Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Experiment Library

Search Ad Asset Test

A test design for responsive search ad assets: the hypothesis, conversions per impression as the metric, guardrails, variation design and reading results.
Experimentintermediate6 min readUpdated

Hypothesis

Adding headlines that carry the offer and a proof point to the responsive search ad will raise conversions per impression, because the ad answers the query and pre-qualifies the click. Searchers scan results quickly. An ad that states what is on offer, and gives one reason to believe it, helps the right searchers choose it and helps the wrong ones skip it. Both effects are good for conversions per impression, even if one of them lowers clicks. Where the added headline is a qualifier, such as a price or a trade-only line, the guide to pre-qualifying search ads covers pinning it and reading the result on qualified leads.

The change must be specific. Decide which headlines will be added, and whether any existing headlines are removed to make room, before building the test. The responsive search ads guide explains how headlines are combined at auction time and why adding assets does not guarantee they show; the ad assets guide covers the separate assets, such as sitelinks, that sit around the ad and should be left unchanged during this test. Record the exact asset list for both arms in the experiment hypothesis template.

Metrics and guardrails

The primary metric is conversions per impression: conversions divided by impressions in each arm. It is the product of click-through rate and post-click conversion rate, so it rewards ads that win the right clicks rather than the most clicks. Use the same conversion actions and attribution settings in both arms, and report the conversion rate after the click with the conversion rate calculator as a diagnostic.

Click-through rate is the first guardrail. A modest fall in CTR is acceptable if it reflects pre-qualification, but a large fall can harm ad rank over time, because expected click-through rate is one of the components the platform describes for Quality Score. CPA is the second guardrail: the new ad may change which auctions are won and at what cost, so check that acquisitions are not becoming dearer. The Quality Score guide explains how the diagnostic is reported and why it should not be the target of a test.

Design

Use the platform's own testing tools, which the Google Ads Help centre documents: a campaign experiment that splits traffic between a base campaign and a trial, or an ad variation that applies a change across selected ads and splits the traffic. Both split at auction or search level within the platform's method, which is the unit of randomisation here. Choose an even split so both arms gather evidence at the same rate.

Design elementChoiceReason
MethodCampaign experiment or ad variationPlatform-managed split of auctions
Unit of randomisationSearch or auction, as the platform splits itFair allocation of comparable queries
VariantsCurrent ad versus ad with offer and proof headlinesOne asset change
Held constantKeywords, bids, budgets, other assets, landing pageOnly the headlines differ
SplitEvenEqual evidence per arm
DurationWhole weeks, enough auctions per armCovers weekly query patterns

Avoid editing the campaign during the test. Bid, budget or keyword changes apply to both arms in some setups and only to one in others, and either way they interrupt the comparison.

Sample size reasoning

Conversions per impression is a small ratio, because most impressions produce no click and most clicks produce no conversion. Detecting a change in it needs many impressions per arm. The inputs are the current ratio, the smallest improvement worth rolling out, and the confidence required; the sample size and significance guide shows how to combine them.

In practice, the limiting factor is often conversions rather than impressions. A high-volume ad group may reach a readable result in a few weeks; a niche one may never get there. For low-volume accounts, consider running the change across several ad groups at once with an ad variation, so that evidence accumulates faster, provided the ad groups share a similar offer and audience.

Reading the results

Read once the planned duration and the planned volume per arm have both been reached. The experiment report shows each arm's metrics and the platform's own view of whether differences are significant; read it alongside your own calculation on conversions per impression. If the trial arm wins on the primary metric and both guardrails hold, apply the change.

Then look at how the result happened. If click-through rate fell but post-click conversion rate rose enough to lift conversions per impression, the ad is pre-qualifying, which was the mechanism the hypothesis described. If click-through rate rose and conversion rate fell, the new headlines attracted clicks the offer could not convert. Check whether the search terms report differs between arms; a large shift in which queries triggered each ad means the arms were not competing for the same searches.

Illustrative example — not a real result

A boiler repair firm adds offer and proof headlines

HeatRight, an invented boiler repair firm, runs a responsive search ad with generic headlines about "fast, reliable repairs". It creates an ad variation that adds "Fixed fee: £89 call-out" and "Gas Safe engineers, rated 4.8" as headlines. Over four weeks, each arm receives about 30,000 impressions. Click-through rate is slightly lower in the trial arm, but post-click conversion rate rises, and conversions per impression move from 0.40% to 0.52%. Cost per acquisition falls from £41 to £34. The firm applies the variation and plans a test on the description lines next.

Pitfalls

Treating asset ratings as the result is the most common pitfall. The ratings compare assets within one ad and describe relative serving performance; they are not a controlled test and they say little about conversions. Use them to choose what to test next, not to decide what won.

Pinning is the second. Pinning headlines restricts which combinations can show, and pinning the new headlines in one arm only changes two things at once. Apply the same pinning rules to both arms. Query mix shifts are the third: if the changed ad starts to show for different searches, perhaps because new words in the headlines affect relevance, the arms are no longer answering the same people. Compare search terms by arm before trusting the result. Finally, do not run this test while automated bidding is still in its learning phase after another change, as the noise will swamp the asset effect.

Common mistakes

Search asset tests are tempting to run on autopilot because the platform offers ratings and suggestions at every turn. For video and social creative, the ad creative hook test applies the same one-change rule, and the experiment library lists the remaining designs.

  • Reading asset ratings as the outcome. They are relative serving labels, not a test result.
  • Optimising for click-through rate. More clicks from the wrong people raise costs.
  • Changing several assets at once. Then nobody knows which headline did the work.
  • Pinning in one arm only. That adds a second variable.
  • Editing bids or keywords mid-test. The comparison restarts whether you notice or not.
  • Skipping the search terms check. Different queries mean different audiences.

Frequently asked questions

Why use conversions per impression instead of click-through rate?

Click-through rate rewards ads that attract clicks, including clicks from people who will not buy. Conversions per impression combines whether people click with whether they convert, so it rewards ads that attract the right clicks.

Can asset performance ratings decide the test?

No. The ratings describe how individual assets performed relative to others within the ad, as the help centre documents them. They are not a controlled comparison and they do not measure conversions, so treat them as hints for the next test.

Should we pin the new headlines to make sure they show?

Pinning in the test arm alone would add a second change. If you pin, pin equivalent positions in both arms, or leave pinning as it is in both and accept that the system decides how often the new headlines appear.

Terms used on this page

Tools to try

Frameworks

Where this page is used

Also related

Sources

  1. Google Ads Help — Google Ads Help

Related