Home / Blog / How AI agents buy

How do AI agents buy from online stores?

4 min read · Foundations · 12 October 2026

AI agents buy from online stores by querying structured protocol endpoints across access, comprehension, trust, transactability, and reliability layers rather than rendering visual pages. Instead of scanning HTML markup for keywords, autonomous crawlers read machine manifests to verify real-time inventory, pricing, and checkout parameters before executing purchases.

The machine-to-machine shift in B2B procurement

Gartner projects 90% of B2B purchases will route via AI agents by 2028, routing over $15 trillion machine-to-machine. B2B buyer workflows are moving from manual search inputs to automated agent queries.

Data from Adobe Analytics shows AI traffic to U.S. retail sites grew 393% YoY in Q1 2026 across over 1 trillion visits. When an autonomous crawler acts on behalf of a buyer, it does not browse categories or evaluate hero banners. It requests machine endpoints directly. Storefronts that fail to expose structured discovery protocols are filtered out during initial evaluation.

To understand the protocol requirements behind this shift, read our overview on what is agentic commerce.

Why traditional SEO fails autonomous AI crawlers

Standard search engine optimization focuses on ranking HTML pages for human clicks. Keywords, meta tags, and surface-level schema markup inform answer engines for text citations, but they fail when an agent attempts transaction execution.

When an AI agent evaluates a product, it looks for deterministic execution rules. A page returning standard Schema.org JSON-LD may state a price, but if the backend cart endpoint requires dynamic session cookies or returns an unhandled 404 on API checkout requests, the agent aborts.

Consider, in simplified form, what an agent expects when it requests /.well-known/ucp:

{
  "version": "2026-01",
  "capabilities": [
    "dev.ucp.shopping.checkout",
    "dev.ucp.shopping.fulfillment"
  ],
  "payment_handlers": [ "..." ],
  "signing_keys": [ { "kty": "EC", "crv": "P-256", "...": "..." } ]
}

A simplified illustration of what the manifest declares (capabilities, schema version, payment handlers and signing keys), not the full specification.

If your server returns raw HTML or an unparsed redirect, the crawler marks the store non-transactable. Learn more about core protocol specs in our guide on what is UCP.

Multi-layer protocol conformance for storefront transactability

To complete an end-to-end purchase, an AI agent requires explicit conformance across five distinct technical layers:

1. Access layer: exposes machine-readable endpoints without aggressive bot-blocking or JavaScript rendering requirements.
2. Comprehension layer: delivers unambiguous product specifications, attributes, and unit economics.
3. Identity and trust layer: validates merchant identity and cryptographic signatures, such as the signing keys a UCP manifest publishes and AP2 payment authorisations.
4. Transactability layer: provides programmatic checkout paths, tax calculations, and payment handling.
5. Reliability layer: maintains continuous state synchronization between store inventory databases and agent queries to prevent price or stock discrepancies.

If any layer breaks, the agent skips the merchant. A failure in reliability, such as stock reporting discrepancies between catalog read and cart submission, triggers an immediate checkout drop.

Preparing your Magento storefront for agentic purchasing

Magento Open Source storefronts do not include native machine syndication rails. Unlike closed platforms that syndicate through proprietary apps, Magento stores require explicit, store-by-store protocol configuration.

Static schema plugins and point-in-time scanners generate audit reports, but they do not maintain real-time endpoint state or prevent protocol drift. As protocol specifications update and store catalogs change, continuous readiness monitoring keeps endpoints compliant. Review implementation steps in our guide to Magento agentic readiness.