Google Ads experiments: how to A/B test a bid strategy or landing page

Quick answer

Google Ads experiments run one change on a copy of a campaign that splits traffic and budget with the original. Use them to test a new bid strategy or landing page without risking the account: a 50% split, one change at a time, and a decision only at statistical confidence. Then apply it to the original or end it.

What is a Google Ads experiment, and when should you use one?

A custom experiment creates a trial version of a campaign, applies the change you want to evaluate, and splits traffic and budget between the original and the trial. What Google Ads Help says, checked on October 17, 2026:

• Custom experiments are available for Search, Display, Video and Hotel campaigns. Shopping and App campaigns cannot use them.
• You can schedule up to five experiments per campaign, but only one runs at a time.
• The original campaign must be active, because the trial depends on it.
• Changes to the original campaign are not reflected in the experiment.
• When you apply, the changes go to the original or the trial becomes a new campaign; performance data is kept either way.

Use one when the change touches money and you are not sure of the effect: moving from manual CPC to automated bidding, adding a CPA target, sending traffic to a new page. For a small, reversible tweak like one more negative keyword, an experiment is overhead.

What is worth A/B testing in Google Ads?

One change per experiment. If the trial swaps both the bid strategy and the page and wins, you cannot tell which one won.

How do you set up a custom experiment in Google Ads?

1. Open Experiments under Campaigns.
2. Click the plus button above the All experiments table, choose Custom and then the custom test for your campaign type.
3. Name the experiment, pick the original campaign and a suffix for the trial campaigns.
4. Select up to two success metrics.
5. Make the change on the trial campaign.
6. Under the experiment split, set the share of traffic and budget. Google recommends 50%.
7. Under scheduling, pick a start date and duration, then save.

Cookie-based or search-based split?

With a cookie-based split, which Google recommends, each person sees only the original or only the trial. With a search-based split, assignment happens on every search, so one person may see both; Google says it may reach statistical significance faster. For a landing page test, cookie-based keeps people from seeing both pages. Display experiments always use a cookie-based split.

How do you run and read an experiment in Ads Editor?

Since October 10, 2026, Ads Editor's Google manager creates experiments on Search and Display campaigns, from the campaign's Edit panel, under Experiment (A/B test).

1. Click New experiment and name it.
2. Choose what the trial changes: the trial bid strategy (with the target CPA or ROAS when it applies) or a new landing page. For the page, you can replace only the ads that use a specific URL, or all of them.
3. Set the trial traffic split from 10% to 90% (50% by default) and, optionally, a start and end date.
4. Ads Editor creates the experiment, applies the change to the trial campaign (which gets the [teste A/B] suffix) and schedules it in Google Ads.
5. Follow original and trial side by side: impressions, clicks, conversions, conversion rate and CPA, with the change and the statistical confidence.
6. Decide: Apply to original brings the change to the original campaign and ends the test; End keeps the original as it was.

Through MCP, Claude or ChatGPT can list, create, apply, end and remove Google experiments under the same rules. Converting a trial into its own campaign stays in Google Ads, because it needs a new budget.

How long should a Google Ads experiment run?

Google does not set a minimum on its experiments page. Three things matter more than the calendar:

• Learning period. A new bid strategy goes through learning, and Google Ads Help says it can take up to 50 conversion events or 3 conversion cycles to calibrate. The first days of a bid strategy trial usually read worse; do not decide on them.
• Volume. Wait for statistical confidence. A 10% split takes far longer to get there than 50%.
• Full weeks. If the business sells differently on weekends, compare whole weeks so you do not mistake a weekday pattern for a trial effect.

Schedule the start ahead of time: Google reviews the trial ads before it begins, and review time depends on the size of the original campaign.

Which mistakes invalidate an experiment?

• Testing two changes at once. The result cannot tell you which one worked.
• Editing the original mid-test. The edit does not reach the trial, and the two stop being comparable.
• Deciding on thin data. A 20% difference on 30 clicks is noise.
• Broken conversion tracking. If conversions double count, an automated bidding trial optimizes toward the wrong number.
• Applying without a note. Log the date you applied it so you can read the effect later, as you would with a Google recommendation.

Experiments are one of the routines in the guide to managing Google Ads for multiple clients. For smaller bid changes by device, hour or audience, see the guide to Google Ads bid adjustments.