How to Make Your Checkout AI-Agent-Ready: The Merchant's Guide to Agentic Commerce

AI-referred traffic to United States retail sites grew 393% year over year in Q1 2026, and those visitors convert roughly 42% better than traditional search traffic, according to Adobe Analytics.

B2Bcentr tracks the infrastructure decisions behind that shift, separating what has actually shipped from what was merely announced.

The gap in this market is no longer demand or model quality, it is merchant plumbing: agents can find products, but most checkouts were built for a human with a mouse, a session cookie, and a saved card.

This guide breaks down exactly what agent-readiness requires, layer by layer, which protocols to support in which order, what it costs to build, and how to measure whether any of it is working.

Key Takeaways

• Agent-readiness starts with catalog data, not protocols. Fix feeds before writing checkout endpoints.
• AI-referred retail traffic grew 393% year over year in Q1 2026.
• Support ACP and UCP together. Multi-protocol merchants see roughly 40% more agentic traffic.
• Verify agents cryptographically with signed headers instead of blocking unknown automated traffic outright.
• Gartner expects AI agents to intermediate over $15 trillion in B2B purchases by 2028.

The Agentic Commerce Market in 2026: What the Numbers Actually Say

Forecasts for this category disagree by orders of magnitude, and almost all of the disagreement is definitional rather than predictive. McKinsey counts AI-orchestrated revenue across the whole commerce chain and lands at $3 trillion to $5 trillion globally by 2030.

Bain includes agent-influenced purchases and projects $300 billion to $500 billion in the United States. eMarketer counts only checkout that happens inside an AI platform and forecasts roughly $20.6 billion of US retail spending in 2026, about 1.5% of total ecommerce.

The narrow number is the honest one for planning purposes today, and it is small. The traffic number is the one that should move your roadmap. Adobe reported 4,700% year over year growth in AI-driven visits to US retail sites across 2025, and the Q1 2026 figure of 393% shows the channel compounding off a much larger base. Over Cyber Week 2025, global sales reached $336.6 billion, with AI and agents driving roughly $67 billion of that.

Then there is the part most consumer coverage skips. Gartner expects around 90% of B2B buying to route through AI agents by 2028, representing more than $15 trillion in spend, and Forrester projects that one third of B2B payment workflows will involve AI agents by the end of 2026. If those hold even directionally, the volume story of agentic commerce is a procurement story wearing a retail costume.

Why Agent Traffic Converts Badly Today

The uncomfortable finding across 2026 research is that agent-referred sessions underperform at the final step not because shoppers are uninterested, but because merchant infrastructure cannot service an agent request.

Product data is incomplete, inventory is stale, pricing lives behind a rendered page, and checkout assumes a browser that can run scripts and solve a challenge.

Trust is the second constraint. Checkout.com data published in June 2026 found that 65% of US consumers would trust AI to compare prices while only 14% would trust it to place an order autonomously, a gap of 51 points.

The same body of research shows what closes it: spending caps (cited by 30% of consumers), instant revocation (29%), and easy cancellation (28%). Average authorized spend before re-approval sits around £177, roughly $224, which is below what merchants assumed.

Merchants know they are behind. A Payments Association survey found 58% of UK online merchants believe agents have already reached their platforms, while only 3% of transactions involve one. Checkout.com put active preparation at 89% of merchants, with 72% agreeing consumers will adopt agent-led shopping faster than merchants can support it.

What "Agent-Ready" Actually Means: The Four Layers

Agent-readiness is not a single integration. It is four dependent layers, and a gap at the bottom blocks everything above it.

1. Data layer: complete, structured, machine-readable product information.
2. API layer: real-time availability, pricing, and cart state exposed programmatically.
3. Identity layer: cryptographic verification of which agent is calling and on whose behalf.
4. Transaction layer: protocol-compliant checkout endpoints and delegated payment credentials.

Most merchants skip to layer four because it is the one with press releases attached. That sequencing fails. An agent that cannot resolve your product's dimensions will never reach your checkout endpoint.

