Ad Rotation in Google Ads 2026: Why One Ad Takes Almost All the Impressions
What the ad rotation setting actually does
You know the pattern: three ads in one ad group, one collected 92% of impressions, the second got 7%, the third barely served. The conclusion writes itself — “two of the creatives are weaker.” It is usually wrong, because ad rotation in Google Ads does not control impressions. It controls how evenly your ads enter the auction, and those are two different things. This piece covers which options are left in 2026 and what each does in Google’s own wording, why Smart Bidding silently overrides your choice, why impressions still diverge under an “even” rotation, why responsive search ads have combination assembly instead of rotation, and how to tell a settings problem from a real account problem in fifteen minutes.
Everything below comes from Google’s help pages: Use ad rotation, Fix unevenly serving ads, plus the Ad Strength and experiment-statistics articles. Where the documentation gives no number, this article says so instead of inventing one.
Two options and their exact wording
The help page opens with a plain fact: “There are 2 settings for ad rotation: ‘Optimize’ and ‘Do not optimize’.” Everything you may remember about rotating evenly for 90 days, optimising for clicks or optimising for conversions is history — Google collapsed the setting into two choices in late September 2017. The even option is flagged in the diagnostics article as “formerly known as ‘Rotate Evenly’.”
| Option | What the documentation says it does | What it is not |
|---|---|---|
| Optimize (UI label: “Optimize: Prefer best performing ads”) | It “optimizes your ads for clicks in each individual auction using signals” — keywords, search terms, devices, locations and more — and prioritises ads that “are expected to perform better than other ads within an ad group”. Over time, “ad serving will become weighted more heavily in favor of the ads that appear statistically likely to perform better” | Not a guaranteed skew: “it’s possible for ad serving to remain relatively even within this setting”, for example when ads perform similarly or the ad group does not get a significant volume of impressions and clicks for some period |
| Do not optimize (“Do not optimize: Rotate ads indefinitely”) | “your ads will enter into the ad auction more evenly for an indefinite amount of time”. There is no 90-day window any more — the option simply does not expire | Not equal impressions: “the percentage of impressions for each ad may not be even due to the quality of the ad”. And it “isn’t recommended for most advertisers” |
Pay attention to how Google explains the skew: through position in the results. “An ad with high quality may appear on the first page of search results, while an ad with low quality may show on the second page of search results, which reaches a smaller number of users.” So even with genuinely even auction entry, a weaker ad physically collects fewer impressions — it gets shown where fewer people are looking. How that position is assembled and how your click price follows from it is covered in our breakdown of Quality Score and Ad Rank.
One more line from the same page worth remembering before you read any report: “no more than one ad from your account can show at a time”. Your ads are not competing for two slots on the same results page — exactly one of them is in each auction.
Which campaign types have it, and where to find it
Google lists the campaign types in a single sentence: “You can use the ‘Ad rotation’ setting in Shopping, Search and Display campaigns.” Performance Max, Demand Gen, Video and App campaigns are not mentioned on that page at all. That absence is not a confirmation that the setting applies there quietly — it is simply an absence, and it should be read as one.
The documented path: the Campaigns icon → the Campaigns drop-down in the section menu → Ad groups → hover over the campaign or ad group you want → the settings icon → Additional settings → Ad rotation → pick Optimize or Do not optimize → Save.
To change many at once: tick the campaigns or ad groups, click Edit, then Change ad rotation from the drop-down. To see what is actually set across the account, Google suggests the filter: “Use the filter icon to show only those campaigns that match specific criteria, including ad rotation.” That is the fastest audit of a whole account without opening campaigns one by one — particularly useful when you inherit an account and do not know its history.
There is also an internal inconsistency worth knowing about. The 2017 announcement says “ad rotation settings will be available at the ad group level — meaning that you can use multiple rotation settings across a single campaign.” The current help page documents the steps under the heading “Set ad rotation for your campaign” and shows no separate ad-group menu. Practical takeaway: verify the real value with the filter at both levels rather than trusting your memory of where you once set it.
