When AI Checks Out for Your Customer, Is Your Shop Ready?

Until recently, shoppers asked AI: "What's good in this price range?" Now they instruct it: "Order two in black, size M." Purchases that complete without a human ever opening your product page — agentic commerce — have moved from demo to shipping product.

This sounds like a big-retailer problem, but small shops feel it first. Agents don't weigh brand recognition the way people do; they weigh readable data.

What actually changed with agentic commerce?

Part of the decision and the checkout is now handled by an AI agent instead of a person. AI used to stop at search and comparison; now it reaches the cart and the payment step. For merchants, one thing changes: a machine reads your product information before your customer does.

Concrete moves are already public:

• OpenAI introduced Instant Checkout inside ChatGPT and released the Agentic Commerce Protocol with Stripe.
• Google published AP2 (Agent Payments Protocol) for agent-initiated payments.
• Card networks are defining how to authenticate non-human buyers, with efforts like Visa Intelligent Commerce and Mastercard Agent Pay.

You don't need to memorize the spec names. The common requirement is simple: price, stock, shipping, and policy must be machine-readable without ambiguity for a transaction to happen.

What can a small shop do this month?

Roughly 80% of the work is organizing information you already have, not installing new technology. Work through this order and agents are far more likely to treat you as a "purchasable" merchant.

1. Unify your product data. Put name, price, options (size/color), stock, and shipping cost in one sheet, and pull anything trapped inside images out into text. Text baked into an image doesn't exist for a machine.
2. Add structured data. Use Product and Offer schema for price, currency, and availability. Check first whether your platform (Shopify, Cafe24, Imweb, WooCommerce) already supports it.
3. State policies as sentences with numbers. Not "contact us about returns," but "returns within 7 days of delivery, unopened, remote-area surcharge applies."
4. Tighten inventory sync. If the feed says in stock, an agent buys immediately. A cancellation call the next day destroys trust faster than anything else.
5. Log order sources. Tag AI-referred sessions and orders so that in a few months you can judge the channel by numbers instead of gut feel.
6. Define your human checkpoint. High-value, custom-made, or B2B orders should require a confirmation step rather than auto-confirming.

What should you be careful about right now?

The two most common failures are blocking real buyers while trying to block bots, and letting a wrong price get executed. Audit your bot-blocking rules and your data accuracy before anything else.

• Over-blocking: Make sure robots.txt or your WAF isn't shutting out AI crawlers and agents entirely.
• Price errors: Missing option surcharges or unapplied discounts get read literally by an agent.
• Fees and settlement: Calculate commission and payout timing before plugging into a new channel.
• Lost differentiation: If only specs and price are readable, that's all you'll be compared on. Spell out "why this product" in plain text — mood photography doesn't reach an agent.

Here's a 15-minute test for today. Ask ChatGPT or Perplexity for recommendations in your category and price band, then paste your own product page text and ask, "Can you place an order with only this information?" Every follow-up question it asks is a gap in your data. Filling those gaps is where preparing for agentic commerce actually begins.