1. The Silent Variant Drop: The Single Biggest Leak in AI Commerce
In our simulated missions (e.g., "Find waterproof trail running shoes in men's size 10 under $120 with fast delivery"), the most widespread point of failure occurred at Stage 2 (Understanding) and Stage 4 (Selection).
Shopify themes predominantly output a single default schema.org/Product JSON-LD block representing the primary SKU. Additional variants (sizes XS through XXL, alternative colorways) are rendered client-side via JavaScript dropdowns.
Frontier LLM RAG crawlers (GPTBot, OAI-SearchBot, PerplexityBot) discard client-side DOM mutations to preserve token budgets. If Size 10 is not explicitly declared as an individual schema.org/Offer with an unambiguous sku and InStock status, the agent concludes the product is unavailable and picks a competitor whose catalog explicitly exposes all sizes.
2. ISO Currency Drift & Geo-Pricing Ambiguity
41.6% of stores operating multi-currency markets (Shopify Markets, Geolocation apps) suffered currency hallucination. When an agent crawls a storefront from an Anycast IP in Northern Virginia, the theme may switch to USD or display converted prices without an ISO currency code anchor (cart.currency.iso_code).
When an AI agent prompts: "Is this item under £80?", an ungrounded currency value causes the agent to discard the candidate or hallucinate an inaccurate price, triggering immediate user bounce.
3. The Mandatory Policy Gate: MerchantReturnPolicy & Shipping
Both Google AI Overviews and OpenAI's commercial purchasing guidelines enforce strict eligibility criteria. Under Google Search Merchant specifications and autonomous shopping protocols, products lacking structured hasMerchantReturnPolicy and OfferShippingDetails objects become ineligible for enhanced AI product cards and high-confidence commercial discovery surfaces.
Merchants often have generous 30-day return policies written in their footer or FAQ, but because it is not codified in machine-readable JSON-LD linked to each offer, AI shopping agents cannot mathematically guarantee the buyer can return the item.
4. Empirical Methodology & Test Protocol
To eliminate bias and measure reproducible machine behavior, Relayeo Research executed this benchmark under the following test harness:
1,284 publicly accessible Shopify stores across Fashion & Apparel (412), Footwear & Athletics (284), Consumer Electronics (216), Home & Lifestyle (198), and Beauty & Cosmetics (174).
Queries simulated using headless agent sessions mirroring OpenAI GPTBot, Google Extended, and Perplexity Sonar token extraction parsers.
Classified as a failure if secondary variant sizes/colors were present in raw HTML/DOM but absent in machine-readable schema.org/Offer blocks.
Tested against Google AI Merchant Return Policy schema standards and OpenAI commercial buying API constraints.
Test whether AI shoppers discover or drop your products.
Run our free, zero-login AI Shopping Mission Simulator. Inspect your variant schemas, ISO currency mapping, and autonomous checkout readiness in 1.1 seconds.