Two decades of buying sequence, intercepted.
Retailers and brands have spent two decades optimizing for a buying sequence that AI shopping agents have now intercepted.
Search returned results. Results drove clicks. Clicks drove revenue. Every layer of the stack, from product feeds to media spend to merchandising, was built around that sequence.
Agentic shopping engines have collapsed it. They interpret product data directly, evaluate options against each other, and decide which products qualify before a shopper ever reaches a results page. The decision that used to happen on the page now happens before the page exists.
This is not happening on a single platform. It is happening simultaneously across seven agentic shopping engines: ChatGPT, Gemini, Perplexity, Meta.ai, Copilot, Grok, and Claude. Each engine maintains its own catalog. Each runs its own evaluation logic. Each is making inclusion decisions about your products right now.
Optimization built for the old sequence cannot reach where decisions now get made. That gap is what ACO addresses.
Every agentic interaction reduces to three components.
ChatGPT looks nothing like Perplexity. Gemini looks nothing like Claude. The underlying structure of how a recommendation gets made does not vary. Every agent identifies, compares, weighs context, and assesses trust. And every agent does that work against three structural components.
The Prompt. A problem, not a keyword.
The consumer's conversational query. Multi-turn, often 10 to 30 words, rarely repeated the same way twice. A shopper does not type "running shoes." A shopper describes a problem, a context, or a goal. There is no keyword to rank for because there is no single keyword in the prompt.
"I need running shoes for marathon training that won't aggravate my plantar fasciitis."
The Product Card. The unit of competition.
The canonical representation of a product, assembled by the engine from product data, structured attributes, content, and cross-source signals. The Product Card is not your PDP, not your feed, not your catalog record. Those are inputs. The Product Card is what the engine constructs from those inputs, in its own catalog, on its own logic. If your product is not on the right Product Card, the engine cannot include it.
Brooks Glycerin GTS 21 · Stability, max cushion · 10mm drop · 37/27 stack · High arch, plantar fasciitis relief, long distance · Strong match
The Offer Card. Where the merchant wins.
The ranked list of merchant offers attached to a Product Card. When the engine has decided which Product Card answers the prompt, it then has to decide which merchant gets the recommendation. Whoever ranks first owns the Product Card. Owning the Product Card is the agentic equivalent of winning the buy box on Amazon, except this dynamic now plays out across every major agentic shopping engine simultaneously.
1. Road Runner Sports, $164.95 (winning) · 2. Fleet Feet, $169.95 · 3. Brooks Direct, $169.95 · 4. Zappos, $174.99
Two metrics decide whether your catalog competes.
Coverage answers the first question: are your products eligible for inclusion? Ownership answers the second: are they winning?
Product Card Coverage
The percentage of your catalog correctly mapped to canonical Product Cards across agentic shopping engines. Governed by canonicalization, the upstream process that resolves each product to the right card inside each engine's catalog. A representative pattern in the field today: 20% coverage, meaning 10,000 of 50,000 SKUs eligible for inclusion.
Product Card Ownership
The percentage of Product Cards where your offer ranks first on the Offer Card. Coverage gets you on the card. Ownership decides whether you gain the recommendation or lose it to a competitor. A representative pattern: 40% ownership on a 20% coverage baseline means winning 4,000 of 10,000 mapped cards, but you are still invisible on the other 40,000 SKUs entirely.
Canonicalization is where coverage breaks. Inconsistent identifiers, fragmented variants, updated model years, and missing structured data all create resolution problems that prevent a product from being mapped at all. When canonicalization fails, the product never enters the consideration set. Ownership is a downstream problem. Canonicalization is upstream of everything.
Defining ACO.
ACO is the discipline of making products understandable, eligible, and competitive in AI-powered shopping environments.
In practice, ACO prepares product catalogs for the way AI shopping agents evaluate, compare, and recommend products.
ACO is a discipline ReFiBuy named and defined. The foundational concepts, canonicalization, owning the Product Card, content and context, and the sequential readiness system, were first laid out by Scot Wingo in the foundational ACO post on Retailgentic. What follows in the ACO Guide builds on that foundation.
ACO runs at three planes simultaneously.
ACO operates upstream of ranking, paid placement, and on-site experience. It is not a layer on top of existing optimization work. It is the foundation that work depends on.
At the SKU level
Product Cards are constructed at the SKU level. Coverage and Ownership are decided one product at a time, by how well each product's data resolves, structures, and contextualizes inside each engine's catalog. Homepages and category pages do not move these metrics.
Across every engine
Each agentic shopping engine maintains its own catalog, its own evaluation logic, and its own Offer Card ranking criteria. A product that achieves coverage on ChatGPT may not be mapped on Gemini. A Product Card you own on Perplexity may rank second on Copilot. Seven engines. Seven competitive environments.
Continuously
Engine catalogs shift. Evaluation criteria evolve. The data that earned coverage last quarter may not earn it next quarter. ACO is governed the way infrastructure is governed: with ownership, monitoring, and continuous correction. Not a one-time project plan.
SEO at the page. GEO at the answer. ACO at the product.
Three disciplines, three layers, all running in parallel. ACO is the layer most teams do not currently own.
- SEO. Page-level, downstream. Optimizes pages for ranking in search results. Inherits the consideration set from the search engine. Unit: the page.
- GEO. Content-level, downstream. Optimizes content for inclusion and citation in AI-generated answers. Inherits the consideration set from the answer engine. Unit: the content asset.
- ACO. Commerce infrastructure, upstream. Governs whether your products qualify for inclusion in the first place. Operates at the decision, not after it. Unit: the product (SKU).
ACO governs what SEO and GEO assume.
SEO and GEO both compete inside the engine's consideration set. ACO governs whether the product enters it at all.
A retailer can run the strongest SEO program in their category and still have most of their catalog invisible to agentic shopping engines, because the Product Cards the engines build are incomplete. The same brand can run a sophisticated GEO strategy and still lose recommendations to a competitor with better-structured product data. Neither discipline was built to address inclusion.
SEO was built for a search results page that ranks indexed pages. GEO was built for an AI answer surface that cites indexed content. Both operate on the assumption that indexing is somebody else's problem. In agentic commerce, indexing is the problem. ACO is what governs it.
No team in the org chart was built to own this.
SEO lives in marketing. GEO is landing in content or brand. ACO does not have a natural home in legacy org charts, because the unit it operates on cuts across four functions at once.
- Marketing owns brand messaging.
- Ecommerce owns the product page.
- Data Engineering owns the product feed.
- Merchandising owns the assortment.
- Product-level eligibility: owner unassigned.
What started as a ReFiBuy framework is now industry consensus.
Major advisory firms have published findings that validate the layered stack and the upstream nature of ACO. The framing is no longer just ours.
Deloitte
Agentic commerce moves the discipline from visibility to action. Eligibility, accuracy, and execution converge inside agent workflows. (Source: Agentic Commerce: Redefining Retail Economics, December 2025.)
McKinsey / QuantumBlack
Traditional SEO is becoming less relevant in an agentic environment. The optimization stack is being redrawn. (Source: The Agentic Commerce Opportunity, October 2025.)