A new Enhancement Proposal (EP) has been submitted to the Universal Commerce Protocol proposing standardized attribution signals. This addresses a critical gap: when an AI agent completes a purchase on behalf of a user, how does the merchant know which touchpoints influenced the conversion? The economics of agentic commerce depend on solving this.
Traditional e-commerce attribution works through tracking pixels, cookies, and click-through URLs. When a user clicks an ad, visits a product page, and completes checkout, the merchant can trace the conversion back to the original marketing touchpoint.
Agentic commerce breaks this model entirely:
The core problem: If merchants can't attribute conversions, they can't pay for advertising. If they can't pay for advertising, the economic model that funds product discovery breaks down.
The Attribution Signals Extension proposes adding optional metadata fields to UCP checkout sessions that capture conversion context:
Identifying where the purchase intent originated:
{
"attribution": {
"source_type": "agent_recommendation",
"source_id": "copilot:shopping-assistant",
"source_context": {
"query": "running shoes for marathon training",
"timestamp": "2026-02-14T08:00:00Z"
}
}
}Capturing the sequence of touchpoints that contributed to the decision:
{
"attribution": {
"influences": [
{
"type": "product_review",
"source": "running-magazine.com",
"timestamp": "2026-02-10T...",
"weight": 0.3
},
{
"type": "price_comparison",
"source": "agent:internal",
"timestamp": "2026-02-14T...",
"weight": 0.5
},
{
"type": "user_preference",
"source": "user:brand_affinity",
"weight": 0.2
}
]
}
}Critical to the proposal: users control what attribution data is shared:
| Level | Description | Data Shared |
|---|---|---|
none |
No attribution data | — |
aggregate |
Category-level only | Source type, general category |
detailed |
Full influence chain | All touchpoints with weights |
The proposal raises several technical questions that will need resolution:
If attribution signals come from agents, how do merchants verify they're legitimate? A malicious agent could claim false attribution to steal advertising revenue. This ties into the broader A2A cryptographic identity work.
An agent might discover products via MCP (querying a product catalog), reason about them using A2A (asking a specialist agent for advice), and complete the purchase via UCP. Attribution needs to span all three protocols.
User preferences and brand affinities might have formed months or years ago. Should those be part of the attribution chain? The proposal suggests a "recency decay" model where older influences have diminishing weight.
This proposal has significant implications for multiple stakeholders:
Attribution signals could enable a new model of AI-native advertising. Instead of buying clicks, advertisers might pay for "recommendation influence" — compensation when an agent cites their content as a factor in a purchase decision.
Agents that provide high-quality recommendations could receive attribution payments, creating an economic model for valuable shopping assistance beyond subscription fees.
Understanding which agent interactions drive conversions helps optimize the UCP integration. If most conversions come from voice assistants, that channel deserves more attention than chat-based agents.
The proposal is in early discussion phase. Key milestones to watch:
The proposal acknowledges these challenges and suggests a phased approach: start with simple source attribution, then expand to influence chains as the model matures.