Score distribution
Number of stores in each score band (0–100).
0-39
0 stores
40-59
5 stores
60-79
31 stores
80-100
63 stores
By store size
Median score per size tier. Tiers are our classification (large brands, mid-size brands, small independents), not measured revenue. Tiers with fewer than 10 audited stores are not shown.
Large brands (n=29)
81
Mid-size brands (n=40)
82
Small independents (n=30)
83
The most common gaps
Share of measurable stores where the check is a gap (partial or fail), %. n = stores where the check could run. Agents are allowed in almost everywhere. The gaps are in the product data.
Shipping & return terms missing or partial in schema (n=99)
93%
Under 85% of products have 80+ word descriptions (n=97)
80%
Ratings not (fully) in structured data, or no reviews (n=99)
72%
Brand / GTIN / SKU incomplete in schema (n=99)
71%
Product images missing alt text (n=99)
68%
Variant barcodes incomplete (n=99)
60%
Product pages don't clearly state shipping & returns (n=99)
45%
Shipping policy empty or thin (n=70)
36%
Product type not set (n=97)
16%
Price not in the HTML on every sampled page (n=99)
15%
What agents can and can't see
Share of measurable stores, %.
Search vs training. Blocking AI training crawlers (GPTBot, Google-Extended, CCBot) is a separate choice and doesn't remove a store from AI answers. Blocking search and user agents (OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot, Claude-User, Googlebot, Bingbot, Amazonbot) does.
Every check, pass / partial / fail
Number of stores per check. "n/a" means the check couldn't run on that store (not measurable). It never counts as a fail.
robots.txt lets AI search & user agents reach products
99/99 pass
Firewall/bot protection lets AI agents load the store
0/99 pass
Sitemap lists products
94/99 pass
llms.txt present (with store-specific facts)
93/99 pass
Agent commerce endpoint (UCP manifest + MCP catalog/checkout tools)
97/99 pass
Product JSON-LD on product pages
91/99 pass
Offers carry price, currency and availability (matching the live price)
87/99 pass
Every variant is described (one Offer per variant or ProductGroup/hasVariant)
87/99 pass
Brand + product identifiers (GTIN/barcode, SKU/MPN)
29/99 pass
Review ratings exposed as AggregateRating
28/99 pass
Shipping & return policy in structured data
7/99 pass
Organization/WebSite schema on homepage
88/99 pass
Descriptions have enough facts for an agent (80+ words)
19/99 pass
Variant option names are meaningful
85/97 pass
Product type / category set
81/97 pass
Price visible in the page HTML (not only via JavaScript)
84/99 pass
Product images have alt text
32/99 pass
Variants have barcodes (GTIN/UPC/EAN)
40/99 pass
Refund Policy has real content
62/99 pass
Shipping Policy has real content
45/99 pass
Product pages state shipping & returns
54/99 pass
Contact details findable
96/99 pass
Standard add-to-cart form with variant IDs
80/99 pass
Accelerated checkout (Shop Pay etc.) on product pages
92/99 pass
No password page, age gate or storefront challenge
98/99 pass
passpartialfailn/a
Check your own store
- Shipping and returns: one plain line on each product page (only what your policy says), real Shipping and Refund policy text, and
shippingDetails/hasMerchantReturnPolicyin your product schema. - Descriptions: 80+ words on your main products: materials, sizes, who it's for, care.
- Ratings: view the page source. If the star rating isn't in the HTML, turn on your reviews app's structured-data / rich-snippet setting.
- Identifiers: fill in variant barcodes (GTIN/UPC/EAN) and make sure brand and GTIN appear in the product schema.
- Access: check
yourstore.com/robots.txtdoesn't disallow AI search agents from/products/, and that bot-protection apps don't challenge them.
Method
- Sample: 100 Shopify stores (99 measurable), chosen by hand. About 30 large brands, 40 mid-size and 30 small independents across the US, UK, EU, Australia/NZ and Canada, in 16 categories. It isn't a random sample of all Shopify stores, so read every figure as "in this sample".
- Data: public storefront data only (robots.txt, product pages and JSON, structured data, policies). No logins, no orders, rate-limited requests.
- Audit: run on 3 October 2026 from a UK IP address. Checks read the server-rendered HTML and structured data, which is what agents that don't run JavaScript see. We didn't test firewalls or bot protection. Each store was scored out of 100 across access, structured data, product content, policies and path to purchase. Checks that couldn't run are excluded from that store's score.
- Publishing: aggregate results only. No store is named. Slices under 10 stores aren't shown.
Get your store checked
We'll run the same audit on your store and send you the top fixes, free. Reply to our post on X with your store URL, or check it yourself with our free Agent Skills, which work with any AI assistant that reads SKILL.md (Claude, OpenAI Codex, Gemini CLI, Cursor, GitHub Copilot and others).
Ask for a free audit on X → Free skills →