AI shopping agents are coming - and most e-commerce sites will be invisible to them. Your beautiful product pages, carefully crafted navigation, and optimized checkout? Agents can't process any of it. Here's the practical roadmap to making your store agent-ready before your competitors do.
Agent readiness has three layers: First, your content must be machine-readable - agents need structured data, not HTML designed for human eyes. Second, your systems must distinguish humans from agents and route them appropriately. Third, you need customer context - enriched profiles that power personalization regardless of whether a human or agent is shopping.
Think of agent readiness as a technology stack with three layers:
Most merchants focus only on the bottom layer - making content readable. But without the top layers, you're serving generic experiences to everyone. Let's build from the bottom up.
llms.txt is a proposed standard that gives AI agents a quick overview of your site. Think of it as robots.txt for AI - a single file that tells agents what your site offers and how to navigate it.
Place this file at your site root: https://yourstore.com/llms.txt
# YourStore
> Premium outdoor gear and apparel for adventure enthusiasts.
## Main Sections
- [Products](/products.md): Full product catalog with categories
- [Deals](/deals.md): Current promotions and discounts
- [Shipping](/shipping.md): Delivery options and policies
- [Returns](/returns.md): Return and exchange policies
- [Support](/support.md): Customer service and FAQ
## Optional
- [Size Guide](/size-guide.md): Detailed sizing information
- [Care Instructions](/care.md): Product care guidelines
The file should include:
Google added llms.txt to Chrome Lighthouse audits in May 2026. Run a Lighthouse audit to check if your file is properly formatted and accessible.
For each page linked in your llms.txt, create a clean Markdown version that agents can process easily. Store these at the same URL with a .md extension.
# Product Catalog
## Categories
### Jackets
- [Ultralight Rain Jacket](/products/ultralight-rain-jacket.md) - $189
- [Insulated Puffer](/products/insulated-puffer.md) - $249
- [Softshell Hoodie](/products/softshell-hoodie.md) - $159
### Footwear
- [Trail Running Shoes](/products/trail-runners.md) - $139
- [Hiking Boots](/products/hiking-boots.md) - $219
## Current Deals
- 20% off all jackets with code WINTER20
- Free shipping on orders over $100
Each product page should also have a Markdown version with structured information agents can parse.
You probably already have some structured data for SEO. Enhance it for agents:
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Product",
"name": "Ultralight Rain Jacket",
"description": "Waterproof, breathable jacket weighing just 200g",
"sku": "URJ-2024-BLU",
"brand": {"@type": "Brand", "name": "YourStore"},
"offers": {
"@type": "Offer",
"price": "189.00",
"priceCurrency": "EUR",
"availability": "https://schema.org/InStock",
"priceValidUntil": "2027-01-31"
},
"aggregateRating": {
"@type": "AggregateRating",
"ratingValue": "4.7",
"reviewCount": "342"
}
}
</script>
Key additions for agent readiness:
Machine-readable pages are good; APIs are better. Create endpoints that agents can call directly:
Search, filter, and retrieve product details programmatically.
OpenAPI specAdd items, apply coupons, calculate totals.
REST endpointsSubmit orders, process payments, confirm purchases.
Secure, authenticatedReal-time stock levels, store availability.
Webhooks for updatesDocument your APIs with OpenAPI (Swagger) specifications. Agents can read these specs to understand how to interact with your store.
Once your content is machine-readable, you need systems that can detect and route different types of traffic.
AI agents identify themselves differently than traditional bots. Look for:
/.well-known/poppy.jsonTraditional bot management blocks automated traffic. For agentic commerce, you want to route agents to appropriate experiences, not block them. A blocked agent is a lost sale.
The Personal Agent Protocol (Poppy) is emerging as the standard for agent-business interactions. Implement the discovery file:
{
"version": "1.0",
"name": "YourStore",
"description": "Premium outdoor gear retailer",
"capabilities": {
"browse": true,
"purchase": true,
"returns": true,
"support": true
},
"authentication": {
"type": "oauth2",
"authorization_endpoint": "/oauth/authorize",
"token_endpoint": "/oauth/token"
},
"channels": {
"api": "/api/v1/openapi.json",
"web": "https://yourstore.com"
}
}
This tells agents:
Serve different experiences based on who's visiting:
| Request Type | Experience | Content Format |
|---|---|---|
| Human browser | Full website with visuals, navigation, UX | HTML, CSS, images |
| Agent (browsing) | Machine-readable pages, structured data | Markdown, JSON-LD |
| Agent (transacting) | API endpoints for cart, checkout, orders | JSON via REST/GraphQL |
| Agent (authenticated) | Personalized data based on customer context | Enriched JSON responses |
This is where most merchants stop - and where the real competitive advantage lies. Knowing an agent is legitimate isn't enough. You need to know about the customer behind the agent.
When a customer (or their agent) authenticates, immediately enrich their profile:
// 1. Agent authenticates via Poppy/OAuth
const customerEmail = authToken.email;
// 2. Enrich with SocialScore
const profile = await socialScore.enrich(customerEmail);
// 3. Profile now includes 500+ signals:
{
"email": "customer@example.com",
"social_presence": { ... },
"interests": ["outdoor activities", "sustainability"],
"behavioral_signals": { ... },
"customer_value_indicators": { ... },
"risk_signals": { ... }
}
// 4. Personalize the experience
const personalizedCatalog = generateCatalog(profile);
This enrichment powers personalization whether the customer browses directly or sends an agent.
Use enrichment data to build actionable customer personas:
These personas should be available to agents in machine-readable format, allowing them to communicate relevant information to their users.
When an authenticated agent requests your catalog, include personalized data:
{
"customer": {
"loyalty_tier": "gold",
"points_balance": 2450,
"personalized_discount": "15%"
},
"recommendations": [
{
"product_id": "URJ-2024",
"reason": "Based on your interest in lightweight gear"
}
],
"recent_orders": [
{"order_id": "12345", "status": "delivered"}
],
"catalog": {
"featured_for_you": [...],
"deals_matching_interests": [...]
}
}
The agent can now tell the customer: "You have 2,450 loyalty points. There's a 15% discount on jackets you'd like, and your last order was delivered."
Not everything needs to happen at once. Here's a prioritized roadmap:
| Priority | Action | Effort | Impact |
|---|---|---|---|
| High | Add llms.txt file | 1 hour | Agent discovery |
| High | Enhance JSON-LD structured data | 1-2 days | Product understanding |
| High | Review bot management rules | 2 hours | Stop blocking agents |
| Medium | Create Markdown page versions | 1 week | Content accessibility |
| Medium | Implement customer enrichment | 1-2 weeks | Personalization foundation |
| Medium | Add Poppy discovery file | 1 day | Agent authentication |
| Lower | Build full API layer | 1-3 months | Full agent transactions |
| Lower | Implement dual routing | 2-4 weeks | Optimized experiences |
SocialScore fills the customer context layer of your agent readiness stack:
When a customer authenticates - whether browsing directly or through an agent - SocialScore instantly enriches their profile with:
This context powers personalization for your chatbots, recommendation engines, and - soon - external shopping agents acting on behalf of your customers.
Test SocialScore's customer enrichment API - see what 500+ signals reveal about your customers.
Free SocialScore Check Book a Demo