How to check if your store is ready for AI shopping agents
A seven-point checklist for testing whether ChatGPT, Claude, Perplexity and other AI shopping agents can find, understand and buy from your online store, with a free automated checker.
Short answer: an online store is ready for AI shopping agents when an agent can (1) reach its pages, (2) read complete product data as structured data, (3) tell variants apart, (4) find the catalogue through a sitemap or feed, and (5) discover how to search, add to cart and check out through a protocol such as UCP, ACP or MCP. You can test all of this for free with the Proxa agent-readiness audit, which gives a score out of 100 and a list of fixes in about 15 seconds.
The rest of this guide explains each check, so you can also do it by hand.
What is an AI shopping agent?
An AI shopping agent is an assistant, such as ChatGPT, Claude, Perplexity or Gemini, that researches products and increasingly completes purchases on a shopper’s behalf. Instead of a person scrolling your category pages, the agent fetches your pages (or calls an API), reads the product data, compares it with other stores and reports back. If it can’t read your price, stock or sizes reliably, it recommends a store where it can.
The seven checks
The Proxa audit scores stores across seven weighted categories. Here’s what each one means and how to check it yourself.
| Category | Weight | What it asks |
|---|---|---|
| Structured data | 25 | Do product pages carry complete schema.org Product and Offer data? |
| AI crawler access | 15 | Does robots.txt or a firewall block AI assistants and AI search crawlers? |
| Protocol readiness | 15 | Does the store publish /.well-known/ucp, or discoverable ACP or MCP endpoints? |
| Catalogue completeness | 15 | Do products have descriptions, prices, images and stock status? |
| Variant data | 10 | Are sizes and colours machine-readable, not just dropdowns? |
| Feeds and sitemaps | 10 | Can an agent find every product through a sitemap or feed? |
| Basics | 10 | HTTPS everywhere and a fast homepage. |
1. Structured product data
Open a product page, view the source and search for application/ld+json. You want a Product with an Offer that includes at least: name, price, priceCurrency, availability, image, an identifier (gtin, sku or mpn), brand and description. Missing price, currency or availability is the most damaging gap, because an agent can’t recommend something it can’t price or confirm is in stock.
See Structured product data for AI shopping agents for a full example.
2. AI crawler access
Open https://yourstore/robots.txt. Look for Disallow: / under user agents such as ChatGPT-User, OAI-SearchBot, Claude-User, Claude-SearchBot, PerplexityBot or Perplexity-User. Blocking these stops assistants from reading your store when a shopper asks about it. Blocking training-only crawlers such as GPTBot or CCBot is a reasonable choice and costs much less.
Also check that your CDN or security plugin isn’t returning 403, 429 or 503 to bots in general. See robots.txt for AI crawlers for a ready-made file.
3. Protocol readiness
Agent commerce protocols let an agent go beyond reading pages:
- UCP (Universal Commerce Protocol): a store publishes a profile at
/.well-known/ucpdescribing its capabilities and endpoints for catalogue, cart and checkout. - ACP (Agentic Commerce Protocol): a checkout protocol that lets an agent create and complete checkout sessions.
- MCP (Model Context Protocol): a general way for AI assistants to call tools; a store can expose product search and stock as an MCP server.
Check by visiting https://yourstore/.well-known/ucp. If it returns a 404 or your homepage, agents can’t discover what they can do on your store. Very few independent stores publish these yet, so having them is an advantage.
4. Catalogue completeness
Agents compare products on facts. Pick five products at random and check each has a real description (not a single line), a visible price, at least one image and a stock status.
5. Variant data
If you sell sizes or colours, each one should exist as data, not just as a dropdown that JavaScript fills in. The best form is a schema.org ProductGroup with hasVariant, where every variant is its own Product with its own sku, price and availability, and variesBy names the dimensions (size, colour).
6. Feeds and sitemaps
Check that https://yourstore/sitemap.xml exists, that it lists product URLs (ideally in a dedicated product sitemap with lastmod dates), and that robots.txt contains a Sitemap: line pointing at it. A product feed (JSON, XML or CSV) linked from the homepage, or the WooCommerce Store API at /wp-json/wc/store/v1/products, also helps.
7. Basics
The whole store should be on HTTPS, and the homepage should start responding in under about 0.8 seconds. Agents work to tight time budgets and may give up on slow stores.
How to check your store automatically
- Go to the free agent-readiness audit.
- Enter your store’s address.
- In about 15 seconds you get a score out of 100, a breakdown by category, and the problems ranked by impact, each with a specific fix.
The audit reads only public pages: the homepage, robots.txt, sitemap, protocol endpoints and up to five product pages. It identifies itself as ProxaAuditBot/1.0, follows robots.txt and never submits forms or touches checkout. It works on WooCommerce, Magento, Shopify and custom stores.
What score is good?
- 80 and above: agents can read most of your store.
- 55–79: agents can read parts of your store, with gaps that cost you sales.
- 30–54: agents will struggle with your store.
- Below 30: most agents will skip your store today.
Common gaps are missing product identifiers, variants that exist only as dropdowns, and no protocol endpoints. The structured-data fixes are usually quick wins on WooCommerce and Magento.
Frequently asked questions
Is the agent-readiness check free?
Yes. The audit is free and needs no account. Repeat audits of the same store within 24 hours return the cached result.
Does it work for Shopify stores?
Yes, the audit runs on any store. Shopify already provides many agent features natively, so the report notes which fixes Shopify handles for you.
Will blocking GPTBot hurt my store in ChatGPT?
Blocking GPTBot only opts you out of model training. ChatGPT’s live browsing (ChatGPT-User) and search index (OAI-SearchBot) use separate user agents. Make sure your rules don’t block those by accident.
How is this different from normal SEO?
Classic SEO gets a page ranked for a person to click. Agent readiness makes the data behind the page usable by software that compares and buys on someone’s behalf. The two overlap heavily: structured data, sitemaps and fast HTTPS pages help both.