Layer 1: Make Your Catalog Machine-Readable

In traditional ecommerce, imagery and copy persuade a human. In agentic commerce the primary reader is a machine that compares structured fields and eliminates anything ambiguous.

A human buyer might call to clarify a missing dimension. An agent simply moves to the supplier whose data answers the question cleanly.

Three benchmarks determine whether your catalog survives that comparison:

• Attribute completeness. Every SKU populates every relevant attribute in structured fields, with consistent units and naming. Google AI Shopping data suggests stores approaching full attribute completion see three to four times higher visibility in AI recommendations than sparse-data catalogs.
• Cross-channel consistency. The price, title, and availability an agent reads in your feed must match your product detail page. Mismatches are treated as unreliability, not as a rounding error.
• Freshness. Agents do not wait for nightly batch syncs. Inventory and price changes need to propagate in near real time.

For B2B sellers the lift is heavier, because most technical documentation, spec sheets, and certifications still live in PDFs that no agent can parse reliably. Converting those into structured attributes is the single highest-leverage project on this list.

The same discipline drives visibility in AI answers generally, which is why this work overlaps with answer engine optimisation and broader AI search visibility programmes.

Layer 2: Expose Real-Time Commerce APIs

Once data is clean, agents need to query it live. That means product feed endpoints, availability checks, and pricing calls that respond in a request cycle rather than a nightly export.

Shopify has reported that AI searches powered by its Catalog API convert at roughly twice the rate of those relying on scraped data, a vendor-stated figure rather than an audited benchmark, but consistent with the mechanism.

Transport matters less than most teams assume. Google's Universal Commerce Protocol supports REST, Agent2Agent, and the Model Context Protocol as options, so you are not locked into one communication layer. If your team already runs MCP servers for internal tooling, that same surface can carry commerce capabilities.

Layer 3: Identify the Agent Before You Trust It

The default merchant posture toward automated traffic has been to block it. That is now a revenue decision. Roughly 78% of financial institutions expect fraud to rise as AI shopping agents scale, so the answer is verification rather than blanket denial.

Visa and Cloudflare's Trusted Agent Protocol, launched in October 2025, is the most widely deployed approach. Agents attach three signed HTTP headers to every request. Signature-Agent carries the URL of the agent operator's public key directory, Signature-Input declares the key ID, validity window, and a tag distinguishing browse-only from payment intent, and Signature carries the Ed25519 signature over the canonical request.

Merchants verify against a Visa-operated directory. No money moves over the protocol itself. Cloudflare verifies these at the network edge, and Akamai announced a parallel integration in December 2025.

Mastercard takes a different architectural route with Agent Pay, issuing an Agentic Token that binds three identities into one credential: the cardholder, the registered agent, and the transaction. Both approaches sit on top of existing EMV tokenization rather than replacing it, and both use Web Bot Auth as the underlying authentication layer.

Adoption so far is concentrated on the processor side, because a processor can implement once and cover its whole merchant base. Direct merchant integrations skew toward large retailers with in-house engineering depth.

For most mid-market merchants, the practical move is to ask your PSP and your CDN what they already verify, then surface that signal in your own bot management rules. This is a natural extension of any serious cybersecurity stack review.

Layer 4: Build the Checkout and Payment Path

This is where the protocols live. Five families now shape the space: OpenAI and Stripe's ACP, Google's UCP, Google's AP2, Visa's Trusted Agent Protocol, and Mastercard Agent Pay. Merchants supporting multiple protocols reportedly see around 40% more agentic traffic than single-protocol merchants.

ACP (Agentic Commerce Protocol) is the checkout standard, released under Apache 2.0 in September 2025. The current stable spec, dated 2026-04-17, covers cart, feed, orders, authentication, and MCP transport. Implementation centres on five REST endpoints: create session, update session, get state, complete purchase, and cancel.

Every response returns full checkout state including line items, totals, and shipping options. Payment moves through a narrowly scoped token, so the agent never touches the raw card number and you remain merchant of record with your catalog, pricing, branding, fulfilment, and dispute handling intact.

