An asset group is the creative unit of a Performance Max campaign: a themed collection of headlines, descriptions, images, logos and videos, pointed at a final URL and seeded with an audience signal. Google assembles ads from those parts for every surface it owns — Search, YouTube, Display, Gmail, Discover, Maps — sizing and combining assets per placement.
It occupies the slot the ad group holds in a Search campaign, and the comparison is the fastest way to understand it: the ad group is a targeting container, the asset group is a story container.
How asset groups work
| Ad group (Search) | Asset group (PMax) | |
|---|---|---|
| Targeting | Keywords you chose | Audience signal plus search themes — a starting point, not a boundary |
| Creative | Ads served as you wrote them | Assets Google assembles differently per placement |
| Placements | Search results | Every Google channel |
| Reporting | Per-keyword, per-ad | Asset-group totals plus per-asset ratings |
Inside one group you supply text assets, images and logos in several ratios, and video — supply none and Google generates video from your other assets, which is rarely the version of your brand you would have chosen. Search themes can be added to steer the system toward query territory the signal alone would miss.
The structural decision that matters is what "one theme" means. The working pattern is to structure by audience story, not product taxonomy: one group per distinct buyer and message — "first-time buyers, discount frame" versus "returning customers, new-range frame" — rather than one group per catalogue branch. Google assembles creative per group, so a group is only coherent if every asset in it could sensibly appear together.
Reading performance happens at two levels. Asset-group columns give you cost, conversions and value, which is enough to compare stories against each other. Individual assets get graded ratings rather than full metrics, so the honest unit of judgement is the group.
Building asset groups by prompt, arriving paused
Building asset groups is exactly the kind of structured, multi-part work an AI agent does well through a connector: it can draft the campaign, the asset groups and the full asset set — headlines, descriptions, images specifications, sitelinks — as one reviewable proposal, created paused so nothing spends until you have looked at it. The useful division of labour: you decide the stories, the agent produces the parts and pushes them once approved.
Review is the other half. PMax changes behaviour quietly, and asset-group totals drift for reasons the interface does not volunteer. A weekly PMax review run by an agent — spend and conversion by asset group, week over week, with anomalies flagged — turns an opaque campaign type into a monitored one. Ask for the comparison in plain terms: which story earned its budget this week, which did not, and what changed.
Be equally clear about what no tool can see. Per-asset conversion attribution inside PMax is limited by the platform itself, and an agent that claims otherwise is inventing data. The honest reading stops at the asset group — which is precisely why the groups should be structured as stories worth judging.
Asset groups inherit their audience logic from signals — defined in audience signals — and sit inside the wider question of running Performance Max with AI.