A full Google Ads asset pack — text and images — assembled from two deliberate passes
A pet-food e-commerce retailer needed a complete Google Ads asset pack for a campaign asset group: responsive search ad text and extensions plus image creatives. AOS assembled it from two passes — text generation, then image generation — each a proven capability in its own right. The caveat is honest: it is composed from two skills run separately, with no single unified asset-pack skill yet.
01 Ground
A modern Google Ads asset group is not one artefact but a pack: responsive search ad text and extensions on one side, image creatives on the other. For a pet-supplies retailer launching a campaign, all of it is needed before the asset group is complete — text alone leaves the visual slots empty, and images alone have nothing to caption. Producing the full set is the gating step between 'we want to advertise this' and 'we have something to put live.'
Each half of the pack is a different discipline. The text side means clustering intent, writing headlines and descriptions tuned to the query, and drafting extensions — a copy and structure problem. The image side means generating on-brand creative that fills the asset group's visual slots — a rendering problem. Asking one undifferentiated process to do both well is how you get mediocre text and off-brand images, or a brittle pipeline where a failure in one half sinks the other.
The retailer's real need was an assembled, complete pack — not a pile of headlines and a separate pile of pictures, but a coherent set covering both. The honest engineering question is whether to wait for a single skill that does everything, or to compose the pack now from two capabilities that each already work well. This use case is the second answer, and it is presented as exactly that: a composed example, not a one-button skill.
- RSA text and extensions assembled from a proven text-generation pass (cf. category RSA build, run cab5a8ec) AOS text-generation capability 80% confident
- Image creatives assembled from a proven image-generation pass (cf. hero-image render to Drive, run 36b26d22) AOS image-generation capability 80% confident
02 Decide
The decision was to compose the full asset pack from two separate, individually proven passes rather than to wait on a single unified skill. The text pass produces the RSA copy and extensions; the image pass produces the creatives; the pack is the assembly of the two. Each underlying capability is one that AOS already runs reliably on its own, so composing them is building on verified parts, not on a new untested mechanism.
This is a deliberately honest architectural stance. The convenient story would be a single 'make me an asset pack' skill, and there isn't one yet. Rather than pretend otherwise or block the deliverable on building it, AOS treats the pack as a composition of two skills run separately — which gets the retailer a complete asset group now, while keeping each capability simple and independently debuggable. The cost of the honesty is that the composition is manual rather than one-click; the benefit is that nothing in the pack rests on an unproven combined pipeline.
03 Deploy
The pack was built in two deliberate passes. First, a text-generation pass produced the responsive search ad copy and extensions for the asset group — the same kind of category-tuned RSA construction AOS performs as a standalone capability. Second, an image-generation pass produced the creatives to fill the asset group's visual slots — the same dedicated image capability used to render and deliver finished assets elsewhere.
The two outputs were then assembled into a single asset pack covering both text and image. There is no unified skill orchestrating this end-to-end; the composition is intentional and stated plainly. Verification therefore rests on the components — the RSA build and the image build each work and are each provable on their own — rather than on a combined run, which is why this is documented as a composed example rather than a single sourced run.