Evidence checklist ยท reviewed 2026-09-08

AI Shopping Agent Readiness

Use this checklist to make product and policy data easier for AI shopping systems to discover, compare, and route to the right human-owned checkout flow.

Choose a store, a channel and an offer

A useful readiness assessment names a real workflow. For example: a US shopper buying one blue medium shirt through a particular assistant, with fulfillment handled by the existing store. Record the channel, market, offer ID, checkout destination and test date. A score for this sample is not proof that every product or region works.

Use an evidence log with four fields for each check: owner, observed result, reference or redacted screenshot, and next review trigger. Leave a check open if the evidence is only a vendor announcement or a feature you cannot access.

Shopify: verify the enabled sales channel

Start in the Agentic sales channel and record its management setting and individual channel controls. Compare an eligible variant's catalog preview with the storefront. Then verify where the shopper finishes the order: Shopify currently documents a store-checkout referral for ChatGPT, while some other channels support direct checkout. Test the route your store actually has.

Sources: channel controls and supported storefronts and checkout routes. The preview helps inspect your catalog; it does not guarantee how a shopping agent ranks or displays products.

WooCommerce: verify product data and session behavior

Compare a public product page with /wp-json/wc/store/v1/products, including currency, variation selection and stock. Follow the official Store API documentation for customer cart and checkout authentication. Keep administrative API credentials on the server. An accessible product endpoint is evidence of data access, not proof of ACP/UCP compatibility.

In a test environment, create a cart, change a quantity, make the item unavailable, and simulate payment failure. Check the payment extension's supported flow, order count and receipt. Save the result and extension version so that a later plugin update has a specific regression case.

B2B and mixed catalogs: test access boundaries first

Use a logged-out session to inspect a contract-priced product. Verify the response, page markup and feed do not expose negotiated offers or restricted products. Test buyer roles, approval limits, minimum quantities and whether an action requests a quote or creates an order.

Shopify's product eligibility documentation excludes B2B-only selling from agentic storefronts. Mixed catalogs need a separate DTC review, and custom price-hiding behavior needs its own public-data check. Do not score an unavailable B2B channel as ready because a DTC route works.

Build your platform-specific action plan

Engineering a protocol integration? Use the ACP/UCP comparison and acceptance test.

Discovery

Product URLs are discoverable

Product and category URLs are linked, included in sitemaps, and not hidden behind fragile client-only navigation.

Add product/category URLs to sitemaps and ensure important product pages render useful HTML.
Product Data

Product structured data is complete

Product pages expose price, availability, SKU, brand, images, reviews, and offer data where applicable.

Audit JSON-LD product markup and fill missing offer, availability, shipping, and return policy fields.
Product Data

Product feed is clean and current

Catalog feeds are up to date, deduplicated, and aligned with on-page product data.

Reconcile feed values with product pages and document update frequency.
Product Data

Agent search and checkout eligibility are explicit

Catalog data distinguishes products that agents may search from products they may purchase and includes required policy links.

Map search and checkout eligibility plus return, privacy, terms, and warning URLs to each agent-commerce feed or channel.
Agent Access

Crawler and AI-agent access is intentional

robots.txt, bot policies, and CDN rules reflect an intentional choice rather than accidental blocking.

Review AI/search bot access, document allow/block decisions, and test public pages from a neutral client.
Agent Access

AI-readable site notes exist

The site has clear machine-readable pointers such as llms.txt or equivalent documentation for key content areas.

Add llms.txt or a concise AI-readable guide that points to product, policy, and support resources.
Policy

Shipping, returns, and trust policies are explicit

Agents and users can find shipping fees, return windows, support contacts, and fulfillment constraints.

Publish structured policy pages and link them from product and checkout flows.
Protocol

Protocol ownership is assigned

A named team owns UCP/ACP monitoring, readiness decisions, and change review.

Assign an owner across ecommerce, SEO, engineering, and payments.
Checkout

Checkout, dispute, and support handoff are tested

A test flow covers agent-mediated checkout, user confirmation, failed payment, dispute handling, and support escalation.

Run a sandbox purchase through authorization, failure, dispute, and support escalation; document each owner and consent point.
Checkout

Agent purchase confirmation is explicit

The shopper confirmation point is clear for price, shipping fees, substitutions, returns, and payment credentials.

Define where shoppers approve price, payment, shipping fees, substitutions, and support handoff.
Monitoring

Agentic commerce changes are monitored

The team has a recurring process for tracking protocol, platform, and search changes.

Set a weekly review cadence and record source-backed changes in a changelog.
Monitoring

Measurement can separate AI-shopping signals

Analytics and logs can identify bot access, product-feed issues, and search/agent referrals where available.

Add log review, search query tracking, and event measurement for product discovery flows.