The Agentic Commerce Playbook: How to Prepare Your Store for AI Shopping Agents

What Is Agentic Commerce?

Agentic commerce is any transaction where an AI agent — ChatGPT, Gemini, Claude, Perplexity, Microsoft Copilot — participates meaningfully in discovery, comparison, checkout, or post-purchase. The agent acts on the buyer's behalf, often without them ever visiting your website.

Four protocols frame the current landscape:

Google's Unified Commerce Protocol (UCP) launched at NRF on January 11, 2026, with 20+ partners including Shopify, Etsy, Target, Walmart, Wayfair, Best Buy, Macy's, The Home Depot, Zalando, Adyen, Amex, Mastercard, Stripe, and Visa. Sundar Pichai framed it directly: "With a few taps via Google Pay, the purchase is complete without ever leaving the conversation." UCP is how your products become buyable inside Gemini and Google Search.

OpenAI and Stripe's Agentic Commerce Protocol (ACP) launched September 29, 2025, under an Apache 2.0 license, with a current stable spec of 2026-01-30. ACP is how ChatGPT discovers and recommends your products. In March 2026, OpenAI ended the original Instant Checkout and pivoted ACP toward merchant-hosted checkout — the buyer clicks from ChatGPT into your site with the cart pre-built.

Microsoft Copilot Checkout (January 2026) brought Shopify-powered purchase flows into the Copilot experience. Microsoft self-reports that early pilots — Keen Footwear and Pura Vida — saw 53% more purchases within 30 minutes and a 194% uplift in conversion likelihood for high-intent queries.

Perplexity's Buy with Pro matters because of the zero-fee Merchant Program. Where ChatGPT's ACP reportedly charges around 4% on in-chat checkout (this figure is unverified), Perplexity charges 0%. That is a real delta for margin-tight categories.

The Consumer Shift Is Real — and Uneven

Riskified's global study of 5,400 shoppers across nine countries (October 2025) is the most-cited number in this space: 73% of consumers now use AI somewhere in their shopping journey — 45% for product ideas, 37% to summarize reviews, 32% to compare prices. 13% have completed a purchase via an AI assistant. And 70% say they are "somewhat comfortable" with AI agents buying on their behalf (Riskified, October 2025).

Cross-reference the number against independent sources and the picture gets more honest. Salesforce found 39% have used AI for product discovery (54% among Gen Z). Adobe's 2025 Holiday Consumer Survey reports 64% using AI assistants more than they used to, 65% more confident in purchases after AI help, and 47% trusting AI. Bain pegs GenAI product research at 30–45%, with 50% cautious about fully autonomous agents. YouGov (July 2025) asked differently and got a tighter answer: 43% had heard of AI shopping agents, but only 14% had actually used one.

The commercial signal is clearer than the survey noise. OpenAI's shopping research reports 52% product accuracy on multi-constraint queries (vs 37% for standard ChatGPT Search), meaning buyers routinely include constraints like price, brand, and delivery window. Adobe's holiday data shows AI-driven traffic to US retailers grew 693% year-on-year in November–December 2025, and AI referrals converted 31% better in holiday 2025 and 42% better in Q1 2026. Shopify's Harley Finkelstein confirms the merchant-side view: orders from AI search platforms to Shopify stores are up 15× since January 2025.

Put differently: far fewer people have bought through an agent than the headlines suggest, but the people who do buy via AI spend more, convert faster, and cost less to acquire. That is the arbitrage.

Salesforce tracked $67B of retail sales (20% of the $336.6B Cyber Week 2025 total) as AI- and agent-driven, with AI-assistant retail traffic up 119% year-on-year in H1 2025. Brands on Salesforce's Agentforce 360 (Pandora, Shark Ninja, Funko) grew sales 32% faster than peers during Cyber Week. The aggregate is still small relative to total ecommerce — but the curve is the only thing that matters.

The Protocols, Explained for Operators

You do not need to implement all four. You need to know which one matters when.

UCP — Google

A /.well-known/ucp discovery endpoint on your domain, three core capabilities (dev.ucp.shopping.checkout, discount, fulfillment), and POST/PUT /checkout-sessions endpoints. Launched US-first, with PayPal coming soon and global rollout in the coming months. It is compatible with A2A, AP2, and MCP. If you already sell through Google Shopping and ship to the US, UCP is where you should spend the first cycle.

