experiments for supported campaign changes and Magnify's proprietary experiment engine for controlled product-title testing.
Check measurement quality and evidence coverage before trusting a winner.
Keep one material question per experiment so the result is interpretable.

Audit my Google Ads account →

Experimentation methodology

Test the decision, not the noise.

What Google Ads A/B testing actually means

A useful experiment compares a control with a treatment while holding the rest of the decision environment as stable as practical. Google Ads has native Experiments for supported campaign changes, but not every optimization idea belongs in that framework. The objective is not to make two versions exist. The objective is to isolate one decision well enough that the result teaches you something reusable.

Start with one business question, not a bundle of unrelated changes.
Choose the primary success metric before the test starts.
Define guardrails such as spend, conversion volume or profitability so a local win does not create a bigger loss elsewhere.

What is not a clean experiment

A before-versus-after comparison is not automatically an A/B test. Seasonality, promotions, budget shifts, learning periods, feed changes and conversion-tracking changes can all move performance at the same time. Changing several variables together may improve the account, but it makes the causal lesson weak.

Do not call a multi-change account rebuild a controlled A/B test.
Do not stop the test the first day one variant looks ahead.
Do not force a winner when traffic or conversion volume is too low.
Do not treat missing or broken measurement as evidence that the control lost.

How Magnify creates value before the experiment

Magnify is designed to answer the higher-value question: what should you test next? The 92-point audit checks the account baseline first. Margin Max prioritizes financially meaningful problems and opportunities. That turns experimentation from random optimization activity into a decision pipeline grounded in the account’s own evidence.

Diagnose the account before creating the hypothesis.
Estimate whether the expected upside is material enough to justify the test.
Use native Google Ads experiments where the platform supports the change cleanly.
Use a different controlled methodology for feed, product or website changes that are not equivalent to a campaign experiment.

Magnify's catalogue experiment engine

For product-title tests across a catalogue, Magnify can build concurrent matched-pair randomized holdouts instead of relying on a weak before/after comparison. Products are matched using baseline traffic and performance context, assignment is locked before launch, challengers are published only to Treatment, and Google publication is verified before the measurement is trusted.

Match comparable products using feed/category context plus impressions, CTR, CVR, ROAS and price.
Randomize within matched pairs and lock Control/Treatment assignment before launch.
Measure CTR, CVR, ROAS, CPA and conversion value across the same baseline and experiment windows.
Use paired difference-in-differences plus a 1,000-iteration bootstrap confidence interval.
Check publication coverage, contamination, traffic, run duration and campaign/bidding drift before allowing a causal result.
Roll out challengers or restore the exact pre-test state and verify the final state in Google.

A practical experiment workflow

The correct sequence is diagnosis → hypothesis → testability check → experiment design → observation → decision → recorded learning. The learning matters because the next recommendation should know what your account has already proven, not restart from generic best practices every month.

Verify tracking and primary conversion goals.
Write one falsifiable hypothesis and one primary metric.
Choose control, treatment, traffic split and a sensible observation window.
Avoid overlapping material changes that contaminate the result.
Evaluate the result with enough evidence instead of declaring a winner from noise.
Record the outcome so future optimization can build on account-specific knowledge.

Testability guide

What you can — and cannot — A/B test cleanly in Google Ads

The exact experiment options depend on campaign type and current Google Ads capabilities. The important distinction is whether the change can be isolated through a native campaign experiment or needs another controlled testing method.

