Agentic commerce · 2026

The funnel grew an agent. Make every product agent-readable.

AI engines research products before shoppers do. Pre-purchase agents already draft carts in ChatGPT Shopping and Perplexity's Buy with Pro. Agent-completed checkout — ACP, UCP, AP2 — is in pilot as of mid-2026. Whichever protocol wins, the PDP groundwork is the same. Start there.

3 stagesno protocol bettingper-product readiness

The three stages

Research is mainstream. Carts are live. Checkout is in pilot.

Agentic commerce is arriving in order, and each stage reads the same product data. The strategic point: you don't have to bet on which checkout protocol wins to start the work that all of them assume.

STAGE 1 · LIVE

Research

A shopper's first touch with a category is increasingly an AI answer, not a results page. ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude, and Copilot answer "best X for Y" with specific products and cited sources. This is where SKU-level tracking measures who wins.

STAGE 2 · LIVE, EXPANDING

Draft carts

Pre-purchase agents assemble options and carts on the buyer's behalf: ChatGPT Shopping, Perplexity's Buy with Pro, and Google AI Mode shopping experiences. The agent picks the SKU, fills the cart, and hands the shopper to your checkout — if it could parse your PDP.

STAGE 3 · IN PILOT · MID-2026

Agent checkout

Protocol-mediated purchase completion: the Agentic Commerce Protocol (ACP, from OpenAI and Stripe, behind ChatGPT's Instant Checkout), Google's Universal Commerce Protocol (UCP), the Agent Payments Protocol (AP2) on the payments layer, and Shopify's agentic storefront work on the merchant side. All in pilot as of mid-2026 — describe per the published specs, not as mainstream shopping behavior.

The work compounds across stages: the schema, crawler admittance, and machine-readable pricing that win citations in stage 1 are exactly what makes a SKU draftable in stage 2 and transactable in stage 3. Protocol-by-protocol detail lives in the ACP guide and the UCP guide.


The readiness measure

The agent-readability score, factor by factor.

One of the two scores every product carries (the other is the citation score — see SKU-level tracking). It answers a single question: could a shopping agent parse this PDP well enough to recommend it, draft it into a cart, or buy it. Scoring detail in the PDP Score docs.

FactorWhat the scan checksTypical failure on a Shopify store
Product JSON-LD completeness name, brand, sku, gtin, offers (price, currency, availability), image, material, review markup where real Theme renders name, image, price — omits gtin, availability, attributes
robots.txt admittance GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended against your robots.txt — per-bot view in AI Agent Lens A blanket block left over from a 2023 scraping scare
Machine-readable price & availability offers data matches the live PDP, variant by variant, including out-of-stock states Stale price in schema; sold-out variant still marked InStock
Discoverable policies Returns, shipping, and warranty pages linked from the PDP and readable without JS Policy buried in an accordion the crawler never expands
Agentic-checkout wiring Where applicable: readiness signals for ACP/UCP-style flows on the merchant's platform Not applicable yet for most stores — scored leniently while protocols pilot

Factor set as of mid-2026; the score is recalibrated as agent behavior and protocol specs change. AI shopping is changing quarterly.


Get ready

Six things to do before the protocols mature.

Each one is concrete, shippable this quarter, and useful at every stage — not a bet on any single protocol.

1 · Grade a best-seller with the free Agentic Readiness Grader — see which factors fail today.~90 s
2 · Complete the Product JSON-LD — the schema for AI search guide covers the fields; the Product Schema Generator drafts the markup.per SKU
3 · Check robots.txt against the AI crawlers in AI Agent Lens — a blocked bot is an unreadable catalog.1 file
4 · Publish llms.txt with the free llms.txt generator — a machine-readable index of catalog and policies.~10 min
5 · Make returns and shipping policies discoverable from every PDP — agents weigh purchase confidence, not just product fit.site-wide
6 · Read the ACP guide and UCP guide — know what each spec asks of a merchant, and what can safely wait.2 guides
NAMING NOTE

On "ACO" — credit where it's due.

Agentic Commerce Optimization (ACO) was coined by ReFiBuy, which builds a six-stage closed-loop catalog-intelligence platform around it for enterprise teams managing complex, multi-system catalogs — Steve Madden is a named customer, per their site as of mid-2026. The neutral definition lives in the ACO glossary entry.

eCommerce Insights agrees with ReFiBuy on the central claim — the SKU is the right primitive — and serves a different team: the D2C brand whose catalog lives in Shopify admin, with public pricing and a 14-day trial. Side-by-side detail in eCommerce Insights vs ReFiBuy; category framing in the what-is-ACO guide.

When the shopper is an agent, your PDP is the salesperson.

Questions buyers ask

What is agentic commerce?

Commerce where AI agents act on the shopper's behalf, in three stages: researching products (mainstream now — engines answer what-should-I-buy queries), drafting carts (live in ChatGPT Shopping and Perplexity's Buy with Pro), and completing checkout through protocols like ACP and UCP (in pilot as of mid-2026). At every stage, an agent can only recommend, draft, or buy what it can read.

Do I need ACP support today?

Probably not yet. The Agentic Commerce Protocol is in pilot with a limited merchant set as of mid-2026, and most agent-led flows still hand the buyer back to your checkout. What pays off today is the groundwork the protocols assume: complete Product JSON-LD, admitted crawlers, machine-readable price and availability, discoverable policies. That work also wins citations right now.

Is this just SEO renamed?

It overlaps, but the reader changed. Google ranks pages for a human to click; an agent parses product data to recommend, compare, or transact, and it fails silently when schema is broken or a crawler is blocked. The practice is still forming — GEO, AEO, and ACO all name slices of it — but the work resolves to measurable, per-SKU readiness rather than rankings.

Which engines can actually complete a purchase today?

As of mid-2026, fully agent-completed checkout is the exception. ChatGPT's Instant Checkout, built on ACP, runs with a limited merchant set; Google's UCP-based flows are in pilot. The common live behavior is research plus draft carts — ChatGPT Shopping and Perplexity's Buy with Pro assemble the cart, then hand the shopper to the merchant's checkout. Treat agent checkout as arriving, not arrived.

What should a Shopify brand do first?

Fix the Product JSON-LD — it is the most common failure. A typical Shopify theme renders name, image, and price but omits gtin, offers availability, material, and review markup, and agents lean on exactly those fields. Run the free Agentic Readiness Grader on a best-seller, fix what fails, then check robots.txt and publish llms.txt.

Does eCommerce Insights implement ACP or UCP for my store?

No — it is not a protocol vendor. eCommerce Insights scores each SKU's readiness for agent-led flows, including agentic-checkout wiring where applicable, and ships the PDP groundwork as reviewable diffs. Protocol enrollment runs through your platform; Shopify's agentic storefront work and the ACP and UCP pilots are linked from the guides.

Start with the groundwork

Whichever protocol wins, the PDP work is the same.

Grade a PDP free in about 90 seconds, or connect the catalog and score every product. 14-day trial, no credit card. The fix layer: PDP optimization.

research · draft carts · agent checkout