Agentic Commerce: When AI Agents Buy on Your Customer's Behalf — And Never Visit Your Site

Adobe Analytics tracked a 4,700% year-over-year increase in generative AI traffic to US retail sites between July 2024 and July 2025. During Black Friday 2025, AI-driven shopping traffic surged 805% year-over-year. Those numbers have been quoted by every e-commerce executive in North America. What most of those executives have not yet internalized is that the next phase does not send traffic to their sites at all.

In Q1 2026, OpenAI and Stripe announced the Agentic Commerce Protocol (ACP), a standard for AI agents to complete purchase transactions programmatically — without a browser, without a product detail page visit, without a checkout flow. Early partners include Shopify merchants, Walmart, and Etsy. The transaction completes inside the agent. The retailer's website is never opened.

This is not a future scenario. It is infrastructure deployed today, and the brands that understand what it means are restructuring their entire marketing and merchandising stack around a single question: when an AI agent is the buyer, what makes it choose you?

What Agentic Commerce Actually Is in 2026

The term "agentic commerce" is used loosely, so let's define it precisely. Agentic commerce is the execution of a purchase transaction by an AI agent acting on a human user's delegated authority, without the human directly interacting with the retailer's digital surfaces.

The anatomy of a typical agentic transaction in 2026 looks like this:

1. A user delegates a task to an AI agent: "Order me more of the protein powder I ran out of last week. Keep it under $45. Make sure it ships in two days."
2. The agent queries the user's purchase history to identify the product category and past preferences.
3. The agent queries structured product APIs from multiple sources — the brand's own catalog API, a marketplace feed, a comparison aggregator — to find candidates matching the constraints.
4. The agent evaluates candidates against explicit constraints (price, delivery window) and implicit constraints (brand trust, review signals, return policy quality).
5. The agent selects the best match, calls the payment API with the user's stored credentials, and confirms the order.
6. The user receives a notification that the order was placed.

The retailer's product detail page was never visited. No marketing pixel fired. No conversion funnel was entered. The decision was made entirely from structured data and entity signals the agent had access to before the transaction began.

This is a fundamentally different purchase process than anything that has existed before — not just in e-commerce, but in retail history. The marketing funnel, in the traditional sense, does not apply.

The Transaction Without a Visit

The implications for brand and marketing strategy are severe and largely unacknowledged in the industry. Consider what does not happen in an agentic transaction:

• The brand's hero images and lifestyle photography are never seen.
• The landing page copy and value proposition are never read.
• The reviews section is never scrolled.
• The upsell and cross-sell modules are never triggered.
• The email capture popup is never shown.
• The retargeting pixel never fires.
• The influencer-partnership landing page never loads.

Every dollar invested in visual merchandising, CRO, email capture, and retargeting infrastructure returns zero in an agentic transaction. The entire conversion optimization stack built over the last fifteen years is bypassed.

What does happen in an agentic transaction:

• The brand's structured product data is queried.
• The brand's real-time pricing and availability are checked.
• The brand's return and shipping policies are evaluated for machine-readability.
• The brand's entity reputation — its prior probability of being a credible source in its category — is consulted.
• The brand's checkout API is called.

Five touchpoints replace the forty-touchpoint funnel. Four of those five touchpoints are infrastructure. The fifth — entity reputation — is the only one that functions like traditional brand equity, and it works through a completely different mechanism than any brand-building tactic previously optimized for.

Which Categories Are First to Go Agentic

Not all categories transition to agentic commerce at the same rate. The determining factor is the degree to which the purchase decision can be resolved from structured data without subjective experience.

Consumer electronics is the clearest early case. When a user asks an agent to "buy me a USB-C hub that works with my MacBook Pro and has at least four ports under $60," the agent can fully resolve that decision from structured catalog data. There is no aesthetic judgment required, no physical experience needed, and the product specifications are either compliant with the constraints or they are not. The agent executes.

Household consumables are the highest-volume early category. Replenishment purchases — the protein powder, the laundry detergent, the printer paper — are precisely the transactions that users most want to delegate. The decision criteria are narrow (same product or functionally equivalent substitute), the price sensitivity is low relative to the value of automation, and the risk of a wrong choice is recoverable. According to McKinsey's 2026 agentic commerce analysis, consumables and household goods will account for the largest share of agentic transaction volume by 2028.

Product Data API Requirements: What Agents Actually Need

The question most operators are asking is: "What do I need to expose?" The answer is more specific than "structured data." Agents have distinct data quality standards, and the failure modes are sharper than they are in human-facing search.