Why Smart Bidding overrides your choice
This is the single point that makes the “Optimize versus even rotation” debate irrelevant for most 2026 accounts. Google states it directly: “If you’re using Smart Bidding, which prioritizes conversions, Google Ads will automatically use the ‘Optimize’ ad rotation setting.” The 2017 announcement said the same thing: “if you’re using Smart Bidding, AdWords will automatically use the ‘optimize’ ad rotation.”
| Bidding strategy | What you selected | What actually runs |
|---|---|---|
| Target CPA / Target ROAS / Maximise conversions / Maximise conversion value | Optimize | Optimize |
| The same Smart Bidding strategies | Do not optimize | Optimize — your choice is overridden automatically |
| Manual CPC, Maximise clicks and other strategies without conversion prediction | Optimize | Optimize |
| The same strategies | Do not optimize | Do not optimize — the setting genuinely applies |
Two consequences follow. First: in a Smart Bidding campaign, switching to “Do not optimize” changes nothing, so the cause of uneven serving is somewhere else. Second: a fair creative A/B test “through rotation” is technically impossible under Smart Bidding, because the platform will keep favouring the ad it predicts will convert better. The right instruments are further down.
Ad rotation is a delivery setting, not a testing tool. Wherever Smart Bidding runs, the choice between the two options has already been made for you, and the only correct way to compare creatives is a dedicated experiment mechanism.
“Evenly” means auction entry, not impressions
The most common reporting mistake is answered by one sentence in the documentation: “Setting ads to rotate evenly doesn’t always mean that ads will serve more evenly. It means that ads will enter the auction more evenly.” And then the reason: “Because ads aren’t guaranteed to win every auction entered, ads don’t always rotate evenly.”
The difference is practical. An equal number of auction entries turns into an unequal number of impressions whenever two ads win at different rates. Winning depends on factors Google lists as “ad quality, bid, competition, etc.” So a 60/40 impression split under even rotation is not a bug and not a support ticket — it is the normal outcome of two ads being differently competitive.
One separate case is easy to confuse with rotation: an ad that is approved but barely serves at all. That is usually not the rotation setting but limited ad serving or an account-level issue, and no rotation change will fix it.
Six explanations Google gives for uneven serving
The “Fix unevenly serving ads” page has no numbered list of causes. It offers six statements, and together they are the whole diagnosis:
- The setting was switched on later than the period you are looking at. “If your ads are serving unevenly even though ‘rotate indefinitely’ is selected, your campaign may have been opted into this setting after the affected timeline.” Days before the switch still follow the old rules.
- Optimisation was running before that. “If ‘Optimize: Prefer best performing ads’ was selected before the affected timeline, this is why your ad is serving unevenly.”
- Or the opposite — even delivery was selected, and that explains it. “But, if ‘Do not optimize: Rotate ads indefinitely’ was selected, this explains your ad’s uneven serving.”
- Even means entry, not impressions. The wording is quoted above.
- An ad does not have to win every auction. “ads aren’t guaranteed to win every auction entered.”
- Quality, bid and competition. “Other factors such as ad quality, bid, competition, etc. may influence how an ad serves.”
There is not a single number on that page: no impression threshold after which a split counts as normal, no period over which optimisation “ramps up”, and no definition of that “affected timeline”. This is not nitpicking about wording — it is a real limit. Any advice of the form “wait for 2,000 impressions and then judge” is somebody’s practice on their own account, not a platform rule.
Three levels that decide what a searcher sees
Rotation arguments usually go sideways because three independent selection levels get mixed together. Separate them once and for all:
| Level | What gets decided here | What you control |
|---|---|---|
| Account | Which single ad from your account can appear at all: “no more than one ad from your account can show at a time” | Account structure and keyword overlap |
| Ad group | Which of several ads in the group enters the auction more often — this is ad rotation proper | The Optimize / Do not optimize setting, but only outside Smart Bidding |
| Ad (RSA) | Which combination of headlines and descriptions is assembled for a specific query | Your asset set and pinning — there is no rotation setting at this level |
When two people arguing about “why it doesn’t rotate” are talking about different levels, they cannot reach agreement. Establish the level first: “one ad took all the impressions” is level two, “one headline never shows” is level three.
