Aizii Learn · AI Search
← Back to Learn

How to check if AI agents can buy from your store

An AI agent can buy from your store only if it can do three things without a human in the loop: find the product, verify the price and stock from machine-readable data, and reach a working checkout. If any one of those fails, the agent either buys blind or stops. The check below walks through each one in order, using the same signals an agent reads, so you find the failing step instead of guessing.

Why agents fail to buy

  1. The product page has no Product markup, so the agent reads a price from the visible page with no authoritative stock signal to check it against.
  2. The markup exists but disagrees with the page. The price in the code is an old sale price, or the availability says InStock for a variant that is sold out.
  3. A crawler is blocked. The search bot that would surface the product cannot fetch the page, so the product never reaches the agent at all.
  4. A bot wall or CAPTCHA stops the agent at the door. The page loads for a human and fails for an automated visitor.
  5. The checkout needs a login before it shows shipping or a total. An agent with no account cannot get past that step.
  6. Shipping, tax, or returns are missing or only appear after the customer enters an address. The agent cannot compute a total, so it cannot complete the order.

The check, in order

1. Read the product page the way an agent reads it

Open a live product page, view the page source, and search for application/ld+json. Look for one Product block with a name, an image, a description, a brand, and an offers block with a price, a currency, and an availability value. Then compare every value with the visible page. The price in the markup must match the price a shopper sees. The availability must match the actual stock for that variant. An identifier such as a GTIN, MPN, or SKU helps an engine match your product to the same item elsewhere, so fill it in where you have one. If the page has two Product blocks with different prices, that is a fail on its own. Pick one source of truth and remove the other.

2. Check that search crawlers can fetch the page

Open your robots.txt file at [yourdomain.com]/robots.txt, replacing yourdomain.com with your actual domain, and confirm it allows the search crawlers that feed AI answers: OAI-SearchBot, Googlebot, PerplexityBot, and ClaudeBot. OpenAI says a site that opts out of OAI-SearchBot will not be shown in ChatGPT search answers, and that a robots.txt change takes about 24 hours to take effect for search. GPTBot is the separate training crawler, so allowing one does not automatically allow the other. ChatGPT-User is different again: OpenAI says it is used for user-triggered actions, not automatic crawling, and robots.txt rules may not apply to it, so do not treat allowing it as the fix for a blocked search crawl.

3. Check for a bot wall

Load a product page in a fresh browser session with no login and no cookies, and watch for a CAPTCHA or a block page. Then check your security and CDN settings for rules that challenge automated visitors. A bot wall that stops an agent is a fail even when the markup is perfect, because the agent never reaches the page to read it. If you run bot protection, confirm it has an allow rule for the search crawlers you allowed in step 2, and check the vendor docs in Sources for the current path before changing a rule.

4. Check the discovery aids

Confirm your store has an llms.txt file at the root and a product sitemap that lists your products. Neither file makes an agent buy on its own, but without them the agent has no catalog summary to orient from and no complete list of product pages to work through. A store with no llms.txt and no product sitemap forces the agent to discover products one page at a time.

5. Walk the checkout as a guest

Add a product to the cart, go to checkout without logging in, and confirm you can see shipping options, tax, and a total before any payment step. If the store forces an account before it shows shipping, that is a fail for an agent with no account. Confirm the shipping and returns policies are published pages with working links, and that a support contact is listed. Google's UCP guide makes the same point for agent checkout: return policies and customer support information must be set, because the merchant remains the seller of record.

6. Check the protocol paths, and confirm their current status

Two protocol paths can let an agent buy without leaving its own interface. The first is Google's Universal Commerce Protocol, which needs a Merchant Center account in good standing with approved products, return policies and support information set in Merchant Center, and the checkout eligibility attribute set on the products you want to offer. The second is OpenAI's Agentic Commerce Protocol, which is an approved-partner path, not a switch you turn on for every store. For both paths, confirm the current availability for your store with the platform before you build against them, since availability and rollout change. A store that has not set up either path can still be bought from by an agent that browses like a human and checks out on the site, provided the markup and the guest checkout pass the steps above.

7. Test with a real prompt, and read the result carefully

Ask an assistant for one of your products by name, then ask for its price and whether it is in stock. Note what it cites and whether the price and stock match your site. Then ask it to buy the product and watch where it stops. A real abort is an agent genuinely attempting a purchase and failing, for example because it cannot verify the price or the checkout blocks it. Friction alone, such as a slow page or an extra click, is not an abort. Do not claim a measured walkthrough rate from one test. One prompt tells you where this product fails today. Repeat the test across products and assistants to find the pattern.

The fundamentals

Say you run a bike accessories brand

You pick your best-selling bike light and run the check. The page source shows a Product block with the right name and price, but the availability says InStock for every color, and the red one has been sold out for weeks. Your robots.txt allows OAI-SearchBot and Googlebot. The checkout works as a guest, and shipping shows before payment. The fix is the availability value: mark the red variant OutOfStock in the data that feeds the markup, then retest the same prompt and confirm the assistant no longer offers the red one. One field, fixed at the source, is what turns a blind buy into a verified one.

What this guide does not cover

This guide covers checking whether an agent can buy from your store today. It does not cover building a protocol checkout integration, which is engineering work with the platform, or the monitoring work of keeping prices and stock accurate as inventory changes. It also does not cover your own brand agent, which sells from the fixed data this check verifies.

Check your own brand now

Run the free scan. It takes seconds, no signup, and shows whether AI crawlers can read your product pages today: aizii.ai/aeo-scan

Run the free scan

Sources

Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.