New Grok Bot commerce auditor measures if shopping agents can actually complete purchases on your store without pretending to be human.

The bot scores 5 areas (20 pts each, 100 total):

1. Access — Can identified agents load pages? Does robots.txt match actual firewall behavior?

2. Catalog — Machine-readable products, prices, availability, SKU/GTIN?

3. Total price — Item + shipping + tax + delivery estimate, not just list price

4. Policy — Numeric returns window, shipping terms, MerchantReturnPolicy markup

5. Checkout — Can reach cart/checkout without spoofing human, including UCP profile or agentic protocol?

Grades A-D. Only A gets "Grok Bot Commerce Optimized" badge.

Amazon scored D (37/100) in first live test:

- Homepage loaded, but product pages inconsistent (Listerine OK, Pantene/Colgate threw HTTP 503 challenges)
- Catalog not machine-readable (0/20)
- Landed cost partial (12/20)
- Returns/policies (0/20)
- Checkout without spoofing (5/20) — no UCP profile at /.well-known/ucp, cart/checkout blocked to identified agents

The fix list is concrete:

- Allow identified agents on product pages, not just homepage
- Publish Product and Offer JSON-LD
- Ship shipping/tax in structured data or numbered policy
- State numeric returns window
- Keep cart/checkout readable without captcha for shopper agents
- Add UCP profile at /.well-known/ucp
- Add llms.txt pointing to feed and policies

If your store blocks honest agents, you lose the sale silently. Agents will route to merchants who publish the facts.