Preparing for agentic commerce.

For years, ecommerce has been built around a familiar journey: a shopper searches, clicks, lands on a product page, filters options, compares reviews, adds an item to cart, and checks out.

Agentic commerce changes that pattern.

In an agentic commerce experience, a customer may ask an AI assistant to do the shopping work for them:

• “Find me a durable carry-on under $300 that fits United's overhead bin rules.”
• “Order the same cat food I bought last month, but only if it is in stock and can arrive by Friday.”
• “Compare three patio dining sets that fit a six-person layout and have good reviews.”

The AI agent may search, compare, summarize, build a cart, verify availability, apply constraints, and eventually help the customer buy. That does not mean merchants disappear. It means the merchant's website, product data, checkout, payment stack, inventory systems, and post-purchase operations need to become understandable to both humans and machines.

The goal is not to “game” AI. The goal is to make your commerce operation clear, accurate, trustworthy, and transactable in a world where AI agents increasingly sit between the customer and the merchant.

Agentic commerce is commerce where AI agents can act on behalf of a shopper. These agents may help customers discover products, compare options, confirm product fit, check price and inventory, create a cart, complete checkout, track orders, initiate returns, or reorder past purchases.

A basic chatbot answers questions. An agentic commerce experience takes action.

That action could be simple, like recommending a product. It could also be transactional, like initiating checkout through a merchant's existing payment and order systems. The customer still matters. The merchant still matters. But the interface between them starts to change.

In practical terms, agentic commerce asks merchants to prepare for three new realities:

1. Your product catalog needs to be readable by AI systems.
2. Your inventory, pricing, shipping, tax, and policy information need to be accurate enough for an agent to rely on.
3. Your checkout and payment stack need to support trusted, secure, auditable transactions that may be initiated through an AI assistant rather than a traditional browser session.

Agentic commerce is still early, but it is no longer theoretical. Major technology companies, payment networks, commerce platforms, and payment service providers are building infrastructure for AI-mediated shopping, checkout, and payment authorization.

That does not mean every merchant needs to rebuild their entire ecommerce stack immediately. In fact, most merchants should not start with a full replatforming project.

The better first move is to become agent-ready.

Agent-ready merchants will be easier for AI systems to understand, easier to recommend accurately, easier to transact with, and easier to monitor. The same work also improves traditional ecommerce: cleaner product data, better structured pages, more accurate inventory, fewer checkout errors, stronger fraud controls, and better attribution.

The merchants that prepare early will have an advantage because agentic commerce rewards operational clarity. Agents need reliable information. They need structured product data. They need accurate inventory. They need deterministic checkout responses. They need clear return policies, shipping promises, and payment outcomes.

If your commerce systems are messy for humans, they will be even messier for agents.

Merchants can think about agentic commerce readiness across seven layers:

1. Product data
2. Website and content accessibility
3. Inventory, pricing, and availability
4. Checkout and cart logic
5. Payments, fraud, and compliance
6. Fulfillment and post-purchase operations
7. Measurement and observability

The best approach is to improve each layer gradually, starting with the systems that already affect your customer experience today.

Agentic commerce begins with product understanding. If an AI agent cannot understand what you sell, who it is for, what problem it solves, whether it is available, and why it is a good fit, it is less likely to recommend it confidently.

Most merchants already have product data, but much of it is written for human browsing or internal operations. Agent-ready product data needs to be clear, structured, complete, and consistent across channels.

Product identifiers

Every product and variant should have stable identifiers — SKUs, variant IDs, GTINs where available, MPNs where relevant, brand names, and category assignments. Avoid changing IDs casually. AI systems, shopping feeds, analytics systems, and checkout systems all depend on stable product identity.

Product titles

Product titles should be descriptive without being stuffed with keywords. A strong title usually includes the brand, product type, defining attribute, size, color, material, model, or use case.

A weak title says: “Premium Pro Series 500.”

A better title says: “Acme Pro Series 500 Stainless Steel 12-Cup Coffee Maker.”