Area	Can you test it?	How to approach it
Bidding strategy	Often yes	A strong candidate for a native campaign experiment when the account has enough conversion evidence. Keep budget, goals and other major variables stable.
Search targeting or campaign settings	Often yes	Test one meaningful hypothesis at a time. Simultaneous keyword, bid, budget and creative changes make the result harder to interpret.
Performance Max settings	Supported in specific experiment types	Use the native PMax experiment options available for the change you are evaluating instead of creating an improvised before/after comparison.
Standard Shopping vs Performance Max	Supported use case	Use the appropriate migration or comparison experiment when available so traffic allocation and measurement are controlled.
Landing page	Yes, with the right setup	Campaign experiments can be useful for some landing-page hypotheses, while dedicated CRO testing can be cleaner for page-only changes.
Product titles and feed treatments	Yes — with Magnify's ecommerce experiment engine	Magnify can run SKU-level title tests and catalogue-level matched-pair randomized holdouts, verify the live Google title, measure Control and Treatment concurrently, and gate causal claims on publication and evidence quality.
Several unrelated changes at once	Technically possible, analytically weak	You may improve performance, but you will not know which change caused the result. Use this for implementation, not for learning.
Very low-volume account	Often not meaningfully	Insufficient traffic or conversions can make the result inconclusive. The correct decision can be to wait, aggregate differently or skip the experiment.

Free after Google Ads connection

checks in one account baseline

Magnify groups the audit into 23 sections and keeps failures, opportunities, healthy checks and unknown evidence visible.

Core areas checked

Conversion tracking, primary goals and conversion-value integrity

Traffic and conversion volume needed to support a useful test

Bidding, budget and campaign structure that can contaminate a comparison

Search-term, keyword and PMax demand quality before testing

Shopping feed and product-data issues that require a different experiment design

Evidence gaps that should stay unknown instead of being forced into a winner

The lead magnet

Magnify diagnoses the account, then runs product experiments with stronger evidence controls

The 92-point audit exposes account conditions that can invalidate a test and Margin Max helps prioritize financially meaningful hypotheses. For ecommerce product titles, Magnify can go further: match comparable products, randomize Control and Treatment, publish challengers only to Treatment, verify the processed Google title, measure both arms over the same calendar window, and refuse a causal claim when publication, traffic, contamination or campaign-context checks fail.

Your account is the case study.

Sign up, link Google Ads and Magnify creates the baseline from your own campaign evidence. The audit stays inside your dashboard so it can become the starting point for optimization — not a one-off downloadable checklist.

Start the free audit →See the 92-point audit page

Frequently asked questions

Is the 92-Point Google Ads Audit free?

Yes. Create a Magnify account, link Google Ads and Magnify builds the 92-point baseline before you need to make any account changes.

Does Magnify change my Google Ads account during the audit?

No. The first analysis is read-only. Magnify explains findings and proposed actions; account changes stay approval-controlled.

What do I need to connect?

For the Google Ads audit, connect the Google Ads account you want analyzed. Merchant Center is requested only when product or feed workflows need it.

Does Magnify have its own A/B testing engine?

Yes. Magnify supports SKU-level product-title tests and catalogue-level concurrent matched-pair randomized holdouts for ecommerce product optimization. The free audit itself does not launch tests or change the account; experiment launch and final rollout remain deliberate actions. For supported campaign-level changes, native Google Ads Experiments can still be the correct execution layer.

How does Magnify avoid false winners in catalogue tests?

Magnify verifies treatment publication in Google, checks matched-pair count, group impressions, publication coverage, arm imbalance, contamination, run duration and campaign-design changes, and uses paired difference-in-differences with bootstrap uncertainty. Results can remain provisional, invalid or inconclusive instead of being forced into a winner.

Can every Google Ads account run a useful A/B test?

No. An account can be too low-volume, have unreliable conversion measurement or contain too many simultaneous changes for a clean result. In those cases, forcing a winner creates false confidence rather than useful evidence.

How long should a Google Ads experiment run?

There is no universal duration. The useful window depends on traffic, conversion volume, conversion delay, variability and the size of the effect you are trying to detect. Magnify’s position is to avoid declaring winners from short-term noise.

Can I A/B test Performance Max?

Google Ads supports specific Performance Max experiment use cases. The right setup depends on what you are testing. Magnify’s role is to diagnose whether PMax is actually the problem and whether the proposed change is worth testing before you create the experiment.

Related Google Ads resources

Magnify · Margin Max

Find the leaks before you scale the spend.

Connect Google Ads and get the free 92-point baseline. Magnify reads first, explains the evidence and keeps account changes approval-controlled.

Get my free 92-point audit →