ACP — OpenAI + Stripe

Four REST endpoints (Create / Update / Complete / Cancel Checkout), HTTPS + Bearer tokens + HMAC-signed webhooks. Licensed Apache 2.0, with roughly 1.3k GitHub stars at the time of writing. ACP now favors merchant-hosted checkout, which means your existing checkout keeps working — ACP feeds the discovery layer. Current discovery partners include Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair. Walmart built a native in-ChatGPT app experience.

AP2 — Google (plus 60+ partners)

AP2 is the payments trust layer. It defines three mandates the buyer signs: the Intent Mandate (a price ceiling and rules), the Cart Mandate (the exact items the agent is allowed to buy), and the Payment Mandate (which links to the first two). It is payment-method agnostic — cards, bank rails, even stablecoins via the A2A x402 extension. Partners include Mastercard, Amex, PayPal, Revolut, Adyen, Checkout.com, Coinbase, Salesforce, ServiceNow, Deloitte, PwC, and Okta. Visa's Chief Product & Strategy Officer Jack Forestell frames the stakes directly: “Agentic commerce is unlocking an entirely new wave of digital interactions, but it can only scale if every player in the ecosystem can trust the agents participating in it.”

MCP — Anthropic

The Model Context Protocol, open-sourced by Anthropic in November 2024, is how agents read context from your store. The important one for merchants is Shopify Storefront MCP: every Shopify store now exposes /api/mcp with no authentication required for public catalog operations. ChatGPT, Perplexity, and Microsoft Copilot all integrate with it. If you run Hydrogen, set proxyStandardRoutes = true and you are done.

How Agents Pick Merchants

Agents do not rank like Google's tenth blue link. They score candidates against three dimensions — structural completeness, semantic density, and trust — then pick the one that clears a confidence threshold, reported to sit around 80% before an agent will recommend or auto-purchase on a user's behalf.

Structural completeness means stable identifiers and populated attributes. Products with a stable GTIN and 30+ populated attributes beat products with the 7 required attributes, even when the extras feel cosmetic. Stale inventory triggers MERCHANDISE_NOT_AVAILABLE errors that hurt your reliability score across future retrievals. Missing return-policy fields directly downgrade your trust score in UCP and ACP.

Semantic density means content an LLM can read. Unique long-form descriptions, material and dimensional facts, context of use, comparisons. Agents do not click thumbnails, they read the JSON.

Trust signals are disproportionately external. Trustpilot became the 5th most-cited domain on ChatGPT in January 2026 (per Promptwatch citation tracking). The platform's "3Rs" framework — Recency, Relevance, Ranking — roughly describes how agents weight reviews: new beats old, category-specific beats generic, and volume plus average rating beats either alone. Agents synthesize across Google reviews, Trustpilot, Reddit threads, and industry sources, which means on-site marketing copy matters less than what the rest of the internet says about you.

The 10-Point Operator Checklist

Tactical. Each item can be scoped in a sprint. If you do nothing else from this article, do these.

A realistic Product JSON-LD block

This is the minimum you should be shipping on every PDP. Replace values with your data, keep the structure.

<script type="application/ld+json">
{
  "@context": "https://schema.org/",
  "@type": "Product",
  "name": "Linen Summer Shirt",
  "image": [
    "https://example.eu/images/shirt-front.jpg",
    "https://example.eu/images/shirt-back.jpg"
  ],
  "description": "100% European linen, pre-washed, unisex fit.",
  "sku": "LS-2026-M-WHT",
  "gtin13": "4006381333931",
  "brand": { "@type": "Brand", "name": "Example Baltic" },
  "inProductGroupWithID": "LS-2026",
  "color": "White",
  "material": "Linen",
  "size": "M",
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.7",
    "reviewCount": "312"
  },
  "offers": {
    "@type": "Offer",
    "url": "https://example.eu/products/linen-shirt",
    "priceCurrency": "EUR",
    "price": "59.00",
    "priceValidUntil": "2026-12-31",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition",
    "shippingDetails": {
      "@type": "OfferShippingDetails",
      "shippingDestination": {
        "@type": "DefinedRegion",
        "addressCountry": "LT"
      },
      "deliveryTime": {
        "@type": "ShippingDeliveryTime",
        "handlingTime": { "@type": "QuantitativeValue", "minValue": 0, "maxValue": 1, "unitCode": "DAY" },
        "transitTime":  { "@type": "QuantitativeValue", "minValue": 1, "maxValue": 3, "unitCode": "DAY" }
      }
    },
    "hasMerchantReturnPolicy": {
      "@type": "MerchantReturnPolicy",
      "applicableCountry": "LT",
      "returnPolicyCategory": "https://schema.org/MerchantReturnFiniteReturnWindow",
      "merchantReturnDays": 30,
      "returnMethod": "https://schema.org/ReturnByMail",
      "returnFees": "https://schema.org/FreeReturn",
      "refundType": "https://schema.org/FullRefund"
    }
  }
}
</script>

The Baltic Angle: Exposed and Ahead

Lithuania has the highest enterprise ecommerce adoption in the EU at 43.03%, ahead of Denmark (38.78%), Sweden (36.72%), and Ireland (36.57%) (Eurostat). At the same time, 86.55% of Lithuanian e-selling enterprises depend on marketplaces — the highest marketplace concentration in the EU — and only 46.9% have a meaningful own-website channel (the lowest).