RSAs have no rotation — they have combination assembly
A responsive search ad is not a finished creative but a pool of assets from which the platform assembles an ad each time. The documented limits: up to 15 headlines and 4 descriptions per ad, with a minimum of 3 headlines and 2 descriptions; character limits are 30 for a headline, 90 for a description and 15 for each path field.
Pinning is the only direct lever over which combinations get built, and it is an expensive one. “Pinning one headline or description causes it to show only in that specific position, preventing other headlines or descriptions from showing in its place.” Google then warns in plain language: “Because pinning reduces the overall number of headlines or descriptions that can be matched to a potential customer’s search, pinning isn’t recommended for most advertisers and can affect ad strength.” And the extreme case: “If you have headlines and/or descriptions pinned to all available positions, unpinned headlines and/or descriptions won’t show” — the rest of your asset pool becomes decoration. The five slots and the price of each pin are covered in pinning RSA headlines.
On serving guarantees: content pinned to “Headline position 1”, “Headline position 2” or “Description position 1” “will always show”, while “Content pinned to Headline position 3 and Description position 2 are not guaranteed to show in every ad”.
Now the reports, because this is where the wrong conclusion about “asset rotation” usually gets made. The performance-label column is gone: “The ‘Performance label’ column has been deprecated as full performance statistics for each asset are now available”, and that full statistics set is “only available for dates on or after June 5, 2025”. The report only includes active assets that have had impressions in the last 30 days, and the data refreshes daily. The key warning: “Occasionally, an asset might not receive any impressions if other assets are predicted to perform better.” Google also tells you where to judge: “Evaluate performance at the asset group level or campaign level, rather than at the individual asset level”, because the ratio metrics “should be used as directional indicators only” and “the sum of individual asset impressions, clicks, or costs may not directly match the corresponding metrics”. For dead assets the only guidance is “Consider replacing assets that receive zero impressions after several weeks” — and “several weeks” is never turned into a number. What is actually worth changing in that report is covered in Ad Strength and RSA asset reporting.
The combinations report answers “what was assembled”, but it comes with one explicit prohibition: “The combinations report shouldn’t be used to create static versions of responsive search ads because they’re not guaranteed to perform the same as your ads”, since “Responsive search ads use real-time signals to generate ads that maximize your performance”. The number of combinations shown and any date window are not stated.
And a word on Ad Strength, which gets mistaken for a rotation lever constantly. The scale is “Incomplete”, “Poor”, “Average”, “Good”, “Excellent”. What it does: it “primarily evaluates the relationship between the assets you provide and the keywords you are targeting”. What it does not do: “it does not measure external factors, such as the benefits of adding new keywords or the quality of your landing pages and offerings”, and most importantly “It isn’t used to calculate Ad Rank, Quality Score, or auction wins”, while “Ad Strength doesn’t determine whether your ad is eligible to serve”. Raising Ad Strength does not switch an ad’s impressions on.
What to use instead of “rotate indefinitely”
If your actual goal is comparing two pieces of copy, 2026 gives you two native mechanisms — and neither of them is the rotation setting.
Ad variations. Available for Search campaigns and responsive search ads. The variation types are “Find and replace”, “Update text” (add, remove or pin headlines and descriptions) and “Update URLs”, and variations “are case sensitive”. The split is entered in the “Experiment split” field, and the mechanics are described like this: “Your ad variation and original ads will be split by assigning cookies to users. That means users may discover only one version of your ad, regardless of how many times they search.” You set a start and end date; once it has started the start date cannot be changed, and “expired ad variations can’t be re-run”. Applying ends the test: “When you apply an ad variation, it creates new ads and your ad variation stops running. You won’t be able to collect any more performance comparison data” — with three choices: pause the original ads, remove them, or keep them. Approval of the new ads “takes up to one business day”.
