Skip to content
Primores AI wiki

Andromeda-Era Creative Strategy — Build a Reusable Template Library, Not Single Ads

Andromeda-Era Creative Strategy

TL;DR: Once glossary/meta-andromeda — Meta’s ad-retrieval engine — can pick the best ad out of tens of millions of candidates, the marketer’s remaining job is to supply a pool worth choosing from. The unit of work stops being the single ad and becomes a reusable, brand-aligned template library: 10-15 genuinely different concepts (format × angle × visual style), each templated across 1:1 / 4:5 / 9:16, rotated every few weeks. This is glossary/creative-is-new-targeting made engine-enforced, and it’s automation/staged-compiler-pattern applied to paid static — an expensive durable template layer feeding a cheap regenerable ad layer.

The shift in one line

The engine got good at finding the right ad; the constraint moved to producing enough distinct ones. (See glossary/meta-andromeda for the technical “why” — and for the careful attribution: Meta built the retrieval engine, agencies made the “creative is the lever” leap.)

That changes what “doing the creative” means. It is no longer “make a great ad and iterate it.” It is “stand up a library of distinct concepts the engine can choose among, and keep the library fresh.” The deliverable is a system, not an asset.

The playbook

1. Build a library, not a hero ad

Aim for 10-15 genuinely different concepts — different along format × angle × visual style, not 15 recolors of one layout. The diversity has to be real, because the engine recognizes and suppresses near-duplicates (this is the same logic behind the mandatory similarity-audit stage in marketing/prescriptive-production-briefs). The glossary/creative-formula-vs-creative-skin distinction is the test: a new formula counts as diversity; a new skin on the same formula does not.

2. Work from archetypes

A working set of static archetypes to span the concept space:

  • Price / offer — the deal, stated plainly
  • Lifestyle overlay — product in an aspirational use-context
  • Social proof / testimonial — review, rating, or quoted result
  • Minimal product — clean product-forward shot, little else
  • Seasonal / promo — time-bound hook (holiday, launch, event)
  • Reveal — curiosity or before/after structure

The archetypes are the columns; your angles (from audience research) are the rows. Filling the grid is what produces real diversity instead of accidental sameness.

3. Template across the ratios

Each concept is built once and templated across 1:1 (feed) / 4:5 (primary feed) / 9:16 (Stories, Reels, TikTok) so a single concept survives auto-placement intact. This is where marketing/meta-ad-placement-safe-zones becomes load-bearing — load-bearing copy has to live in the band that survives every placement, or auto-placement clips it.

4. Rotate on a cadence

Refresh the live set every 2-3 weeks. Creative fatigue windows have compressed (vendors quote 2-3 weeks vs. a former 6+, though those specific numbers are unverified vendor claims — treat the direction as real, the magnitude as untested). A library makes rotation cheap: pull the fatigued concepts, push fresh ones from the same template system.

Why this is the staged-compiler pattern

The library splits cleanly into the two layers of automation/staged-compiler-pattern:

LayerWhat it isEconomics
Template libraryThe brand-aligned concept system (archetypes, formulas, render templates)Expensive, durable, compounding — the part worth investing in
Each adA skin/variant rendered from a templateCheap, regenerable — recompile, don’t hand-craft

If an ad underperforms, you regenerate it from the template. If the template is wrong, that’s the expensive event — it goes back through the durable layer. Same asymmetry as source vs. build artifact. And the way you actually fill the library at volume is glossary/human-anchored-ai-multiplication: anchor on premium human-shot input, let AI multiply formats and skins around it — which keeps the output on the safe side of the AI-recognizability penalty.

Dogfooding it

We run this on our own paid-social work. A breadth-set wave — multiple distinct skins across several hooks, structured as a formula-vs-skin matrix and rendered from a single build spec — is itself an Andromeda-aligned creative pack: the spec is the reusable template system, each rendered card is a regenerable ad. Productionizing it as an HTML/CSS template library rendered headlessly turns “make ads” into “run the library,” which is exactly the shift this page argues for.

Honest scope

  • The engine is Meta-primary and confirmed. glossary/meta-andromeda is documented by Meta; the retrieval-stage mechanics are not in dispute.
  • The creative-strategy specifics are agency/directional. Archetype lists, the “10-15 concepts,” and the 2-3 week rotation are practitioner consensus, not platform doctrine. Useful as a starting structure, not a guarantee.
  • No performance guarantees. Every circulating ROAS/lift number (e.g. “7% conversion lift from AI images,” “22% ROAS lift,” “65% higher ROAS at 20+ ads/month”) is vendor-reported and unverified — see the warning box in glossary/meta-andromeda. The defensible claim is structural (the lever moved to creative volume/diversity), not numerical.

Key Takeaways

  • The unit of work is the reusable template library, not the single ad — because the retrieval engine chooses from a huge pool and rewards real diversity.
  • Span the concept space with archetypes × angles; ensure diversity at the formula level, since near-duplicates get suppressed.
  • Template each concept across 1:1 / 4:5 / 9:16 and keep load-bearing copy in the placement-safe band; rotate the live set every 2-3 weeks.
  • It’s the staged-compiler split applied to paid static: durable template layer (invest here) feeding a cheap regenerable ad layer (recompile, don’t hand-craft).
  • Treat the engine as confirmed and the strategy specifics — and all the lift numbers — as directional vendor color.

Sources

By Andrej Ruckij