How ad rotation works
When an ad group contains more than one eligible ad, something has to decide which one is served. Ad rotation is that setting, and it offers two positions. Optimise lets Google favour the ads it expects to perform better, using signals from the auction such as the query, the device and the person’s context. Do not optimise serves the eligible ads more evenly over time.
The setting lives at campaign level and applies to the ads inside it. It does not control the assembly of pieces inside a single responsive ad — that happens separately, at asset level, and continues regardless of what rotation is set to.
One detail changes the whole picture: when a campaign uses an automated bidding strategy, the bidding system is already choosing what to serve in order to hit its target, so the rotation setting has little left to control. In most modern accounts, rotation is a smaller lever than it looks.
Why ad rotation matters
It decides whether the account is exploiting what it already knows or exploring what it does not. Optimise gets more from proven copy sooner, which suits a campaign with steady volume and a clear winner. Even rotation gives a new ad room to gather data before an older, better-established one crowds it out.
For accounts with modest traffic, the practical difference is smaller than the debate suggests. A campaign that receives few clicks a day will take a long time to prove anything either way, and the setting cannot manufacture the volume that a real comparison needs.
Common mistakes with ad rotation
The most common is using even rotation as a testing method. Serving ads evenly is not the same as a controlled experiment: the ads still face different queries, devices, times of day and competitors, so the two sides are never comparing like with like. Any conclusion drawn from it is weaker than it appears.
The second is switching to even rotation inside a campaign that uses automated bidding and then expecting the split to hold. It largely will not, and the resulting report is confusing rather than informative.
The third is leaving a long-dead ad running beside a new one because rotation will supposedly sort it out. Rotation chooses between what you give it; it does not remove tired copy, and an ad nobody has looked at in a year will keep collecting impressions.
How to act on it
Leave rotation on Optimise for almost every campaign, and treat the alternative as a specialist tool rather than a default. That single decision removes a setting people spend far too long adjusting.
When you genuinely want to compare copy, use a method built for it. Ad variations and campaign experiments split traffic deliberately and report both sides against each other, which is what a test needs in order to be trusted. Keep the number of ads per ad group small so each has enough traffic to matter, and remove ads you have already decided against instead of asking a serving setting to do it for you.