That combination is unusually exposed to agentic disintermediation. When an AI agent routes buyers based on structured data and external reviews rather than marketplace placement, brands that rely on Pigu.lt, 220.lv, Kaup24, or Hansapost for discovery suddenly compete on the agent's terms, not the marketplace's. The Lithuanian ecommerce market sits at $3.18B in 2025 and is projected to reach $6.54B by 2030 at a 15.46% CAGR — a fast-growing market about to get a new intermediary layer.

Consumer adoption is ahead of the curve: Microsoft's AI Diffusion Report puts Latvia at 55%, Lithuania at 52%, and Estonia at 51% — all top-10 globally. Lithuania's own curve went from 17% in 2023 to 35% in 2024 to 52% in 2025. Poland sits at 7% for comparison. Eurostat data (December 2025) shows 32.7% of EU residents used GenAI in 2025, with Estonia in the top three.

The regional proof point is Zalando — one of only two European UCP launch partners. Zalando calls itself the #1 fashion platform referred by conversational AI, scaled AI-generated product content from 0 to 90% of catalog, and reports a 13% uplift in items-in-bag from precision matchmaking (Zalando FY 2025 report). PHH Group (Pigu.lt, 220.lv, Kaup24, Hansapost) implemented Algolia AI search in 2025. Vinted remains the region's most globally visible property but has not disclosed agentic work. Baltic operators should move faster than the EU average because the ceiling is higher and the adoption curve is already steeper.

The Reality Check

Not all of this is upside. Deloitte's October–November 2025 retail executive survey (n=330) found 81% of retail executives believe GenAI will weaken brand loyalty by 2027. When an agent picks for the buyer based on price, delivery window, and trust score, the brand becomes a supplier, not a relationship.

Walmart blocked external agents from its checkout in late 2025 and built its own Sparky experience instead. Amazon CEO Andy Jassy has publicly expressed skepticism about letting third-party agents transact through Amazon. Attribution breaks — traditional UTM and last-click models do not capture agent-mediated purchases cleanly. The widely cited figure that 93% of Google AI Mode searches end without a click has no primary Google disclosure and should be treated as directional rather than authoritative.

Regulation is the other wall. The EU AI Act becomes fully applicable on August 2, 2026. Agents making autonomous purchase decisions on behalf of consumers may fall under high-risk classification, with fines up to 7% of global turnover for prohibited practices and 3% for standard violations. The UK CMA has stated that businesses are "fully liable" for AI agent conduct on their behalf. If your agent makes a bad decision on a customer's money, you own it.

Two more quiet risks worth naming. SEO industry studies (non-peer-reviewed) show organic click-through rates down 61% on queries that trigger AI Overviews — brand media planning was built on clicks, and clicks are thinning. And delivery vendors such as nShift have reported that agents will default to the cheaper or faster option when prices tie, which means logistics becomes the tiebreaker in an increasingly flat competitive set. Treat that as a planning input, not a proven law.

Next 30/60/90: What to Do

Days 1–30

• Audit every PDP for Product + Offer + AggregateRating + Review JSON-LD. Fix missing fields first.
• Update robots.txt to allow ChatGPT-User, Claude-User, PerplexityBot, OAI-SearchBot. Disallow /cart and /checkout.
• Publish /llms.txt pointing to product feeds, shipping, return, and FAQ pages.

Days 31–60

• Ship an ACP product feed if you sell to US ChatGPT users. Refresh every 15 minutes.
• Flip native_commerce: true in Merchant Center. Add handling_cutoff_time and video_link.
• Audit Trustpilot and Google Reviews — aim for 100+ reviews and an 80% response rate on each.

Days 61–90

• Pilot an agent-assisted checkout flow. If you run Shopify, expose /api/mcp and test against ChatGPT and Perplexity.
• Update your attribution model to capture agent-mediated referrals separately from organic and paid.
• Run an EU AI Act compliance review before August 2, 2026. Document agent policies, user consent flows, and mandate handling.