Complete attribute sets. Human shoppers forgive missing product attributes — they rely on images, infer from category context, or ask customer service. Agents do not forgive missing data. An attribute field that is null, incomplete, or inconsistent across variants causes the agent to either skip the listing entirely or flag it with low confidence. In a competitive category where multiple brands have complete data, missing attributes are a disqualifier. The average product catalog has roughly 23% of SKUs with incomplete attribute data, according to Salsify's 2026 product data quality report. In agentic commerce, that 23% is invisible by definition.

Real-time pricing and availability. Agents checking product catalogs validate that the data they receive is current. If an agent queries a product, selects it, and then receives a "price changed" or "out of stock" error at checkout, it logs the brand as an unreliable source. Repeated failures cause systematic deprioritization. The acceptable latency between an inventory or pricing change and its reflection in the catalog API is roughly 15 minutes — the same standard that Google Shopping enforces for Shopping Ads. Brands whose catalog data lags significantly beyond that window will see agentic transaction routing move to competitors with fresher feeds.

Machine-readable policies. Return policy, shipping policy, and warranty terms are decision inputs for agents. A return policy written as a prose paragraph in an HTML div is not parseable. A return policy exposed as structured JSON — with fields for return window, restocking fee, and accepted return conditions — is. Agents systematically assign a lower policy confidence score to brands whose terms require natural language parsing, because the interpretation error rate is higher. The practical implication: if your return policy is "see our full policy at [link]," you are at a data quality disadvantage versus the brand that exposes a structured policy object in their API response.

Checkout and payment API compatibility. The transaction itself requires an API endpoint. The three primary checkout protocols in use by agents in 2026 are the OpenAI/Stripe ACP, Shopify's Storefront API with agent permissions, and the emerging Agent-to-Merchant (A2M) standard being developed by the Commerce Working Group. Brands on Shopify or major e-commerce platforms gain ACP compatibility through their platform. Direct-to-consumer brands using custom checkout infrastructure need to build an explicit API layer. Brands that do not expose a programmatic checkout endpoint are simply excluded from the transaction, regardless of how good their product data is.

For a deeper treatment of how structured product data and schema markup interact with AI discovery, the schema markup and entity context framework provides the technical foundation that underpins the agentic data layer.

Pricing and Availability Real-Time Signals

The freshness requirement for agentic commerce creates an operational challenge that most e-commerce teams have not fully internalized. Traditional e-commerce operates on a crawl-and-cache model: Google or a comparison engine crawls your product feed periodically, and the data served to users may be hours or days old. Human shoppers see a price, click through, and discover the current price at checkout — friction that is accepted as normal.

Agents do not accept that friction. An agentic transaction is designed to complete without a human confirming the final details. If the data the agent used to make the selection does not match the data at checkout, the transaction fails, the user is notified of a failure they expected would not happen, and the brand's data reliability score drops.

The operational requirement is therefore not just "accurate data" but "accurate data in near-real-time." For brands with dynamic pricing — promotions, flash sales, clearance — this means the pricing feed needs to update as frequently as pricing changes. For brands with perishable or limited inventory, availability signals need to reflect real-time warehouse status, not day-old reports.

Several brands have solved this through event-driven architecture: pricing and inventory changes trigger an immediate API event that updates the catalog feed, rather than relying on scheduled batch jobs. The additional infrastructure cost is modest — it amounts to a webhook layer on top of existing inventory management systems. The commercial cost of not having it is that your product gets deprioritized by agents that have logged your feed as a source of inaccurate data.

Payment and Checkout APIs: The Final Gate

The checkout layer is where most direct-to-consumer brands currently fall down. Building for agentic commerce requires exposing a programmatic checkout API that agents can call with a pre-validated cart and stored payment credentials, completing the transaction without a human entering card details.

The key implementation considerations:

1. Authentication and agent identity. The checkout API needs to accept agent authentication tokens, not just human login credentials. The OpenAI/Stripe ACP uses a specific token format that brands need to support. This is not complex — it is roughly equivalent to implementing an OAuth integration — but it does require deliberate implementation rather than being included by default in standard checkout platforms.

2. Cart validation before commitment. A robust agentic checkout API validates the entire cart — inventory, pricing, shipping availability, policy compliance — before committing the transaction. Agents expect a validation step that returns either a confirmed transaction or a structured error with a reason code. Checkout APIs that return unstructured error messages or silent failures cause agents to retry or abandon, both of which erode brand reliability scores.

3. Confirmation and fulfillment webhooks. Once a transaction is committed, the agent needs a structured confirmation payload it can present to the user. Order ID, expected delivery date, and total cost are the minimum. Agents that receive ambiguous or partial confirmations flag the transaction as uncertain, which creates a poor user experience and reduces the brand's agentic commerce routing frequency.

4. Returns and cancellation APIs. The complete lifecycle of a transaction includes returns and cancellations. Brands that expose a returns API — allowing an agent to initiate a return on the user's behalf with the same frictionlessness as the original purchase — have a measurable advantage in agent-driven replenishment categories. Users who delegate purchasing also want to delegate returns, and the agent's ability to deliver that complete lifecycle drives the preference signal for brands that support it.

