In short
The learning status shown on a campaign normally clears within a week or two of a change, but that label and a settled campaign are not the same thing. A campaign is genuinely settled when its cost per order and return stop swinging week to week, which usually takes longer, and takes longer still on a store with few conversions or a long gap between click and purchase. Expect a couple of weeks on a busy account and a month or more on a small one. Any target, bidding or tracking change starts it again.
The status label and a settled campaign are different things
The learning label tells you the bidding strategy is recalibrating after a change. It clears on its own timetable, generally within days to a couple of weeks, and once it clears the campaign is bidding normally rather than performing predictably.
What you actually care about is when the numbers stop moving for reasons other than demand. That takes as long as it takes to accumulate enough conversions for the weekly figures to be stable, which is a function of your order volume rather than of any timer inside Google.
A useful way to think about it: the label is about the algorithm, the settling period is about your sample size. A campaign producing several orders a day gathers a readable sample quickly. A campaign producing four orders a week will show wild weekly swings for a month or more, and none of that variation means anything.
So judge the campaign on a window long enough to contain a real sample, and resist reading the first days at all. The most expensive habit in a young account is reacting to the learning period as though it were performance.
Three things decide how long it takes
Conversion volume is the first and largest. The bidding needs examples, and examples arrive at the rate your store sells. Everything else being equal, a campaign with ten times the conversions settles far faster.
Conversion delay is the second, and it is the one most often forgotten. If your buyers typically take a week between the first click and the order, then a week of data is not yet a week of outcomes, and the earliest a change can be fairly assessed is a full cycle later. Categories with considered purchases, furniture, high value electronics, anything requiring a fitting or a quote, carry the longest delays.
The size of the change is the third. Nudging a target slightly is a smaller adjustment than doubling a budget or switching bidding strategy, and it disturbs less. Large changes made at once produce longer, sharper learning periods than the same change made in steps.
Seasonality sits behind all three. A change made in the week demand shifts will look like the change caused whatever happened, and separating the two after the fact is usually impossible.
What quietly restarts it
More than you would expect. Changing the bidding strategy or moving a target by a meaningful amount is the obvious one. So is a large budget change, because the campaign has to work out what a different volume of auctions looks like.
Less obvious: changes to the conversion actions themselves. Adding a conversion action to the primary column, changing the attribution window, switching the counting option or altering what value is sent all change the objective, and the strategy has to relearn against the new definition.
Also less obvious: structural edits. Adding or removing asset groups in Performance Max, changing the listing groups, replacing the product feed or moving products between campaigns all shift what the campaign is bidding on. In search campaigns, wholesale keyword changes have the same effect.
And tracking breakage restarts it involuntarily. If purchases stop being recorded for two days because of a theme deployment, the strategy sees a collapse in conversions and adjusts. When tracking returns, it adjusts again. That double disturbance is often mistaken for a campaign problem when it was a tag problem.
How to make it shorter and what to do meanwhile
Give the campaign more signal rather than more attention. Enhanced conversions increase the share of orders that can be attributed to a click, and consolidating duplicate conversion actions into one clean purchase action makes every conversion count for more.
Consolidate structure too. Splitting a modest budget into several campaigns divides the conversions among them and lengthens every learning period simultaneously. Fewer, better fed campaigns settle faster than many thin ones.
Change one thing at a time and write down the date. A change log with the date, the change and the reason turns an unreadable account into one where you can attribute movements to causes, and it costs a minute per edit.
While it runs, do the work that does not disturb it: fix feed errors, improve product pages, write and test new assets ready for the next cycle, clean up negatives at account level. All of those improve the campaign without resetting it. Then hold the settings until the review date you set at the start.
The same change, two stores
- Store A orders per week
- About 400
- Store A time from click to order
- Under a day for most orders
- Store A readable after
- Roughly two weeks
- Store B orders per week
- About 20
- Store B time from click to order
- Around eight days
- Store B readable after
- Closer to six weeks
- Same change made
- Target lowered by a fifth
Illustrative comparison. The lesson is that the waiting period belongs to the store rather than to the platform, so two accounts making an identical change need different review dates.
Related questions
Should I pause a campaign that is performing badly during learning?
Only if it is spending at a rate you cannot afford or is clearly serving on the wrong queries. Pausing and restarting throws away the learning and begins the period again. Where the issue is cost rather than relevance, reducing the budget disturbs less than switching the campaign off.
Does a small target change restart learning?
Small adjustments cause a smaller disturbance than large ones, and repeated small adjustments add up to a campaign that is permanently adjusting. Set a review interval, make one change per interval, and let each one play out before deciding whether it worked.
Why does my campaign look worse after learning ended?
Usually because the early period included an unrepresentative sample, often the cheapest and easiest conversions available. As the campaign broadens, the average cost per order rises towards a truer figure. Compare against a full cycle rather than against the first week.