Agentic commerce readiness checklist for 2026

Making a store work for AI shopping agents sounds like a project. In practice it is a list: a few dozen specific things that are either right or not, most of them settings rather than code. This is that list, ordered by how much each item affects whether an agent can complete a purchase.

Work through it top to bottom. The checklist below saves your ticks in this browser, so you can come back to it.

How should you use this checklist?#

Top to bottom, because the groups are ordered by the stage of the purchase they unblock. An agent that cannot get past your bot manager never reads your product data, so fixing structured data first produces no visible result. The order follows the same logic as the revenue model in how much revenue you lose when AI agents can't buy: the first failing stage is the one that costs you.

Each item says why it matters and links to the fix. If you are not sure whether an item is true for your store, the free scan checks most of them for you.

0 of 30 done

1. Let agents in

• Verified AI agents are allowed on product, basket and checkout pagesThe most common reason agents fail. Allow signed agents before any rule that challenges automation. How to fix
• No CAPTCHA on browsing, basket or checkoutA CAPTCHA stops every agent. Keep it for sign-up and password reset. How to fix
• An unsigned request claiming to be an agent gets no further than a signed oneOtherwise a rule is trusting the user-agent string, which lets scrapers in. How to fix
• robots.txt allows search and assistant agents on catalogue pagesA leftover Disallow: / hides the whole store. How to fix
• Every page redirects to HTTPS in one stepAgents give up on long redirect chains and refuse insecure pages. How to fix

2. Make products findable

• Your sitemap lists product URLsCrawlers find the whole catalogue, not just what the homepage links to. How to fix
• Product pages are reachable by plain linksProducts only reachable through script-driven navigation are invisible to many agents. How to fix

3. Make product data readable

• Every product page has Schema.org Product data in the HTMLIt is how an agent reads what the product is without guessing. How to fix
• Each product has an Offer with a price and a currencyNo price block means the agent cannot price or compare the product. How to fix
• The price in the markup matches the price on the pageTwo different prices and the agent trusts neither. How to fix
• Availability uses a Schema.org value such as InStockFree text such as 'in stock' is not understood. How to fix
• Products carry a GTIN or MPN, and a brandLets an agent match the product across stores. How to fix
• Shipping cost and return policy are in structured dataAgents answer 'delivered by Friday, under $150?' from this. How to fix
• A return policy is declared for productsOne of the questions shoppers most often ask an assistant. How to fix

4. Feeds and protocols

• Your Merchant Center and ChatGPT feeds agree with your pagesFeed and page disagreeing on price or stock reads as an unreliable seller. How to fix
• Eligible products are opted into Google's agentic checkoutSet native_commerce(checkout_eligibility) in Merchant Center. How to fix
• A UCP profile is published, if your platform supports itHow Google's agent discovers what your checkout can do. How to fix
• The UCP profile's endpoints respondA profile pointing at dead endpoints is not support. How to fix

5. Checkout, payment and after the order

• Every button and link has an accessible nameAgents find 'Add to basket' by its name. An icon with no label is invisible. How to fix
• Every checkout field has a labelAn agent cannot fill a field it cannot identify. How to fix
• At least one payment method an agent can use is offeredExpress wallets and delegated payment let an authorised agent finish the purchase. How to fix
• An order-status page is linked from the storeAgents asked 'where is my order?' need somewhere to look. How to fix
• Customers can start a return without emailing a personA returns path an agent can find and use. How to fix

6. Trust

• No hidden text on pages instructs AI agentsHidden instructions are treated as prompt injection and erode trust. How to fix
• Reviews and comments are moderated for instructions aimed at agentsUser content can carry injection attempts on your behalf. How to fix
• Agents are shown the same prices as peopleDifferent prices for agents are found, and reported. How to fix

7. Measure and keep it working

• AI assistant referrals have their own channel in analyticsSo you can see how shoppers from ChatGPT and others convert. How to fix
• Agent traffic and refusals are visible in your CDN or bot managerAgents mostly do not run analytics scripts; logs show them. How to fix
• An agent test runs in your deploy pipelineBot rules and themes change weekly; catch regressions before customers do. How to fix
• Production is checked on a scheduleProduction bot rules differ from staging more often than teams expect. How to fix

Why does access come first?#

Because an agent that is refused at the door cannot be helped by anything else on this list. Bot managers are configured to stop automated traffic, and AI agents are automated traffic. Many stores admit agents on the home page and refuse them at the basket, where security rules are strictest.

The fix is narrow: allow verified agents, the ones that prove their identity with a Web Bot Auth signature, on the paths a shopper needs, and keep everything else in place. Our guide to why AI shopping agents get a 403 at your checkout has the steps for Cloudflare, Akamai, DataDome and others.

Which items have the biggest impact for the least effort?#

Bot rules, robots.txt, and the price block in your product template. Each is usually a single change that affects every product:

• One bot-manager rule that skips challenges for verified agents on commerce paths.
• One robots.txt group that lets search and assistant agents read your catalogue; see robots.txt for AI agents.
• One template fix that puts a complete Offer in your product page's JSON-LD; see product data AI agents can read.

The items that take longest are usually feed consistency across a large catalogue and returns flows that currently need a person. Start them early, but do not let them hold up the quick wins.

How often should you re-check?#

After every platform, theme, app or security change, and on a schedule in production. The items on this list break for ordinary reasons: a theme update drops structured data, a new app adds a checkout step, a security engineer tightens a bot rule after an attack. None of those looks like an agent change when it ships.

Re-running a checklist by hand after every change does not happen in practice. Put the checks in your deploy pipeline and run them against production on a schedule; testing your store with AI shopping agents in CI shows how.

What is not on this list?#

Things that sound important but rarely decide whether an agent can buy. llms.txt is the clearest example: 2026 studies found AI crawlers rarely request it, so it sits below everything here (see does llms.txt work?). Implementing every protocol yourself is another: most platforms provide UCP and feeds for you, and your effort is better spent on the data underneath. And no checklist replaces trying it: the most reliable test is an agent attempting the purchase.

Frequently asked questions

What should I fix first to make my store ready for AI agents?

Access. Make sure your bot manager allows verified AI agents on product, basket and checkout pages, and remove CAPTCHAs from the purchase path. Nothing else helps an agent that is refused at the door.

Do I need a developer to complete this checklist?

For most items, no. Bot rules, robots.txt, feed settings and many structured data fixes are configuration. Template changes and returns flows may need a developer.

Is llms.txt on the checklist?

No. Studies in 2026 found AI crawlers rarely request llms.txt and no clear link to AI citations, so it ranks below every item here.

How long does it take to make a store agent-ready?

The highest-impact items, bot rules, robots.txt and the product price block, often take a day. Feed consistency and self-service returns can take longer on large catalogues.

Does the checklist save my progress?

Yes, in your browser only, using local storage. Nothing is sent to us, and clearing your browser data resets it.

Sources

1. Merchant listing (Product, Offer) structured data — Google Search Central
2. Prepare your Merchant Center account for UCP — Google for Developers
3. Verified bots — Cloudflare docs
4. Overview of OpenAI crawlers — OpenAI for Developers

Written by the tragent team and checked against the sources listed above on October 19, 2026. See how we write and correct posts in our editorial policy.