By Carson Cobb. Data collected November 16 and 17, 2026.
AI agents are starting to shop for people. McKinsey’s 2026 survey found that the share of US AI users who completed a purchase with AI rose from 10% to 17% in a year (McKinsey, October 2026). Shopify said orders from AI searches grew 3.3 times year over year in Q3 2026 (PPC Land).
The checkout side is still settling. In March 2026 OpenAI shifted ChatGPT away from its own Instant Checkout toward product discovery, with merchants using their own checkout, and said Shopify product data reaches ChatGPT through Shopify Catalog with no extra setup (OpenAI, March 2026). Either way, agents need to find, understand and compare your products first.
Shopify has given every store the basic agent setup. I checked 617 stores to see what’s there by default and what stores have added.
The short version
- The defaults are everywhere. 100% respond on the UCP file, 97% have agents.md, 98% have llms.txt, 97% have an agentic discovery sitemap.
- Almost nobody has rewritten them. 94% of agents.md files use Shopify’s default template. 3% replaced it.
- Almost nobody blocks AI. 1% fully block any major AI crawler in robots.txt.
- The product data agents need is thinner. 23% put a GTIN in product structured data, and 11% put return or shipping terms there.
The agent setup, store by store
| Item | Stores | Share of 617 |
|---|---|---|
| UCP file responds (/.well-known/ucp) | 616 | 100% |
| llms.txt responds | 604 | 98% |
| agents.md responds | 601 | 97% |
| Agentic discovery sitemap listed | 600 | 97% |
| Shopify’s agent banner in robots.txt | 474 | 77% |
| agents.md uses Shopify’s default template | 583 | 94% |
| agents.md replaced with a custom file | 18 | 3% |
| Any major AI crawler fully blocked | 8 | 1% |
The product data agents read
| Item | Share of product pages |
|---|---|
| Product structured data | 86% of 598 |
| Price and availability in structured data | 96% and 97% of 516 |
| GTIN in structured data | 23% of 516 |
| Return policy or shipping details in structured data | 11% of 516 |
| Variant data with ProductGroup | 23% of 516 |
What this means
The gap isn’t access. Shopify opened the door for every store, and almost none closed it. The gap is what an agent finds when it walks in. It can read the price and stock. It usually can’t confirm the exact product across sellers, the return window, or which variant matches what the shopper asked for, unless it reads and interprets the page text.
A rewritten agents.md is the cheapest fix and the rarest. It’s a plain text file where a store can tell agents its real policies, which products suit which needs, and what not to claim. I covered what’s in the default in Shopify agents.md.
What to do with this
- Leave the defaults on. Don’t block AI crawlers unless you have a specific reason.
- Customize agents.md. Policies, product guidance, sizing help, and claims you can and can’t make.
- Fill GTINs and variant names. The data agents use to match the exact product.
- Publish return and shipping terms in Merchant Center and in structured data.
- Test it like a shopper would. Ask an AI assistant to find and compare your product, and see what it gets wrong.
Getting a store ready for AI agents is part of my AI Search and Conversion Audit. The AI Visibility Check is a quick first look.
How this was done
- Stores: 617 Shopify brand stores in six product areas: supplements (123), beauty (127), food and drink (154), pet (41), home (76) and apparel (96). This is my full dataset of 620 Shopify brand stores, minus client stores and one store that blocked automated reads.
- When: November 16 and 17, 2026.
- How: I requested robots.txt, agents.md, llms.txt, /.well-known/ucp and the sitemap index for each store. A file counted as the default template when it started with Shopify’s standard heading; a store that edited text inside the template would still count as default. A crawler counted as blocked when robots.txt disallowed its whole site. Product data came from the first product in each store’s public feed, read as the server first sends it.
- Limits: a file responding doesn’t mean an agent uses it. One product page per store.
No store or brand is named in this report.