Search ad experiments

A search ad experiment compares a proposed change against an alternative under broadly similar conditions. Begin with the decision the result must support—for example, whether to use a different page offer or bidding goal. Before splitting traffic, choose the business outcome and the improvement that would justify the change.

Candidate tests can also compare ad variations or targeting options. Match the type of change to the decision you need to make; do not treat every campaign adjustment as the same experiment.

Google Ads Experiments lets you prepare a draft of an existing campaign, then run it against the original as an experiment. The versions run simultaneously, with traffic assigned randomly and persistently to users, rather than relying on a before-and-after comparison.

Choose the experiment format to fit the question: use a campaign experiment for campaign changes, test ad variations when the question is about ad copy, and consider a custom experiment for a custom design. Google Ads Scripts names SEARCH_CUSTOM as an experiment type.

Ask one question at a time

A useful hypothesis names the change, expected outcome and acceptable downside. For example: “A different quote offer may increase accepted enquiries without pushing cost per accepted enquiry above our limit.” This is a testable proposal, not a forecast.

Keep the main measure close to the decision. Orders may suit a retailer; a service business may need to assess which enquiries it could accept. Click-through rate can help explain attention, but cannot establish whether the change produced useful work.

Choose a comparison that fits the question

For a campaign-level comparison, define the original and proposed versions, then check whether Google Ads Experiments supports that design. Prepare the proposed changes in a campaign draft before launching the experiment against the original.

Keep targeting, keywords, conversion actions and other settings aligned unless they are the planned change. Record any necessary difference so you can account for it when interpreting the result.

Set the traffic split when creating the experiment, and check that both versions are serving before treating the comparison as underway. A Google Ads Scripts ExperimentBuilder example uses a 50% traffic split; it is an example setting, not a universal default.

Traffic is randomly assigned to users and remains consistent for their later impressions, while the original and experimental campaigns do not compete against each other in auctions. The daily budget is divided according to the traffic split, so include that allocation in your spending plan.

Changes during a run make interpretation harder. Keep the original active and note any delivery problems or material changes. Use the All experiments tab to track experiment creation.

A campaign experiment compares the versions among the traffic they receive. It cannot establish whether advertising created customers who would otherwise have arrived through another route. For that question, use an appropriate holdout and measure total outcomes; comparable areas with brand ads withheld in some areas provide the comparison described above.

Set the measurement and stopping rules first

Record the intended change, primary outcome, traffic split, spending limit, review period and decision threshold. Include the time needed for clicks to become orders or assessed enquiries. State what would justify adopting the trial, keeping the original or gathering more evidence.

Assess the estimated difference and uncertainty against the improvement the business needs. Google Ads reports statistical significance for experiment results, but a statistically significant difference still needs to meet the business threshold you set.

Do not treat an early lead or a real-time reading as a dependable result while data is still accumulating. Set an evidence requirement before launch, based on the primary outcome and the size of improvement that would matter to the business.

Allow time for conversions to arrive after the ad interaction before reading recent results as complete. If the planned review period ends before the evidence requirement is met or delayed conversions have arrived, classify the result as inconclusive or extend it only within the stated limit.

Make a bounded decision

Compare the planned measure, reported statistical significance, uncertainty, spend and later business outcomes. Record the settings, dates, split, delivery problems and material changes. Conclude adopt, keep the original, extend within a stated limit or inconclusive, according to what the evidence supports.

Adopt the trial only if its result clears the pre-set business threshold and the evidence is sufficient for the decision. Keep the original when the trial does not justify its downside, and avoid treating an inconclusive result as proof that the versions are equal.