Most ecommerce creative briefs still ask for a finished set: several banners, a handful of videos, and the usual size adaptations. Amazon’s reimagined Responsive eCommerce Creative (REC), announced on September 30, 2026, changes the unit that matters. The system can assemble ads from product listings, Brand Store content, campaign data, and uploaded or generated images and videos, then adapt the mix to audience signals, objectives, and placements.
This is more than another ad-format update. If the platform chooses which image, clip, product fact, offer, or conversational prompt appears in a given impression, should brands still spend most of their production budget on one locked banner set?
ONCE’s view: finished films still matter, but they should no longer be the only unit of procurement for responsive commerce media. Brands need a governed component system in which every asset has a job, a source of truth, defined transformation rights, safe combinations, and an accountable approver.
What Amazon has actually changed
Amazon’s September 30 launch announcement introduced in-banner video, multi-image carousels, Alexa for Shopping prompts, and placements on Alexa+ and Fire Tablet. Carousels can display three to six products. Prompts may be generated from product detail pages, Brand Store content, and campaign data. Amazon’s current REC specification adds that an ad can contain up to 20 ASINs, skip products that are out of stock, and pair an image or video section with an ecommerce section containing live prices, badges, coupons, ratings, or Prime information.
The evidence does not prove that the entire advertising industry has become autonomous. It does show that Amazon is joining product data, brand content, media assets, and delivery logic inside the same assembly system. Product titles, claims, imagery, Store structure, and ASIN relationships are therefore becoming upstream creative inputs rather than back-office catalog details.
That connects directly to our earlier argument about building a product-family map that keeps AI from confusing related SKUs. Automation does not repair ambiguous product truth; it distributes it faster.
The new failure mode: correct assets, incorrect combinations
An image, a feature clip, and a promotional line can each be accurate in isolation while becoming misleading together. A lifestyle image may show Model A while the price belongs to Model B. A “waterproof” demonstration can appear beside an ASIN rated only for splash resistance. A premium brand film may be cropped until only an unexplained detail remains. A claim approved in one market can surface in another where it is not permitted.
Amazon now offers controls for logos, color palettes, approved imagery, and copy guardrails, plus previews filtered by ASIN, placement, size, language, and asset. That is useful, but it also makes the governance requirement visible: uploading assets is not the same as approving every combination the system can create.
There is also a scope discrepancy buyers should not ignore. The September 30 announcement lists in-banner video across multiple locales, while Amazon’s current US and UK REC specification pages label video and conversational prompts as US-only. Until Amazon reconciles those pages, do not build a multi-market production budget around the broader list. Verify availability in the target account, for the target ASINs and placements, before committing spend.
The ONCE Responsive Commerce Creative BOM
A production brief for REC—or any responsive ad system—should specify at least seven fields:
| Field | Procurement and approval question |
|---|---|
| Product truth source | Which ASIN, listing version, and approved specification does this component represent, and who maintains it? |
| Component job | Does it establish identity, show context, prove a feature, enable comparison, or explain an offer? |
| Transformation rights | Must it remain original, or may it be cropped, resized, animated, or generatively modified? |
| Safe combinations | Which SKUs, claims, and scenes may appear together, and which pairings are forbidden? |
| Surface and locale | Which on-Amazon, off-Amazon, Fire, or Alexa placements, sizes, languages, and markets are approved? |
| Fallback order | What replaces the asset when video, AI enhancement, or prompts are unavailable? |
| Preview and owner | Which ASIN × placement × language states must be checked, and who approves and refreshes them? |
The BOM is not paperwork for its own sake. It separates the platform’s ability to generate a combination from the brand’s permission to publish it. Amazon may own the assembly engine; the brand still needs to own facts, rights, and approval.
Four production routes
1. Reuse listing assets
Reuse is reasonable when the listing is accurate, existing images are clean, videos are already separated by feature, and the product has no high-risk claims. The production task becomes metadata, permissions, fallback logic, and preview coverage rather than a new shoot.
2. Commission a modular proof shoot
Choose this route when the listing has pack shots but lacks operation, scale, material behavior, or outcome evidence. Do not commission only a 30-second edit. Capture independent close-ups, actions, results, and environmental context. Each clip should remain truthful when silent, cropped, or placed before and after different components.
3. Disable AI enhancement or restrict transformations
Use tighter rules when color, proportions, construction, safety language, health claims, trademarks, or legal disclosures must remain exact. Amazon notes that its AI opt-in may create variations an advertiser did not supply and that images may still be resized even when AI enhancement is off. “Do not alter” therefore needs a practical crop-safe zone and minimum legibility standard, not just a checkbox.
4. Keep a custom finished film
Launch narratives, emotional storytelling, complex demonstrations, and sequences whose meaning depends on order should remain authored. Responsive components can cover conversion and placement breadth; a finished film can carry meaning and brand memory. They are different jobs.
Measure cost per verified usable combination
Compare production options with a denominator that reflects what can actually run:
Cost per verified usable combination = (product-truth cleanup + modular production + platform configuration + preview and QA) ÷ combinations that pass product-accuracy, brand, locale, and destination checks.
A shoot that returns 40 clips but only eight safe combinations is not necessarily efficient. Twenty traceable, crop-safe components with clear permissions may create more usable inventory. This formula does not claim an industry benchmark; it simply forces hidden rework and mismatch costs into the buying decision.
A controlled way to start
Do not begin with the whole catalog. Select one product family, one priority locale, and one campaign objective. Reconcile the product truth first, complete the BOM for every component, and preview the core ASIN, placement, language, and fallback states. Scale only when an error can be traced to a field, owner, and revision action.
Google’s own responsive display guidance makes the cross-platform principle explicit: assets are combined without human assistance, so they must work together across many configurations. Amazon’s move brings that principle closer to live commerce data and conversational interfaces.
The durable advantage for a production partner will not be “making more assets.” It will be connecting truthful capture, product data, transformation permissions, and combination learning into an operating system. The buyer’s question changes with it: not merely “Does this ad look good?” but “Can these components be used correctly a thousand times?”
Sources and evidence boundary
Verified facts in this article come from Amazon Ads’ “Scale premium display ads with Responsive eCommerce Creative” (September 30, 2026), Amazon’s current REC specifications, Amazon’s March 25, 2026 prompts announcement, Google Ads Help’s responsive-display best practices, and Marketing Dive’s July 6, 2026 coverage of the Cuisinart Creative Agent pilot. The BOM, four production routes, and cost formula are ONCE analysis, not platform guarantees. We have intentionally not repeated vendor-reported performance percentages as independent industry evidence.
