What is Asset group?

An asset group is where the creative lives in a Google Ads Performance Max campaign. It holds headlines, long headlines, descriptions, images in several shapes, logos, videos, a business name, a final URL, a call to action, and an audience signal.

From those components the platform assembles ads and places them across every surface the campaign type reaches — search results, shopping, display inventory, video, discovery feeds, maps and email surfaces — choosing the arrangement and the destination surface per opportunity.

It is the same pattern as dynamic creative and responsive search ads, extended to span surfaces rather than to vary within one. The advertiser supplies range; the platform supplies arrangement and placement.

It is not an ad group. An ad group organises keywords and the ads serving against them, on one surface, with the advertiser deciding which ad shows for which query. An asset group has no keywords, spans surfaces, and hands both decisions to the platform.

It is not a placement control. An advertiser cannot say that a given image is for display and a given video is for the video surface. The assembly decides.

It is not a targeting boundary. The audience signal inside it is a starting point, not a fence.

It is not a unit of testing in the usual sense. Two asset groups in the same campaign are not an experiment: the platform allocates between them on its own prediction, so the comparison is contaminated before it starts.

The decision is the same one that governs every pooled-allocation arrangement: things go in the same container when you want the platform to choose between them.

One asset group per genuinely distinct offering. A different product line, a different service, a different customer with a different problem — each needs its own creative and its own destination, so each needs its own group.

Not one per small variation. Splitting by colourway, by city or by a minor audience difference divides the conversion signal across groups that each need volume, and slows the whole campaign's learning without buying control that exists.

Not one group for everything. A single group covering products that need different messages produces assembled ads that fit none of them, because the components were written for different things and can be combined freely.

The test is whether the components in a group would make sense in any combination. Where they would not, the group is holding two campaigns.

Everything that applies to assembled creative applies here with the additional burden of covering several surfaces.

Independence. Any headline can appear with any image, so nothing may refer to anything else.

Every shape. Landscape, square and portrait images, and video in the orientations the surfaces use. Where video is not supplied, the platform will generate something from the other assets, and an automatically generated video is a fallback rather than an asset — supplying real video is one of the clearer improvements available.

Variety of argument. Different problems named, different objections answered, different proof offered. Rephrasings give the assembly nothing to choose between.

A real audience signal. Customer lists and converters are the strongest seeds; inferred segments contribute little. The signal matters most in the first weeks, before the campaign has conversions of its own.

Correct final URLs. The destination is part of the creative, and a group pointing at a generic page gives up the relevance its specific components were for.

The reporting is thinner than a conventional campaign's and it is not empty, and the difference between an account that learns from Performance Max and one that does not is usually how carefully this is read.

Asset performance ratings say which components the platform is using and which it rates poorly. A consistently low-rated asset is a candidate for replacement — with a different argument, not a rewording.

Search themes and search terms, where exposed, say what queries the campaign is answering. This is where an account discovers it is buying its own brand name at scale, or reaching a category it did not intend.

Placement and surface breakdowns say where the money went, at whatever granularity the platform offers. This is where cheap inventory with no outcomes becomes visible.

Exclusions still work. Brand-suitability exclusions, negative keywords at the account level where supported, and audience exclusions remain the advertiser's reliable instruments.

The most common complaint about this campaign type is that it absorbs budget and credits itself with conversions the account was getting anyway — branded search in particular.

The complaint is frequently justified and it is not a creative problem. The remedies sit outside the asset group: separating brand into its own campaign so the comparison is honest, excluding the brand terms where the platform permits it, and measuring incrementally rather than by attribution.

What the asset group controls is whether the creative is any good, which components the assembly has to work with, and whether the destination supports what the ad promised. Those are worth doing well, and they do not settle the incrementality question, which needs a test rather than a report.

Because the campaign type exposes less than a conventional one, the work is scheduled rather than triggered — nothing will alert an account that its creative has gone stale or that its delivery has drifted.

Read the asset ratings monthly and replace the weakest with different arguments rather than rewordings.

Read the search terms and themes for categories the business does not serve and for brand traffic the campaign is absorbing.

Read the placement breakdown for inventory that consumes spend and produces nothing.

Refresh the audience signal as the customer base changes, since a seed built from last year's customers describes last year's business.

Check the final URLs still resolve to the pages the components were written for, which is the failure that arrives silently after a site change.

Re-supply video when the library dates, rather than leaving the platform to generate something from stills.