Industry benchmark · 2026-08-11
Magento stores are not ready for shopping agents.
We ran the same six agent-readiness checks against 1,370 live Adobe Commerce and Magento storefronts — no volunteers, no sign-ups, no self-selection. This is what an AI shopping agent finds when it arrives at a typical Magento store today.
Check by check
Where the storefronts fail.
Each bar is the share of stores passing that check, counted only across stores where the check could be decided. The two product-data checks are the ones that decide whether an agent can quote your price and promise delivery.
-
UCP-01 /.well-known/ucp manifest present and valid JSON 0%
0 pass · 1,286 fail · 84 not decidable
-
MCP-01 Advertised MCP endpoint reachable 0.1%
1 pass · 1,369 fail
-
BOT-01 robots.txt policy for AI shopping crawlers 44.6%
606 pass · 752 fail · 12 not decidable
-
LLMS-01 /llms.txt present 10.3%
138 pass · 1,205 fail · 27 not decidable
-
FEED-01 Product structured data completeness (25+ agent attributes) 6.8%
30 pass · 413 fail · 927 not decidable
-
SCHEMA-01 Product, Offer, AggregateRating JSON-LD on a product page 9.9%
44 pass · 399 fail · 927 not decidable
Distribution
How many checks a store passes.
Readiness is not a bell curve. Most storefronts sit at the bottom of the range, which is why a store that fixes even the product-data layer moves ahead of most of its category.
| Checks passed | Stores | Share | Bar |
|---|---|---|---|
| 0 / 6 | 663 | 48.4% | |
| 1 / 6 | 617 | 45% | |
| 2 / 6 | 73 | 5.3% | |
| 3 / 6 | 12 | 0.9% | |
| 4 / 6 | 5 | 0.4% | |
| 5 / 6 | 0 | 0% | |
| 6 / 6 | 0 | 0% |
Method
How this was measured.
The sample
Domains were assembled from public sources — the HTTP Archive / Chrome UX Report crawl, public top-site lists and agency portfolios — then verified as running Magento or Adobe Commerce by fetching the homepage and reading platform fingerprints. Stores were not contacted and did not opt in, so the sample is not biased toward merchants already interested in agentic commerce.
The checks
The same six checks the public scanner runs, unchanged: the UCP manifest, an advertised MCP endpoint, robots.txt policy for AI shopping crawlers, /llms.txt, product structured-data completeness against 27 agent-relevant attributes, and Product/Offer/AggregateRating JSON-LD on a sampled product page.
Refusals are not failures
When a store answered our request with 401, 403, 429 or 503, we recorded a refusal rather than a failed check — we never observed the artifact, so asserting anything about it would be a claim the merchant can see is wrong. Those 3 stores sit outside the pass-rate figures above.
What is not published
Only aggregates. No domain, brand or individual result appears on this page or in any material derived from it — a scanned store's result belongs to its owner, and we publish it only with permission. If your store is in this sample and you want its result, run the scan yourself and it is yours.
Your store
See where you sit against these numbers.
The scan is free, takes about fifteen seconds and runs the same six checks against your storefront. You get your score against this benchmark; the full findings come by email after an engineer has reviewed them.
Scan your store