A campaign experiment in Google Ads is a controlled test in which a copy of an existing campaign, carrying one change, runs at the same time as the original and shares its traffic. At the end you compare the two and decide whether to apply the change, so you learn what a new bid strategy or setting does before committing your whole budget to it.
How a campaign experiment works
You start from a live campaign, called the base, and create a trial version through the Experiments section of Google Ads. Older guides call this “drafts and experiments”, because you used to build a draft first; the idea is the same. In the trial you change one thing: a switch from manual CPC to Maximise Conversions, broader match types, a new landing page, or different location settings.
You then choose how to split traffic, usually 50/50, and whether to split by search (each auction goes to one arm at random) or by cookie (each user is kept in one arm so they see a consistent version). Cookie-based splits are better for changes that affect what a person sees across visits, such as ad copy or landing pages.
Both arms run in parallel over the same dates, so seasonality, competitor activity and news affect them equally. Google Ads reports the difference in clicks, conversions, cost and other metrics, with a confidence indicator to show whether the gap is likely to be real. You can apply the trial’s changes to the original campaign, convert the trial into a new campaign, or end it.
Google also offers specialised experiment types, such as video experiments and experiments built for Performance Max. At the time of writing (October 2026), the available types depend on the campaign type, so check the Experiments page in your account.
Why it matters
Before-and-after comparisons are unreliable in paid search. If you change the bid strategy in March and conversions rise in April, you cannot tell whether the change worked or whether demand rose with the season. An experiment removes that doubt, which matters most when the change is expensive to get wrong, such as moving a large account onto Smart Bidding.
It also protects the business. Only part of the budget is exposed to the change while you learn, and if the trial underperforms you end it and nothing has happened to the original campaign.
Common mistakes
- Testing several changes at once, so a result cannot be traced to any one of them.
- Ending a test after a few days. Automated bidding needs time to settle, and conversion lag means recent clicks have not finished converting.
- Running experiments on campaigns with very few conversions, where no difference will ever reach a confident result.
- Editing the base campaign during the test, which changes the control and spoils the comparison.
- Judging on cost per click when the decision should rest on cost per conversion or conversion value.
How to act on it
Write down the change, the metric that decides the outcome and what result would make you apply it, before you build the experiment. Choose a campaign with steady conversion volume, split traffic evenly, and plan to run for at least four weeks, longer if your customers take time to convert. Leave both arms alone while it runs.
When it ends, look at conversions and cost per conversion first, then check the confidence indicator. If the result is unclear, treat that as an answer too: the change made no reliable difference. Keep a log of every test so you do not repeat old ones. The same principles apply to any A/B testing, and structured testing is part of how I run PPC management each month.