Experiments. Custom experiments are available “for App campaigns and Search and Display campaigns”, and the documentation gives exactly one duration number: “we recommend allowing it to run for at least 4 to 6 weeks to collect enough data to evaluate the results”. The statistics underneath are spelled out: “Jackknife resampling is applied to bucketed data to calculate the sample variance of the percent change of a metric”, then “Two-tailed significance testing is then run using the 95% confidence interval”, with “Twenty buckets… in the control arm and twenty buckets… in the treatment arm”. There is an honest caveat too: “If there are too many buckets, then it might take too long to get statistically significant results.” How to frame hypotheses in that interface is in experiments in Google Ads, and the method for comparing RSAs specifically is in ad A/B testing when Google assembles the ad.
Note how different these tools are in nature: ad variations split users by cookie inside one campaign, while an experiment splits traffic between two arms and computes significance. The rotation setting does neither — it only changes delivery priority inside an ad group.
A 15-minute diagnosis, step by step
- Check the campaign’s bidding strategy first. Smart Bidding means Optimize is effectively forced and the setting is not worth discussing. That is thirty seconds of work.
- Read the real value with the ad rotation filter rather than from memory. You will also see whether different campaigns carry different values.
- Compare the date the setting last changed with your report period. This is precisely the “opted into this setting after the affected timeline” case: if you flipped it a week ago and are looking at a month, the skew stays in the data.
- Segment impressions by day and find where the curves separated. A sharp break on a specific day is almost always an account edit, not “optimisation”.
- Check whether the ad is approved with restrictions. An ad approved with limitations will collect few impressions under any rotation setting.
- Compare the ads on quality and bid, not just impression counts — a position difference explains a reach difference, exactly as the first-page / second-page wording describes.
- For RSAs, open the asset and combinations reports and confirm you are not looking at level three: “why doesn’t this headline show” is not an ad rotation question.
- Check your pins. If every position is pinned, unpinned assets never show — which looks exactly like “rotation is broken”.
- If you need a verdict on creative, launch an ad variation or an experiment with a 4-to-6-week horizon instead of flipping rotation and drawing conclusions from a three-day slice.
Five expensive mistakes
- Comparing the CTR of two ads while Optimize is on. The samples were not collected randomly: the platform deliberately gave more impressions to the ad it predicted would earn more clicks. Any conclusion from that comparison is biased by construction.
- Turning on “Do not optimize” in a Smart Bidding campaign and waiting for evenness. The choice is overridden automatically, so you spend weeks waiting for something that will not happen.
- Keeping even rotation “so the ads can gather data”. The option has no time limit, it does not accumulate anything by itself, and Google does not recommend it for most advertisers.
- Building a static ad out of the “best” RSA combination. This is the one scenario the documentation explicitly rules out: combinations “are not guaranteed to perform the same as your ads”.
- Trying to fix serving through Ad Strength. It is not used to calculate Ad Rank, Quality Score or auction wins, and it does not determine serving eligibility.
What Google’s help pages do not say
This section saves time in any rotation argument: below are the questions the platform has no public answer to, which means any number attached to them belongs to a specific account rather than to a rule.
- No volume threshold after which Optimize starts visibly skewing delivery. There is only “as data is accumulated”.
- No definition of “affected timeline” on the diagnostics page — even though three of its six explanations rest on it.
- No normal impression share. Google never says which split between ads should be considered expected.
- No consistent name for the setting: “Optimize” / “Do not optimize” on one page, “Optimize: Prefer best performing ads” / “Do not optimize: Rotate ads indefinitely” on another, “optimize” / “rotate indefinitely” in the announcement.
- No clarity on the level: the announcement promises ad group level, the current page documents campaign level.
- No list of unsupported campaign types — only Shopping, Search and Display are named, and the rest are simply not mentioned.
- No split range for ad variations, neither minimum nor maximum percentage — and one page calls the split a share of budget while another calls it a share of traffic.