Agents need to understand the product quickly. Clear titles help.

Product descriptions

Descriptions should do more than market the product. They should explain what the product is, who it is for, what it includes, what it is compatible with, and what tradeoffs a customer should understand.

For agentic commerce, the best product descriptions answer the questions a customer would ask before buying:

• What is this product used for?
• What size, capacity, material, or fit does it have?
• What is included in the box?
• What is not included?
• What products is it compatible with?
• What are the common reasons someone returns it?
• What type of customer is it best for?
• What are the care, installation, or maintenance requirements?

Product attributes

Attributes matter more in agentic commerce because agents compare options. Color, size, dimensions, weight, material, ingredients, compatibility, warranty, certifications, age range, fit, power requirements, and care instructions should be structured whenever possible.

A customer may ask an agent for a “non-toxic toddler table under 30 inches wide” or a “dishwasher-safe stainless steel water bottle that fits in a car cup holder.” If your product data does not include those attributes, the product may not appear in the comparison.

Media

High-quality images still matter. So do alternate views, lifestyle images, dimensional diagrams, spec sheets, installation guides, and videos. AI systems may rely on the text around media, filenames, alt text, captions, and product metadata to understand what the media shows.

Reviews and Q&A

Customer reviews and product Q&A become useful context for AI systems. They reveal how real customers describe fit, quality, use cases, and objections. Merchants should not manipulate reviews, but they should structure and maintain them so they are easy to understand.

Compatibility, substitutes, and accessories

Agentic commerce will increase comparison shopping. Merchants should make it easier for agents to understand which products go together:

• This printer works with these toner cartridges.
• This stroller is compatible with these car seats.
• This grill cover fits these grill models.
• This cable is a substitute for this discontinued cable.
• This bundle is better for beginners than buying separate parts.

This is not just merchandising. It is machine-readable context.

In traditional ecommerce, inaccurate inventory creates customer frustration. In agentic commerce, inaccurate inventory can also cause agents to lose trust in your store.

Agents need to know whether an item is actually available, whether the price is current, whether a promotion applies, and whether the product can arrive when the customer needs it.

Real-time or near-real-time inventory

Inventory should be updated frequently enough that your website, product feeds, marketplaces, ads, and checkout systems are not contradicting each other. For fast-moving products, daily updates may not be enough — consider intraday updates or API-based updates for price, promotions, availability, and local inventory.

Variant-level availability

Do not only mark the parent product as in stock. Agents need to know whether the exact size, color, configuration, or location-specific item is available. A jacket is not “in stock” if only XS remains and the shopper asked for XL.

Clear fulfillment promises

Availability is not only about stock. It is also about delivery. Agent-ready merchants should expose clear fulfillment logic:

• Ships today
• Ships in 2–3 business days
• Available for pickup at this store
• Available for local delivery
• Backordered until a specific date
• Preorder available
• Not available in certain regions
• Oversized shipping applies

Agents will increasingly optimize for customer constraints. “Arrives before Friday” may matter more than “lowest price.”

Consistent pricing and promotions

Pricing should match across product pages, feeds, cart, checkout, and confirmation emails. Promotions should be deterministic. If an agent applies a discount code or sees a sale price, checkout should return a clear success or failure reason.

Ambiguous promotions create broken checkout experiences. Examples include:

• “Buy more, save more” without clear rules
• Member-only pricing with unclear eligibility
• Region-specific discounts that only appear after address entry
• Promo codes that conflict silently
• Bundles that recalculate unpredictably

Agents need clear cart logic. Customers do too.

Your website is still important. Even as more shopping starts inside AI assistants, product pages, category pages, help pages, policy pages, and structured data remain critical.

Use structured product data

Product pages should include structured data for product name, image, price, availability, brand, GTIN where available, ratings where appropriate, shipping, returns, and offers. Structured data should match what the customer can see on the page. Do not publish one price in structured data and another price visually. Do not mark a product as available if the visible page says it is sold out.

