Asset groups in Performance Max: how to structure them correctly
Asset groups are structured by product economics, not by assortment: one group for high-margin items, a separate one for low-margin ones, and a third for seasonal or sale items. In Performance Max the asset group is the unit that replaces the ad group: it gathers the texts, images, videos and products between which the algorithm chooses. The minimum the platform requires is 3 headlines, 1 long headline and 2 descriptions, but the minimum is rarely enough for meaningful learning. Audience signals are a hint for the algorithm, not a restriction: they do not narrow delivery. The most common mistake is a single asset group for the whole assortment, in which high- and low-margin items share one budget and one target.
My name is Hristiyan Panov — a performance marketing consultant with over 5 years’ experience and over €2 000 000 managed ad budget for more than 90 companies on Google and Meta. Below is what to split by and what each group needs to contain.
What is the basis for splitting?
By margin and by objective, not by the categories in the website menu. Splitting by category looks tidy but solves nothing — Google chooses inside the group anyway.
Splits that work:
- High versus low margin. It allows different return targets.
- Bestsellers versus the rest. It protects the budget from being spread across items with no history.
- New customers versus all. When you have an acquisition goal, not a goal for total revenue.
- Seasonal versus year-round. Different periods, different budgets.
How many assets are needed?
Enough variety for the algorithm to have something to choose from — several headlines with a different angle, several descriptions, images in different ratios and at least one video. If you do not upload a video, Google will generate its own from your assets, and it is usually weaker than one made on purpose.
Variety is measured in messages, not in the number of files. Five headlines that say the same thing are one headline.
Are audience signals targeting?
No. They are a hint about where the search should start, not a restriction. Google will go outside them if it finds a result elsewhere.
That is why it makes sense to order them by the value of the data inside them. First your own lists: buyers from recent months, people who started checkout, visitors to product pages. After them a custom segment, described by the real queries through which people find you and by the URLs of the competitor pages they view. Demographics and general interests come last, because they carry the least information about intent.
That the signal really is only a hint can be seen in the report: compare which segments the conversions come from with what you supplied. A mismatch is normal and in itself is not a reason to remove the signal.
This is also the reason not to rely on them for exclusion. If you do not want brand traffic in Performance Max, this is done with negative keywords, not with signals. The same applies to unwanted pages — they are stopped with a list of excluded URLs at campaign level.
How do you read the report when it is so closed?
In three places: the asset group report, the combinations that are actually shown, and the search terms report at campaign level. They are not as detailed as in Search, but they show whether the algorithm has headed where you wanted.
The important number, however, is not there. It is the share of new customers — see how to measure whether PMax really brings growth.
When is an asset group too small?
When it does not gather enough conversions to exit learning. Then the split has done harm — it is better to merge and split again when volume grows.
The signs are visible before that. Impressions gather in a handful of items, while the rest of the product set sit with no impressions. The cost per conversion jumps up and down between weeks for no reason in the account. The status goes back into learning after almost every change, because there is not enough data to stabilise it.
The check is simple: open the group report and see how many products actually receive impressions versus how many are in the set. If only a handful are working, the group has no life of its own.
When merging, merge with the group with the closest economics, not the largest — otherwise the low-margin items again run against the same target as the high-margin ones. And when you split again, do it by one criterion at a time.
A practical rule of thumb: do not create a new asset group if you cannot give it enough budget and products to live on its own.
What changes when assets are added?
Changes to assets usually do not restart learning as strongly as changes to budget or bidding. That is why refreshing the creatives is the safer way to revive a group that is getting tired.
Ordered by risk, the changes look like this: the heaviest are the objective, the bidding strategy and the budget; after them comes the composition of the product set; the lightest is adding a new asset.
That is why you should add rather than replace everything at once. Leave the working headlines and images in place and launch the new ones alongside them — that way you have a baseline for comparison and the algorithm does not lose its history. Removing an asset with accumulated data is the riskier of the two directions.
Which to remove is read in the asset report: which combinations are actually shown and which elements barely enter them. Make one change at a time and wait before touching the next, otherwise there is no way to separate the effect of the creative from the effect of the budget.
FAQ — Frequently asked questions about asset groups
How many asset groups in one campaign?
As many as you can feed with budget and products. Two well-fed ones beat six hungry ones.
Does each need a separate feed?
No, but a separate product set is needed — this is the mechanism that determines which items enter the group.
Can I exclude products?
Yes, through product sets. Items outside the set are not shown by this group.
What about pages I do not want advertised?
There is a list of excluded URLs at campaign level. It is useful for pages such as “terms” and “careers”.
How long should I wait before judging?
At least long enough for it to exit learning — usually a few weeks at sufficient volume.
Conclusion
Performance Max does not give you the levers you are used to, but asset groups remain real control. How you split them determines where the money goes — and that is a decision the algorithm does not make for you.
If PMax is spending and you are not sure on what, start with an independent review.
Request a quote for a digital audit →
Related articles: Performance Max · Google Shopping · disapproved items
