In the previous article, we established a key judgment: Agentic Commerce's first battleground is not payment but discovery and recommendation. Whether an AI agent accurately describes your product and recommends your brand when answering user questions depends on a precondition — whether your product data is machine-consumable.
Traditional e-commerce optimization targets human consumers: beautiful product photography, compelling copy, smooth checkout experiences. B2A (Business-to-Agent) optimization targets AI agents: structured data formats, verifiable factual claims, real-time accurate pricing and inventory.
This is not either/or — you need both. But most brands have invested heavily in B2C optimization while leaving B2A almost entirely blank.
The Four Pillars of B2A
AI agents consume your product data through four primary entry points. Each entry point serves different agent scenarios, but together they form a complete "AI-readable" infrastructure.
Pillar 1: Schema.org Structured Markup
Schema.org is AI's fact-verification layer. When AI needs to answer "Is this brand trustworthy? Is the price reasonable? Are returns easy?" it relies on structured data embedded in your web pages.
Most brands' Schema markup remains at a basic level: a Product type with name and price. This is far from sufficient. AI agents making recommendation decisions need a deeply nested semantic graph — Product with Brand, Offer with ShippingDetails and MerchantReturnPolicy, AggregateRating with ReviewCount.
Common Schema problems and their consequences: missing availability means AI may recommend out-of-stock products; prices inconsistent with the actual cart price erode trust; missing AggregateRating puts you at a disadvantage in comparisons; missing return policy Schema reduces recommendation confidence. Partial Schema can be worse than none — AI sees incomplete data and assumes missing fields mean "not provided."
Actionable checklist:
- Validate every product page in Google Rich Results Test
- Confirm offers.availability syncs with actual inventory
- Confirm offers.price matches cart price (including promotional pricing)
- Add shippingDetails and hasMerchantReturnPolicy
- Add aggregateRating if you have a review system
- For Shopify stores: check whether your theme outputs complete JSON-LD — most Shopify themes' default Schema does not include shipping and return information
Pillar 2: llms.txt
llms.txt is AI's directory navigator. It tells AI: "If you can only read 20-50 pages from this website, these are the most important ones."
llms.txt is not a sitemap clone. Sitemaps list all pages for full-site crawling. llms.txt curates high-value pages for AI to quickly understand your brand and products within a limited context window.
The file should be placed at the root directory /llms.txt, formatted in Markdown, with a brand positioning summary at the top followed by categorized links — each with a one-line description that AI uses to decide whether to read deeper.
Common mistakes include: copying all sitemap URLs (curate 20-50 instead), listing links without descriptions, including login-required pages, placing the file in a subdirectory, and omitting a brand positioning summary.
Important note: llms.txt is not a requirement for Google AI Overview. Google's AI features primarily rely on Schema.org and Google Merchant Center data. llms.txt primarily serves ChatGPT, Perplexity, Claude, and other third-party AI platforms.
Pillar 3: Product Feed
Product Feed is the core data source for AI shopping. If Schema is structured data embedded in web pages, Product Feed is a product data stream independent of web pages — it enters AI systems' product indexes directly.
Major feed channels include Google Merchant Center (the data source for Google Shopping and Google AI Overview product recommendations), Shopify Agentic Storefronts (Shopify's AI channel management, activated by default in 2026), and OpenAI Product Data (which currently primarily ingests data through Shopify and Google Merchant Center streams).
Feed quality requirements span price accuracy (feed prices matching actual site prices), inventory synchronization (out-of-stock items promptly marked), variant completeness (each size/color/style submitted individually), image quality, GTIN/MPN identifiers, shipping and return information, and review data.
A common problem: many brands' feeds are manually uploaded or infrequently synced, meaning AI may see last week's prices and last month's inventory. In Agentic Commerce, inaccurate data equals no recommendation.
Pillar 4: UCP Manifest
UCP Manifest is the entry point for AI agents to discover a merchant's capabilities. Placed at /.well-known/ucp/manifest.json, it tells agents: "I support catalog queries, cart building, discount application, checkout initiation."
Current status: UCP remains early-stage, with Google and Shopify driving standardization. For most small to mid-size brands, UCP Manifest is not an immediate priority — Schema, llms.txt, and Product Feed take precedence. But for brands with annual revenue exceeding $5 million, monitoring UCP development and early preparation is worthwhile.
How the Four Pillars Work Together
Each pillar serves a different AI consumption scenario: Schema provides structured facts when AI reads your web pages; llms.txt provides quick orientation within limited context; Product Feed makes you discoverable in product searches independent of your website; UCP Manifest tells agents what transaction capabilities you support. None replaces another — they are complementary data layers.
The Gap Between "AI Readable" and "AI Recommended"
Having B2A infrastructure does not guarantee AI recommendation. It is a necessary condition, not a sufficient one. AI recommendation decisions also depend on brand fact accuracy in AI training data, competitor data completeness, AI platform commercial relationships, and user history and preferences. But without B2A infrastructure — missing Schema, stale Feed, no llms.txt — AI physically cannot read you, and the question of recommendation does not arise.
What Comes Next
B2A infrastructure enables AI to read you. But being readable does not equal bringing users to your website, and certainly does not equal generating orders. Between UCP's discovery end and ACP's checkout end lies an entire "evidence gap" — what AI recommended, whether users clicked, what happened after arrival, whether conversion occurred — that no protocol or platform can fully measure. In the next article, we dissect this gap and why it matters critically for brands.
FAQ
Q1: What is the difference between B2A and B2C?
A: B2C (Business-to-Consumer) is experience optimization for human consumers — beautiful images, compelling copy, smooth checkout. B2A (Business-to-Agent) is data optimization for AI agents — structured formats, verifiable facts, real-time accuracy. Both are needed simultaneously.
Q2: What are the most important Schema properties for e-commerce?
A: For e-commerce, the critical Schema properties are: Product.offers.price (accurate pricing), Product.offers.availability (inventory status), Product.aggregateRating (ratings), Product.offers.shippingDetails (shipping info), and Product.offers.hasMerchantReturnPolicy (return policy). Most brands are missing the last three.
Q3: Is llms.txt useful for Google AI Overview?
A: Limited. Google AI Overview primarily relies on Schema.org and Google Merchant Center data, not llms.txt. llms.txt primarily serves ChatGPT, Perplexity, Claude, and other third-party AI platforms. Do not position llms.txt as a "requirement for appearing in Google AI results."
Q4: If I use Shopify, how much B2A infrastructure is handled automatically?
A: Shopify auto-generates basic Schema (Product + Offer) but typically excludes shipping details and return policies. Shopify's Agentic Storefronts auto-handle Product Feed sync. llms.txt requires manual creation. UCP Manifest does not need manual handling currently — Shopify is integrating it. Overall, Shopify merchants' B2A infrastructure is approximately 40-50% complete out of the box.
Q5: What is the ROI of B2A optimization?
A: Direct ROI is higher AI recommendation probability. Indirect benefits include improved Google Rich Results display quality, better Google Shopping performance from enhanced Product Feeds, and future-readiness for UCP/ACP integration. Shopify data shows products appearing in AI catalog searches convert at approximately 2x the rate of general AI search — entering AI product indexes is itself a high-value action.