Return Policy Machine-Readability: The Underrated Differentiator

Return policy has always been a conversion factor in human e-commerce. The standard insight — "free returns increase conversion" — is well documented. In agentic commerce, return policy functions differently. It is not a conversion nudge; it is a machine-read data field that influences shortlist generation.

The distinction matters because the machine does not respond to marketing framing. "Hassle-free returns within 30 days" is marketing copy. An agent sees it as a string to parse. The structured equivalent — `{"return_window_days": 30, "return_shipping_cost": 0, "restocking_fee": 0, "conditions": ["unused", "original_packaging"]}` — is a data object the agent can evaluate precisely against the user's preferences.

Brands that have invested in structuring their policy data gain a concrete advantage when agents compare two otherwise similar products. The brand with structured policies is evaluated accurately. The brand with prose policies is evaluated with uncertainty — and agents in high-confidence decision mode favor certainty.

The implementation path is straightforward. Policies already exist in most brands' content management systems. The work is extracting the key variables into a structured JSON object and exposing it in the product API response alongside price and availability. Total engineering effort: typically four to eight hours for a standard checkout platform integration.

The Brand-First Agentic Decision

There is one factor in agentic commerce that does not reduce to data fields: entity authority. When an agent is choosing between two products with similar structured data, similar pricing, and similar policies, it falls back on a prior belief about the brand's trustworthiness and category relevance. That prior belief is formed from training data — from the density and authority of mentions, reviews, and citations in the corpus the agent's underlying model was trained on.

This is the agentic commerce equivalent of the share-of-model problem. Brands with high entity authority get the benefit of the doubt. Brands with weak entity authority do not get the benefit of the doubt — and in a commodity category with multiple competent suppliers, the benefit of the doubt is often the deciding factor.

The entity authority mechanism in agentic commerce works through three channels:

Training data density. How often and in what context has the brand been mentioned in the web content the model was trained on? A brand with thousands of positive review mentions across Reddit, tech media, and consumer publications has a higher prior probability of being a good choice than a brand mentioned a handful of times. This is not manipulable in the short term — it is a function of the brand's actual presence in authoritative digital spaces over time.

Review signal aggregation. AI systems synthesize review signals from multiple platforms — Amazon, Google Shopping, Trustpilot, Reddit, and vertical-specific review sites. A brand with hundreds of recent positive reviews across multiple independent platforms has a higher entity confidence score than a brand with excellent on-site testimonials but sparse third-party coverage. The trust signals, reviews, and UGC analysis documents how this aggregation works in detail.

Citation in comparison content. When an agent has been trained on comparison content that positions a brand favorably in its category — "brand X is the best option for Y use case" — that positioning becomes a prior in the agent's brand evaluation. Comparison pages on vendor domains, third-party listicles, and media coverage of category rankings all feed this prior. Brands that have invested in agentic-era comparison content show measurably higher brand-preference rates in agentic selection tasks.

What "Discovery" Means Without a Screen

In traditional e-commerce, discovery is a visit — a product detail page that a human reads, evaluates, and either converts on or exits from. In agentic commerce, "discovery" is the moment an agent's query returns a product candidate. It is invisible to the brand unless the brand has instrumentation on its catalog API.

This creates a measurement gap that most operators have not yet bridged. The funnel metrics that brands currently report — sessions, page views, time on site, conversion rate — capture none of the agentic discovery process. A brand could be appearing in thousands of agentic shortlists per day, converting at a high rate, and reporting that data to leadership as "direct traffic" with an unknown source, because the agentic transaction never generated a session.

The measurement approach for agentic commerce requires three new instrumentation points:

1. Catalog API query logging. Every time an agent queries your product catalog API, that is a discovery event. Logging query volume, query attributes, and response characteristics gives brands visibility into agentic demand signals that are invisible in standard analytics. The agent querying your API is the equivalent of a human landing on your category page — it represents expressed intent.

2. Checkout API entry rate. The ratio of catalog queries that result in a checkout API call is the agentic conversion rate. This metric tells you how often your product is shortlisted versus actually selected. A low checkout-to-query ratio indicates that your product passes initial filtering (it appears in shortlists) but is losing the selection step — typically due to pricing, policy terms, or attribute gaps.

3. Attributed revenue by agent source. Checkout API calls should include an agent source identifier when supported by the protocol. OpenAI/Stripe ACP includes source metadata that brands can use to attribute revenue to specific agentic channels. Tracking this over time reveals which agent ecosystem is routing the most transactions to your brand, and allows targeted optimization for high-volume sources.

For GA4 users, the AI search referral tracking setup guide covers the current state of analytics configuration for AI-driven traffic, including what agentic referral data looks like in practice.

