How many asset groups should a Performance Max campaign have?

As many as your conversion volume can support and no more. Each group needs enough conversions inside your review window to be readable: a group producing 30 conversions in a 30 day month gives you 30 ÷ 30 = 1 per day, which is already thin for a weekly review. Start with one group and split only when the reporting gives you a concrete reason to.

Can I set a different ROAS target for a single asset group?

No. Bidding and targets are campaign level settings in Performance Max. Asset groups differ in creative, final URL, audience signal and listing group only. If a product set genuinely needs its own target, it needs its own campaign.

What happens if an asset group has no video?

Google has generally assembled a basic video from the group other assets so video inventory can still be served. That fills the requirement without being a deliberate piece of creative, so upload your own if the placement matters to your brand. Check the current requirement in the interface, since asset specifications change.

Do listing groups work the same way as Shopping product groups?

The subdivision logic is familiar: you split by category, brand, product type, item ID or custom label, and each split leaves an everything else node. The difference is the consequence. In Performance Max the listing group also decides which creative set represents those products, so an unplanned everything else node means the wrong message on real inventory.

Is it worth restructuring asset groups in a campaign that is already running?

Only with a reason you can state in a sentence. Editing an asset group assets or its listing group restarts learning on that group, so a restructure has a real cost. Do it when the current split cannot answer a question you need answered, not because a wider structure looks more sophisticated.

Performance Max asset groups explained

Performance Max asset groups are the level inside a PMax campaign where your creative actually lives: one asset group holds one set of headlines, descriptions, images, logos and video, and in a retail campaign it also holds the slice of the product feed those assets are allowed to advertise. Budget, bid strategy and target ROAS sit above it, at campaign level. That division is the whole reason the object is worth understanding, because it tells you exactly what you can and cannot control here.

What the asset group controls and what it does not

A PMax campaign has two working levels. The campaign carries budget, bidding, the ROAS or CPA target, locations, languages, the schedule and campaign level exclusions. The asset group carries creative, the final URL, the audience signal and, when the campaign is attached to a Merchant Center feed, a listing group that decides which products belong to it.

Two consequences follow, and most structural mistakes come from missing them:

You cannot bid differently per asset group. Pulling your best sellers into their own group does not give them their own ROAS target. A product set that genuinely needs a different target needs a different campaign.
You can report differently per asset group. Cost, conversions and conversion value break out per group, so the split you choose today becomes the reporting axis you live with for the rest of that campaign life.

So the test before creating a group is not only whether those products deserve their own creative. It is whether you will read that row separately, and act differently when you do.

How many Performance Max asset groups a campaign needs

Start from your conversion volume, not from the shape of your catalogue. Group level reporting is only readable if each group accumulates enough conversions inside the window you actually review.

Run it with your own figures. Suppose the campaign records 300 conversions in a 30 day month. If you split it three ways and the split were perfectly even, that is 300 ÷ 3 = 100 conversions per group. Real splits are never even; a common shape is 180 conversions in the group holding the hero category, 90 in the second and 30 in the tail, which still sums to 300. The top group is producing 180 ÷ 30 = 6 times the conversions of the tail group.

Now test whether that tail row is readable. 30 conversions across the month is 30 ÷ 30 = 1 conversion per day, so a seven day review window holds 1 × 7 = 7 conversions on average. You can see a total collapse at that volume. You cannot separate a mild dip from ordinary variation. Split the same campaign six ways instead and the tail halves again: 15 conversions a month is 15 ÷ 30 = 0.5 per day, and a week of review contains 0.5 × 7 = 3.5 conversions. The row still exists, but it can no longer answer a question.

That gives a working rule instead of a number pulled from the air: create a group when it will still carry enough conversions to be read at your review cadence, and when the split maps to something you would genuinely handle differently. In practice that argues for few groups per campaign, and for a new campaign starting with one.

Listing groups: deciding which products a group can advertise

In a retail campaign the listing group is what ties the asset group to inventory. It behaves like the product group tree in Shopping: you subdivide by category, brand, product type, item ID or the custom labels on your feed, and every subdivision leaves an "everything else" node behind.

Three things matter more than the mechanics of clicking through that tree:

The "everything else" node is where products hide. Anything you did not explicitly place lands there. If that node sits in a group whose copy and imagery were written for one category, you are advertising the rest of the catalogue with the wrong message, out of the same budget.
Your split rule has to exist in the feed first. If you want to separate by margin band, season or stock depth, that distinction must already be a value on the product before it can become a listing group. Custom label fields are what most retailers use for it, and a stale or inconsistent feed will undermine any structure built on top of it, which is why feed quality work belongs before structure work.
Keep the split clean across campaigns. Letting the same products sit in more than one Performance Max campaign makes both serving and reporting harder to reason about. Before relying on any particular behaviour there, check Google official documentation for how the current version resolves that overlap.

The asset side: what each group has to carry

Every group needs its own text, images, logos and, ideally, video. The interface states the required minimum for each asset type and will not let you publish a group that is short of it. Rather than memorising counts that change between product updates, read the counter in the editor while you build, and treat the ad strength indicator as a completeness check rather than a performance forecast: it reflects how much raw material the system has to assemble combinations from, not whether those combinations will sell anything.

Video is where groups are thinnest. Google has generally assembled a basic video from your other assets when you supply none, which fills the slot without representing the brand deliberately. If video inventory matters to you, upload something you chose. Producing it is the recurring cost of a wide structure: one group refreshed quarterly is four creative cycles over twelve months, and five groups on the same cadence is 5 × 4 = 20 cycles in that same year. That ongoing creative maintenance load, not the initial build, usually decides how many groups an account can sustain.

What one catch all group costs you later

The default build is one campaign, one asset group, the whole feed, one set of assets. It is fast, it works, and it is the reason so many PMax campaigns become impossible to diagnose six months in.

With a single group, the asset group table has one row. Channel and placement opacity is a known limitation of the campaign type and a subject of its own; the part people miss is that their own structure is the other half of the visibility problem, and that half is within their control. Per asset performance ratings survive, so some creative signal remains. What does not survive is any way to tell whether accessories are carrying the account return or dragging it, because no row isolates them.

Once the rows exist, comparing them is simple: divide the conversion value a group generated by the cost it accumulated in the same window, and read the groups against each other rather than against a target you invented. If you want to set that internal threshold first, work it through with a ROAS calculator before touching the campaign.

A build order that survives the first month

Structure decisions are cheap before launch and expensive afterwards, because editing an asset group listing group or its assets restarts the learning the campaign has done on that group. A sequence that holds up:

Write the split rule in one sentence before opening the campaign builder. If you cannot state it, you do not have a rule yet and your groups will overlap.
Name each group after the rule rather than after the creative. "High margin" tells you what the row means in three months; "Summer visuals v2" does not.
Give the tail an explicit home. Decide which group owns "everything else" instead of discovering later that it was the one built for your flagship line.
Fill every asset slot the editor asks for, then stop. Padding a group with weak variants to chase a better strength rating just adds combinations you would not approve individually.
Hold the structure through the learning phase. Judge the split on the reporting it produces, not on how it looks in the builder.

Layering automation on top of a Performance Max campaign is a separate discussion, and one covered in our piece on AI for Performance Max campaigns. No automation layer chooses your asset groups for you. That decision is human, it is made once and lived with for a long time, and it sets the limit on what every report you read afterwards can tell you.