- No significance flag in the variation report: the 95% confidence interval is documented for experiments, not for ad variations.
- No count of combinations in the combinations report, and no stated period they cover.
- No number inside “several weeks” — the only guidance for replacing an asset that gets zero impressions.
So the honest answer to “what should I set” is: for the vast majority of accounts, leave it alone. Under Smart Bidding the value is fixed for you anyway. On manual bidding or Maximise clicks, “Do not optimize” is technically available, but Google does not recommend it for most advertisers and the resulting data still is not a clean A/B because the ads are differently competitive. It makes deliberate sense in two cases: you need predictable auction entry for brand or legal reasons, or you run manual bids on narrow semantics and can wait months. Everywhere else the winning combination is Optimize plus a separate experiment when you need a verdict — and the 2017 announcement adds a note on volume: “using an optimized ad rotation with three or more ads per ad group can increase both clicks and impressions”. If you run paid traffic at scale, see how Google Ads agency accounts are set up and which services cover the operations around the account. More breakdowns of Google Ads mechanics are in the blog.
FAQ
How many ad rotation options are left in Google Ads in 2026?
Two: “Optimize” and “Do not optimize”. The documentation states it plainly — “There are 2 settings for ad rotation”. Options such as rotating evenly for 90 days or optimising for conversions were removed back in September 2017.
Why does one ad take almost all the impressions?
If Optimize is on, that is the design: priority goes to ads that “are expected to perform better than other ads within an ad group”. If even delivery is on, the skew comes from the fact that evenness applies to auction entry rather than impressions, plus the influence of quality, bid and competition.
Does the rotation setting work under Smart Bidding?
No — your choice is overridden: “If you’re using Smart Bidding, which prioritizes conversions, Google Ads will automatically use the ‘Optimize’ ad rotation setting.” Switching to “Do not optimize” in such a campaign changes nothing.
Which campaign types have the rotation setting?
Google names Shopping, Search and Display. Performance Max, Demand Gen, Video and App campaigns are not mentioned on that page at all, and that silence should not be read as confirmation either way.
Can I force my ads to serve in exactly equal shares?
No. “Setting ads to rotate evenly doesn’t always mean that ads will serve more evenly. It means that ads will enter the auction more evenly”, and an ad is not guaranteed to win every auction. Equal impressions are not promised under any setting.
Then how do I compare two versions of copy correctly?
Through ad variations, where users are split by cookie, or through an experiment, where significance is computed: two-tailed testing against a 95% confidence interval with twenty buckets in each arm. The recommended experiment length is “at least 4 to 6 weeks”.
How long should an ad variation run before I decide?
No specific number of days is documented for variations — only a start and an end date. The “4 to 6 weeks” figure belongs to experiments, and it is sensible to treat it as a floor for variations too, otherwise you are reading noise.
What happens to the original ads when I apply a variation?
New ads are created and the test stops: “it creates new ads and your ad variation stops running.” You choose one of three outcomes — pause the originals, remove them, or keep them. Approval of the new ads takes “up to one business day”.
Do responsive search ads have rotation?
Not as a setting. Inside an RSA the platform assembles headline and description combinations in real time, and your only levers are the asset pool and pinning, which “isn’t recommended for most advertisers and can affect ad strength”.
Can I build a static ad from the best combination in the report?
Google explicitly says not to: “The combinations report shouldn’t be used to create static versions of responsive search ads because they’re not guaranteed to perform the same as your ads.” That combination won under specific conditions, not in general.
Will raising Ad Strength get my ad serving again?
No. “It isn’t used to calculate Ad Rank, Quality Score, or auction wins”, and “Ad Strength doesn’t determine whether your ad is eligible to serve”. It is feedback on asset diversity, not an eligibility lever.
After how many impressions is a split considered normal?
No such threshold exists in the documentation — only “as data is accumulated” and the note that delivery may stay even when ads perform similarly or volume is low. Any specific figure is somebody’s experience on their own account rather than a platform rule.