Building for the Agent Economy: The 6-Step Playbook

If you are an e-commerce operator with a meaningful portion of revenue in agentic-ready categories, here is the prioritized implementation sequence:

1. Conduct a catalog data completeness audit. Pull your entire product catalog and evaluate attribute coverage by SKU. For each product category, define the minimum required attributes for agentic evaluation (electronics: processor, RAM, storage, connectivity specs, dimensions; apparel: material, care instructions, size chart with numeric measurements; consumables: weight/volume, ingredients/materials, compatibility). Identify and remediate the SKUs with missing or inconsistent attributes. A realistic timeline for a mid-size catalog (5,000 to 50,000 SKUs) is four to eight weeks.

2. Implement schema.org Product and Offer markup with real-time signals. Deploy JSON-LD markup on all product pages with dynamically populated pricing and availability fields — not baked-in values that go stale. The `availability` property should reflect live inventory (In Stock / Out of Stock / Limited Availability). The `price` property should reflect the current unit price with any active promotions applied. This step bridges the gap between your structured data layer and the web crawlers that feed AI training pipelines.

3. Expose a structured policy API. Build a machine-readable policy object for your return and shipping terms. At minimum, expose return window, return shipping cost, restocking fee (if any), and return conditions. Shipping policy fields should include carrier options, estimated delivery windows by region, and cutoff times for same-day or next-day dispatch. Surface this policy object in your product API response alongside price and inventory.

4. Establish real-time inventory and pricing webhooks. Instrument your inventory management and pricing systems to push updates to your catalog API feed within 15 minutes of any change. For brands on Shopify, this is natively available via Shopify's inventory webhooks and the Storefront API. For brands on custom infrastructure, this typically requires adding a webhook event emitter to the inventory management system and a feed refresh trigger on the catalog API.

5. Enable agentic checkout protocol support. Implement support for the OpenAI/Stripe ACP if you sell through ChatGPT-integrated channels, and enable agent permissions on Shopify Storefront API if you are on Shopify. For DTC brands on custom checkout infrastructure, build a programmatic checkout API endpoint that accepts agent authentication tokens, validates the cart, processes payment with stored credentials, and returns a structured confirmation payload. Engage your payment processor about agent payment token support — Stripe, Adyen, and Braintree all have developer documentation for agentic payment flows.

6. Instrument agentic funnel metrics. Add logging to your catalog API and checkout API to capture query volume, shortlist-to-checkout rate, and attributed agentic revenue. Build a weekly dashboard tracking these metrics alongside traditional e-commerce conversion metrics. The first three months of data will establish your baseline agentic conversion rate and identify the drop-off points that represent the highest optimization opportunity.

What Operators Should Do This Quarter

The transition to agentic commerce is not a future disruption to prepare for — it is a current revenue channel that most operators are participating in imperfectly, without measuring it, and without optimizing for it. Three actions matter most in the near term:

Fix catalog data completeness now. This is the foundational requirement and the one with the longest lead time. A catalog data audit and remediation project for a mid-size catalog takes months, not days. Every week of delay is a week of agentic transactions routing to competitors with complete data. The ROI on catalog data quality work in an agentic commerce context is immediate: each additional SKU with complete attribute coverage is a candidate that was previously invisible to agent queries.

Get on the ACP and Shopify agent infrastructure. The OpenAI/Stripe Agentic Commerce Protocol is live and processing transactions today. Shopify merchants can enable agent permissions through the Storefront API with a configuration change. If you sell on Shopify and have not enabled agent API access, you are leaving agentic transactions on the table with minimal implementation cost to capture them.

Start measuring agentic discovery. You cannot optimize what you do not measure. Even before full agentic checkout infrastructure is in place, brands can add logging to catalog API queries to understand how much agentic demand they are already seeing. That data makes the business case for the larger infrastructure investment in checkout API support and policy structuring.

The brands that win agentic commerce are not necessarily the largest brands or the best-funded ones. They are the brands with the most complete, accurate, and machine-readable product data infrastructure. In a transaction where no human reads a landing page, the landing page cannot save you — but a complete, fresh, structured catalog API absolutely can.

Takeaway: Agentic commerce collapses the marketing funnel into a single citation decision made entirely from structured data. When an AI agent buys on a customer's behalf, it never visits your product page, never sees your hero image, never reads your value proposition copy. It queries your catalog API, evaluates your pricing and policies as machine-readable data, checks your entity authority against training-data priors, and either routes the transaction to you or to a competitor whose data is more complete. The brands that win this era built their product data infrastructure for machines — complete attribute sets, real-time pricing and inventory, structured policy APIs, and programmatic checkout support. The brands that invested only in human-facing conversion optimization are entering the decade with the wrong assets.