What is Agentic Commerce Optimization? ACO, defined neutrally.

ACO in one sentence

Agentic Commerce Optimization is the practice of preparing product catalogs so AI shopping agents — including agents that draft carts and complete purchases — can evaluate, compare, and recommend them. ACO emphasizes SKU-level product-data readiness over marketing-content optimization, and it overlaps heavily with AEO and GEO applied to ecommerce. That is the canonical definition published in the glossary.

The wording matters. ACO is about preparing catalogs, not writing content. Its unit of work is the SKU, the variant, the metafield, the availability flag. Its reader is an agent parsing structured data, not a human reading copy.

Who coined ACO and why

ReFiBuy introduced the term in its public materials during 2024–2025, reflecting a specific bet: that AI shopping agents would become a sales surface in their own right, requiring optimization work distinct from classical SEO, GEO, and AEO. The bet was defensible — by mid-2026, ChatGPT Shopping, Perplexity's Buy with Pro, and Google's AI Mode shopping flows had all moved from preview to live or limited release, and agent-completed checkout protocols entered merchant pilots.

eCommerce Insights credits the originator and takes no position on ReFiBuy's tooling or pricing here; the direct comparison lives at eCommerce Insights vs ReFiBuy. The term is useful as shared vocabulary whether or not a brand ever evaluates their platform. Note also that the minority "Agent Engine Optimization" reading of AEO overlaps substantially with ACO — the disambiguation is in the AEO guide.

ACO names a shift most brands have not priced yet: optimizing for buyers who are not human.

How ACO differs from GEO and AEO

GEO and AEO optimize for human-facing generative surfaces — a person reads the answer and decides. ACO optimizes for agent-facing surfaces — software reads the catalog and acts. Four practical consequences:

• The unit of optimization. GEO and AEO target pages and passages. ACO targets product records — SKUs, variants, metafields, offers blocks.
• The freshness requirement. Humans tolerate stale content; agents do not. An agent comparing three products against a budget needs current price and real availability. ACO makes freshness a first-class concern.
• The comparability surface. GEO and AEO optimize for citation; ACO optimizes for comparison. An agent compares products on weight, material, price, and lead time — attributes missing from the catalog are missing from the comparison.
• The API exposure. Some agents read public pages; increasingly they read structured feeds and protocol endpoints. ACO programs consider both paths.

Agentic commerce in three stages

The strategic point: the PDP and catalog work that wins citations in stage one is the same work that makes a SKU draftable in stage two and purchasable in stage three. Clean schema, admitted crawlers, machine-readable price and availability, discoverable policies. Brands do not have to bet on which protocol wins — the groundwork is shared. The full framing is on the agentic commerce solution page.

Why catalog data — not content — is the ACO primitive

A content-first program writes better PDPs for human readers and the engines that cite on their behalf. ACO asks a different question: when an agent reads the structured data behind a PDP, is it complete enough to act on? Most catalogs fail this test at scale. Common failures in eCommerce Insights audits as of mid-2026 (illustrative): variant-level attributes missing on 30–60 percent of SKUs, GTINs absent on long-tail products, metafield coverage concentrated on best-sellers, availability flags correct at the parent level but wrong at the variant level, and priceValidUntil set to a default far-future date that engines treat as unreliable. None of these are content problems. They are structured-data hygiene problems, and ACO is largely the discipline of fixing them.

A catalog readiness checklist

Check	What an agent needs	Where
Product JSON-LD	name, sku, gtin13, brand, image, full offers block	per SKU
Availability	Accurate at variant level, schema.org/InStock vocabulary	per variant
Price freshness	Current price, ISO currency, honest priceValidUntil	per offer
Crawler access	GPTBot, PerplexityBot, ClaudeBot, Google-Extended admitted	robots.txt
Policies	Shipping, returns, warranty discoverable as linkable pages	site-wide
Attributes	Material, size, weight, compatibility in additionalProperty / metafields	per SKU
Checkout wiring	Protocol readiness where piloted — feeds, endpoints, opt-in flags	optional · 2026

The first six rows are measurable today — they are the inputs to the agent-readability score, and the free agentic readiness grader checks any PDP against them in about 30 seconds.

The protocol layer: ACP, UCP, AP2

Three protocols define agent-completed checkout as of mid-2026, all in pilot. The Agentic Commerce Protocol (ACP), maintained by OpenAI, Stripe, and Meta, backs ChatGPT Instant Checkout — the spec is published at agenticcommerce.dev. Google's Universal Commerce Protocol (UCP) serves AI Mode and Gemini, with Shopify, Etsy, Wayfair, Target, and Walmart among launch partners. The Agent Payments Protocol (AP2) sits on the payments layer alongside both. Shopify brands mostly should not implement either protocol directly — Shopify's own Agentic Storefronts integration wraps UCP, and ACP requires a compatible PSP. Score readiness, ship the groundwork, and let the platform handle the wire format.

What to do this quarter

1. Run the readiness checklist against your top 50 revenue SKUs.
2. Fix variant-level availability and identifier gaps first — they break agent comparisons silently.
3. Verify crawler admittance and policy discoverability.
4. Baseline citation performance too: agents recommend what engines already trust. Start with the product AI visibility pillar.
5. Watch the protocol pilots; revisit checkout wiring when your platform's integration exits early access.