Make product pages crawlable

Important product information should not be hidden behind scripts that crawlers cannot reliably process. Server-rendered or easily accessible product content is usually safer for important fields like title, description, price, availability, and variant information. Make sure your robots.txt, sitemap, canonical tags, and noindex rules are not accidentally blocking important product pages, images, or policy pages.

Create clear policy pages

Agents may need to answer customer questions about shipping, returns, warranties, subscriptions, cancellation windows, sizing, privacy, and support. Policy pages should be written in plain language and kept current. Good policy content includes: shipping methods and timelines, return window, refund timing, restocking fees, final sale exceptions, warranty terms, subscription cancellation rules, contact options, international restrictions, store pickup rules, and the damaged or missing item process.

Agents will struggle with vague policy language. Customers already do.

Consider an agent information page

Some merchants may choose to publish an agent-facing information page or file that summarizes how AI systems should understand the store — links to product feeds, sitemap locations, policy pages, brand guidelines, customer support rules, and allowed crawler behavior. Emerging files like llms.txt may become useful as signage for AI systems, but treat them as optional and experimental. They are not a replacement for official product feeds, structured data, APIs, or commerce protocol integrations.

In agentic commerce, checkout may not begin with a shopper clicking “Add to Cart” on your website. An AI agent may request a cart or checkout session based on the customer's intent. That means your commerce stack should be able to create, update, validate, and complete checkout sessions through reliable APIs.

The merchant should remain the source of truth for cart state. An agent can pass customer intent, product selections, address information, delivery preferences, or payment authorization details, but the merchant should calculate the final order.

Your system should be able to answer:

• Are the requested items still available?
• Is the requested quantity allowed?
• What is the current price?
• Does the promotion apply?
• What are the shipping options?
• What taxes and fees apply?
• What is the final order total?
• Is the payment authorized?
• Was the order accepted, declined, or placed on hold?
• What error should be returned if something fails?

Agent-ready checkout should include clear error handling. For example:

• Item out of stock
• Quantity unavailable
• Shipping address unsupported
• Payment declined
• Promotion not eligible
• Price changed
• Tax calculation unavailable
• Fraud review required
• Merchant account unavailable
• Checkout session expired

Agents need actionable responses. “Something went wrong” is not enough.

Payments are one of the most important parts of agentic commerce. A customer may authorize an AI agent to purchase a product within certain limits. The payment stack needs to support that authorization securely, with clear consent, scoped permissions, fraud controls, and a reliable audit trail.

Merchants should talk to their payment service provider, gateway, fraud vendor, and ecommerce platform about agentic payment readiness. Important questions include:

• Can we accept delegated or scoped payment tokens?
• Can our PSP support network tokens or agent-specific payment credentials?
• Can we authorize payment only after the final cart is validated?
• Can we handle maximum amount limits, expiration windows, or customer-defined constraints?
• Can we distinguish trusted agent traffic from suspicious automation?
• Can we pass agent-related signals into fraud decisioning?
• Can we handle refunds, partial refunds, cancellations, chargebacks, and order updates cleanly?
• Will this change our PCI scope?
• Are there regional requirements for strong customer authentication or step-up verification?

Merchants should avoid designing a payment flow where an agent-supplied cart total is blindly trusted. Your system should always validate price, tax, shipping, inventory, and eligibility before authorizing or capturing funds.

The safest path for many merchants will be to use existing PSP-supported methods rather than directly handling sensitive card data. That allows merchants to prepare for agentic commerce while limiting unnecessary compliance complexity.

Agentic commerce introduces a new challenge: not all automation is bad. Some automated traffic may come from legitimate AI agents helping real customers. Other automated traffic may come from scrapers, bots, fraudsters, fake storefronts, credential attacks, or malicious systems trying to manipulate shopping flows. Merchants need to separate trusted agent activity from harmful automation.

Request authentication

Agentic checkout APIs should authenticate requests. Do not expose sensitive cart, checkout, payment, or customer operations through unauthenticated endpoints.

