Skip to main content
Agentic commerce is what happens when an AI assistant — ChatGPT, Google Gemini, Perplexity, or a brand’s own copilot — can recommend and complete a purchase on behalf of a user. For your store to participate, four things need to be true: AI can find your products, you speak the right commerce protocol, AI can trigger a checkout, and you can see and measure what AI drove. Asva AI covers all four.

​
The agentic commerce stack

  1. Discovery — The AI can find your products via catalog feeds, AEO/GEO, and .well-known/ucp
  2. Protocol compliance — You speak the right commerce language (UCP or ACP)
  3. Checkout — The AI can trigger a payment on your behalf
  4. Attribution — You can see and measure what AI drove

​
Universal Commerce Protocol (UCP)

UCP is an open standard backed by Google that lets AI assistants discover, evaluate, and purchase from any merchant using a shared commerce language. When you implement UCP, Google AI Mode, Gemini, and any UCP-compatible agent can browse your catalog and complete purchases without the user leaving the AI experience. How it works: 
AI Agent
  │
  ├─ 1. Fetches /.well-known/ucp  →  discovers your capabilities
  ├─ 2. Calls /ucp/products       →  browses your catalog
  ├─ 3. Calls /ucp/checkout       →  initiates a cart/session
  ├─ 4. Triggers payment          →  your PSP processes it
  └─ 5. Gets order status         →  confirms purchase to user
Where UCP surfaces:
  • Google AI Mode (Shopping)
  • Google Gemini
  • Any UCP-compatible AI agent
Key files you need to host:
  • /.well-known/ucp — capability manifest (required; publicly accessible, no auth)
  • Product catalog endpoint — your structured product feed
  • Checkout endpoint — creates and completes orders
UCP builds on Google Merchant Center. If you already have a Merchant Center feed, you’re halfway there. See the Google Merchant Center guide.

​
Agentic Commerce Protocol (ACP)

ACP is OpenAI’s protocol for enabling ChatGPT and other OpenAI-powered agents to initiate and complete purchases via tool calls — without the user leaving the conversation. How it works: 
User: "Buy me those running shoes"
  │
ChatGPT
  ├─ Calls search_products tool   →  your catalog endpoint
  ├─ Presents options to user
  ├─ User confirms
  ├─ Calls initiate_checkout tool →  your checkout endpoint
  └─ Purchase completes in-conversation
Where ACP surfaces:
  • ChatGPT (Instant Checkout)
  • Any OpenAI-powered agent with function calling
Key concepts:
  • Tool definitions — JSON schemas that describe your commerce tools to the agent
  • Checkout state machine — create → update → complete flow
  • Idempotency — safe retries with idempotency keys to prevent duplicate orders
ACP and UCP are complementary. Many brands implement both for maximum coverage across Google and OpenAI surfaces. See UCP vs ACP for guidance on which to prioritize.

​
Dark agentic traffic

When a user finds your product on ChatGPT or Gemini and then visits your site directly, that session typically arrives with no referrer — and lands in GA4 as “Direct” or “(not set)”. This is dark agentic traffic: real, high-intent visits that your analytics can’t see. Why this happens:
  • AI assistants use server-side requests that strip the browser referrer
  • Referrer headers are removed by ChatGPT’s click proxy
  • UTM parameters are rarely appended by AI surfaces
  • GA4’s default channel grouping has no “AI” category
What Asva does: Asva’s attribution snippet analyzes session fingerprints, timing patterns, and UTM signals to identify AI-sourced sessions and surface them as a distinct channel in your analytics. 
Without Asva:          With Asva:
Direct → 35%           Direct → 18%
Organic → 28%          Organic → 28%
Paid → 22%             Paid → 22%
                       AI Commerce → 17%  ← previously hidden
See Dark traffic attribution for implementation details.

​
Readiness Score

The Readiness Score is Asva’s 100-point audit of your site’s agentic commerce readiness. Run it against any domain via the Readiness API or the free browser tool. What gets checked:
CategoryWhat we checkWeight
Discoverability.well-known/ucp, feeds, entity schema30%
CatalogProduct data quality, variants, pricing25%
CheckoutUCP/ACP endpoint health, payment support25%
AttributionAsva snippet, GA4 custom dims, dark traffic20%
Score grades:
ScoreGradeMeaning
80–100AAgent-ready. Appearing on major AI surfaces.
60–79BMostly ready. A few gaps to close.
40–59CPartially implemented. Missing key endpoints or feeds.
0–39DNot yet agent-ready. Start with the quickstart.

​
AEO and GEO

Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) are the practices of making your brand and products easy for AI systems to find, cite, and transact with. For commerce, that means:
  • Structured product data (JSON-LD Product and Offer schema)
  • Entity consistency — the same brand name, legal entity, and address everywhere
  • UCP/ACP protocol compliance so agents can complete purchases
  • Evidence blocks with official sources so AI cites you, not competitors
See the full AEO/GEO guide for implementation tactics.

​
Glossary

TermDefinition
UCPUniversal Commerce Protocol — Google’s open standard for agentic checkout
ACPAgentic Commerce Protocol — OpenAI’s protocol for ChatGPT Instant Checkout
.well-known/ucpCapability discovery manifest; must be hosted at your domain root with no auth
Dark trafficAI-sourced sessions misattributed as Direct in analytics
AEOAnswer Engine Optimization — structured data and entity clarity for AI discoverability
GEOGenerative Engine Optimization — visibility in ChatGPT, Gemini, and Perplexity
Readiness ScoreAsva’s 100-point agentic commerce audit
Full definitions at asva-ai.com/glossary.