Two implementation details matter. First, publish and verify a /.well-known/acp manifest that AI crawlers can reach. Second, if you are on Stripe, enable the shared payment token flow.

Other PSPs integrate through the Delegated Payment Spec, which is more work. Shopify merchants on a paid plan with Shopify Payments largely inherit the integration rather than building it.

UCP (Universal Commerce Protocol), launched at NRF in January 2026, has broader scope: discovery, cart, checkout, orders, and post-purchase. Merchants publish a /.well-known/ucp manifest and expose capabilities for Checkout, Orders, and Catalog. There is no central approval committee.

AP2 (Agent Payments Protocol) does not move money. It produces evidence, through three signed mandates expressed as W3C Verifiable Credentials. An Intent Mandate captures what the user authorized including budget ceiling and category, a Cart Mandate binds a specific SKU and total to that intent, and a Payment Mandate authorizes the charge.

Google contributed AP2 to the FIDO Alliance in May 2026. For merchants selling into enterprise procurement, this is the layer that makes agent purchases auditable. Our full breakdown of agent payment protocols covers how these compose in a single transaction.

Set Expectations: Discovery First, Checkout Second

OpenAI retired Instant Checkout in March 2026, roughly five months after launch, with reporting suggesting fewer than about fifteen Shopify merchants ever went live against it.

The stated reason was that buyers preferred to complete purchases where their saved payment methods and order history already lived.

The protocol survived the product and now underpins discovery and merchant apps inside ChatGPT. The strategic read for merchants is direct: in 2026, the return on agent-readiness comes primarily from being discoverable and transactable from an AI surface, with the purchase completing on infrastructure you control.

Build for discovery plus redirect, and treat in-chat conversion as upside rather than as the business case.

What This Costs to Build

Budget conversations stall because nobody publishes clean benchmarks. The following are directional US market ranges rather than quoted prices, and they assume an existing ecommerce stack rather than a replatform.

Headcount is the larger line. Directional US base salary ranges for the roles that carry this work: a commerce or backend engineer building the endpoints runs roughly $130,000 to $185,000, a payments integration engineer $150,000 to $210,000, a PIM or product data specialist $75,000 to $115,000, and a technical SEO or AEO lead $95,000 to $140,000.

Fractional specialists in agentic commerce typically bill $150 to $300 per hour, which is often the right first move for teams under 50 people.

On transaction economics, the reference point most merchants cite is the retired Instant Checkout arrangement, where OpenAI charged a 4% transaction fee that combined with Stripe processing to reach an effective take rate near 9.2%, roughly one third of typical Amazon Marketplace costs.

Product discovery in ChatGPT Shopping currently carries no listing fee. Treat all of these as moving targets.

Fraud, Liability, and the Questions Nobody Has Answered

Cryptographic agent identity reduces chargebacks from unauthorized transactions by giving you proof of who initiated a request and with what intent. It does not settle liability.

If an agent buys the wrong item within a valid mandate, the signature is cryptographically sound and the dispute is commercial rather than technical.

Three controls are worth implementing before you accept meaningful agent volume: cap spend per session rather than per transaction, maintain an allowlist of verified agent operators, and require human confirmation above a threshold that matches the roughly $224 average authorized spend consumers report.

Also recalibrate fraud models, because agent traffic will trip velocity and device-fingerprint rules that were tuned on human behaviour.

Measure It Or You Will Not Believe It

AI referrals are frequently misattributed or buried in standard analytics setups, which makes the channel look smaller than it is and starves the project of budget. Instrument the following before you build anything:

• Verified agent sessions, segmented by operator and by browse versus pay intent.
• Feed error rate and attribute completeness, tracked as a percentage of catalog.
• Agent-initiated cart creation rate against completion rate.
• Time to propagate a price or inventory change to your public endpoints.
• Revenue per agent session compared to organic and paid search.