Signed requests and verification

Where supported, verify that requests came from trusted platforms and were not tampered with.

Idempotency

Agent-driven systems may retry requests. Your checkout APIs should support idempotency so duplicate requests do not create duplicate orders, duplicate charges, or conflicting cart states.

Rate limiting

Agent traffic can scale quickly. Rate limits should protect your systems without blocking legitimate commerce partners.

Session expiration

Checkout sessions, payment tokens, and customer authorizations should expire. Do not allow an agent to complete an old cart with stale pricing or unavailable inventory.

Least privilege

Agent permissions should be limited to the task. A customer may authorize “buy this item if under $100 and delivery is available by Friday.” That should not become permission to buy any item at any price.

Fraud scoring

Fraud systems should begin capturing agent-related signals: source, platform, session type, device context, token type, shipping mismatch, abnormal cart behavior, repeated retries, and high-risk product categories.

Prompt-injection awareness

As AI agents read websites, reviews, descriptions, files, and policy pages, attackers may try to hide malicious instructions in content. A fake product listing could attempt to manipulate an agent's ranking, redirect a shopper, or extract sensitive data. Treat user-generated content, reviews, third-party marketplace content, and external product data as untrusted inputs. AI-facing systems should validate outputs, restrict actions, and require explicit authorization for high-risk steps.

The sale does not end at checkout. Agentic commerce will also affect order tracking, delivery updates, returns, cancellations, exchanges, warranties, and customer support.

If an AI assistant helps place an order, the customer may later ask the same assistant:

• “Where is my order?”
• “Can I return this?”
• “Did the refund go through?”
• “Can I reorder the same product?”
• “Can I change the shipping address?”

To support that experience, merchants need clean post-purchase data and workflows. Agent-ready post-purchase operations include:

• Order confirmation
• Order status
• Shipment tracking
• Delivery status
• Cancellation rules
• Return eligibility
• Return label creation
• Refund status
• Partial refund support
• Exchange options
• Warranty claims
• Customer support escalation

Merchants should make sure these updates flow back into the systems where the customer expects to see them. If an AI assistant shows stale order information, the customer may blame the merchant.

One of the biggest challenges in agentic commerce will be visibility. Today, merchants can usually see traffic sources, campaign performance, product page behavior, cart conversion, checkout errors, payment declines, fraud decisions, and order outcomes.

In agentic commerce, some of the customer journey may happen outside the merchant's website. The customer may compare products, narrow choices, ask questions, and build intent inside an AI assistant before the merchant ever sees a request. That creates an Agent Visibility Gap.

The Agent Visibility Gap is the difference between what the merchant can measure in traditional analytics and what actually influenced the customer's decision through AI-mediated discovery and purchase.

Merchants should begin building a measurement layer for agentic commerce. Useful metrics include:

• AI referral traffic
• Agent crawler activity
• Product feed coverage and errors
• Structured data errors
• Product eligibility by channel
• Price and availability mismatches
• Agent checkout sessions created and completed
• Agent checkout errors
• Payment authorization success rate
• Payment decline rate
• Fraud review and chargeback rates
• Out-of-stock errors
• Delivery promise misses
• Return rate by agent-referred order
• Customer support contacts after agent-referred orders
• Product questions agents cannot answer
• Top products surfaced by AI referrals
• Products with high AI interest but low conversion

This should become a dashboard, not a one-time report. Agentic commerce will evolve quickly, and merchants will need to understand what is working, what is breaking, and where customers are being lost.

Most merchants do not need to solve everything immediately. A practical 90-day plan can create a strong foundation.

First 30 days: clean up the foundation

Start with product and site readiness.