What Changes Next

Two shifts are already in testing. Cross-merchant orchestration, where an agent assembles a basket spanning multiple merchants from a single instruction, is being trialled through protocol extensions in 2026 with full production more likely in 2027. And agent-to-agent procurement, where a buyer's inventory system negotiates directly with a supplier's sales agent, moves the requirement from clean feeds to full API-level contract, pricing, and approval logic.

Gartner also expects around 40% of agentic commerce projects to be cancelled by 2027, which is the useful counterweight to the forecasts above. The projects that survive will be the ones that fixed data and APIs rather than the ones that chased a protocol announcement. Merchants building marketplace-style supply relationships should read this alongside our guide to building and scaling a B2B marketplace.

Conclusion

B2Bcentr covers agentic commerce as an engineering problem rather than a prediction market, because that is where the decisions actually get made. Making a checkout agent-ready means four sequential layers: a complete and consistently structured catalog, real-time commerce APIs, cryptographic agent identity through Trusted Agent Protocol or Agent Pay, and protocol-compliant checkout via ACP and UCP with AP2 supplying the authorization record. The market data supports urgency without supporting hype. AI-referred traffic is compounding fast, autonomous checkout remains a small and trust-constrained slice, and B2B procurement is the volume event on the horizon. The merchants who capture that spend will be the ones whose product data was machine-readable before the protocols stabilised, because every layer above it depends on that groundwork.

Read Next:

• Agent Payment Protocols Compared: AP2, ACP, x402 and MPP in 2026
• Does the EU AI Act Apply to Your AI SDR? Article 50 Explained (2026)
• What Is AI Search Visibility? A B2B SaaS Guide (2026)

FAQs

1. What does it mean to make a checkout AI-agent-ready?

Making a checkout AI-agent-ready means exposing four capabilities to autonomous agents: machine-readable product data with complete structured attributes, real-time APIs for availability and pricing, cryptographic agent identity verification, and protocol-compliant checkout endpoints that accept a scoped payment token. The merchant remains merchant of record throughout and keeps catalog, pricing, fulfilment, and dispute handling.

2. Which agentic commerce protocol should merchants implement first in 2026?

The agentic commerce protocol merchants should implement first in 2026 is ACP if buyers discover products inside ChatGPT, and UCP in parallel for Gemini and Shopify surfaces. Merchants supporting multiple protocols see roughly 40% more agentic traffic. Both depend on the same groundwork, so clean feeds and real-time inventory should be finished before either integration begins.

3. How much does it cost to make an ecommerce checkout agent-ready?

The cost to make an ecommerce checkout agent-ready typically ranges from about $55,000 to $300,000 for a mid-market merchant, split between catalog data remediation, endpoint development, and ongoing feed operations. Shopify merchants on paid plans inherit much of the integration. Fractional specialists billing $150 to $300 per hour are often the more efficient path for smaller teams.

4. How do merchants verify that an AI shopping agent is legitimate?

Merchants verify that an AI shopping agent is legitimate by checking cryptographic signatures attached to each request. Visa and Cloudflare's Trusted Agent Protocol uses three signed HTTP headers verified against an operator directory, distinguishing browse-only from payment intent. Mastercard Agent Pay binds cardholder, agent, and transaction into a single Agentic Token. Both use Web Bot Auth underneath.

5. Is agentic commerce worth investing in for B2B merchants?

Agentic commerce is worth investing in for B2B merchants because Gartner expects around 90% of B2B buying to route through AI agents by 2028, representing more than $15 trillion in spend, and Forrester projects one third of B2B payment workflows will involve agents by end of 2026. The prerequisite work, structured specifications and real-time APIs, delivers value regardless of protocol outcomes.

Disclaimer: This content is provided for informational purposes only and does not constitute legal, financial, or compliance advice. Protocol versions, governance arrangements, and partner counts cited here reflect publicly announced milestones as of August 2026 and are moving quickly. Adoption figures come from vendor and foundation announcements with differing methodologies and should be treated as directional signals rather than guaranteed outcomes.