• Audit product data completeness
• Confirm stable SKU and variant IDs
• Add or improve GTINs, MPNs, brand fields, and product attributes
• Review product titles and descriptions for clarity
• Add compatibility, sizing, dimensions, care, materials, and use-case information
• Check structured product data on major product pages
• Confirm price and availability match between page, feed, and checkout
• Review robots.txt, sitemap, canonical, and noindex settings
• Make shipping, return, warranty, and support pages easier to understand
• Identify high-value or high-volume SKUs for an agentic commerce pilot
• Start tracking AI referral sources and crawler behavior in analytics

Goal: make your catalog and website easier for both humans and machines to understand.

Days 31–60: improve operational reliability

Focus on inventory, pricing, checkout, and payment readiness.

• Improve frequency of inventory and price updates
• Review local inventory, pickup, delivery, and backorder logic
• Define a source of truth for product, price, inventory, and promotion data
• Document checkout session behavior
• Map all checkout failure states
• Review tax, shipping, discount, and fee calculations
• Ask your PSP about delegated payments, tokenization, network tokens, and agentic payment support
• Ask your fraud provider how it plans to identify trusted agent activity
• Review PCI scope with your payments or compliance team
• Confirm payment page script monitoring and tamper controls
• Build a simple internal scorecard for agent-readiness by product category

Goal: make sure your systems can return accurate answers when an agent asks, “Can this customer buy this product right now?”

Days 61–90: prepare for controlled pilots

• Create a pilot group of products with complete data and reliable availability
• Test checkout edge cases: out-of-stock, price changes, shipping restrictions, declined payments, expired sessions
• Confirm idempotency behavior for repeated requests
• Confirm order update and refund workflows
• Test analytics tagging for AI-referred or agent-assisted orders
• Create dashboards for agent traffic, errors, conversion, payment outcomes, and fulfillment issues
• Red-team agent-facing content for manipulation or prompt-injection risks
• Train customer support teams on agent-assisted orders
• Create an escalation path for payment, fraud, fulfillment, and technical issues
• Decide which agentic commerce integrations are worth pursuing first

Goal: move from general readiness to controlled experimentation.

Agentic commerce is not only a marketing project. It touches too many systems. The right working group usually includes:

• Ecommerce leadership
• Product data or PIM owner
• Merchandising
• SEO or content team
• Paid media or marketplace team
• Web development
• Payments team
• Fraud and risk team
• Analytics team
• Legal or compliance
• Fulfillment or operations
• Customer support

A single owner should coordinate the roadmap, but the work is cross-functional. For smaller merchants, this may be one ecommerce lead working with an agency, developer, PSP, and platform provider. For larger merchants, agentic commerce should become part of the broader digital commerce roadmap.

• Do not rip out your ecommerce platform just because agentic commerce is emerging.
• Do not create a separate, disconnected product catalog only for AI.
• Do not let agents bypass your normal order validation, fraud screening, payment authorization, or compliance processes.
• Do not assume every AI crawler, bot, or agent is trustworthy.
• Do not treat experimental files or unofficial standards as replacements for structured data, product feeds, and reliable APIs.
• Do not ignore the customer relationship. Even if discovery starts in an AI assistant, the customer's trust still depends on your product accuracy, payment experience, delivery promise, return process, and support quality.

The merchants best positioned for agentic commerce will not necessarily be the largest. They will be the clearest.

• Clear product data.
• Clear availability.
• Clear pricing.
• Clear policies.
• Clear checkout responses.
• Clear payment authorization.
• Clear order updates.
• Clear measurement.

Agentic commerce rewards merchants whose systems can be understood and trusted. That is good news, because the work required to prepare also improves the ecommerce experience you already have today.

Agentic commerce is not a distant future where AI replaces merchants. It is a near-term shift in how customers discover, evaluate, and purchase products. The AI agent becomes a new interface. The merchant still owns the commerce experience.

To prepare, merchants should focus on becoming agent-readable, agent-transactable, and agent-observable: better product data, more accurate inventory, stronger checkout APIs, secure payment readiness, better fraud controls, reliable fulfillment data, and clearer measurement.

The merchants who start now will not just be ready for AI agents. They will be better merchants overall.