The Protocol War and the Evidence Gap: AI Commerce Infrastructure's Missing Layer

Over the past twelve months, a quiet standards war has unfolded in AI commerce infrastructure. Three protocols appeared nearly simultaneously, each backed by some of the technology industry's most influential companies:

UCP (Universal Commerce Protocol) — driven by Google and Shopify. Positioned as the "discovery layer" of AI commerce: product catalog queries, inventory checks, price comparisons, product detail retrieval. When an AI agent helps a user find "waterproof running shoes under $200," UCP provides the standardized data interface.

ACP (Agentic Commerce Protocol) — led by OpenAI, with deep Stripe integration. Positioned as the "execution layer" of AI commerce: add-to-cart, initiate checkout, process payment, confirm order. When an AI agent completes a purchase on behalf of a user, ACP provides the secure transaction execution pipeline.

AP2 (Agent Payments Protocol) — proposed by Google. Positioned as the "authorization layer" of AI commerce: cryptographically signed agent authorization, payment permission delegation, audit chains. When an AI agent needs to spend money on a user's behalf, AP2 provides the cryptographic proof that "I am authorized to do this."

These three protocols are building toward the same vision from three different starting points — discovery, execution, authorization: AI agents can complete the full commercial transaction cycle on behalf of humans.

What the Protocols Solve, and What They Do Not

Each protocol is precise within its domain:

• UCP knows which product catalogs an agent queried and under what filter conditions
• ACP knows which checkouts an agent initiated and what the payment status is
• AP2 knows whether an agent's authorization is valid, how much was spent, and whether the audit chain is intact

But between them lies an enormous evidence gap.

A typical AI commerce path looks like this:

• User asks AI: "Find me waterproof running shoes"
• AI agent calls UCP to query multiple merchants' product catalogs → UCP observable
• AI synthesizes information and recommends three options to the user → Answer layer — protocol cannot observe this
• User clicks an AI-recommended link and arrives at a merchant's website → Visit layer — protocol cannot observe this
• User browses product pages, compares sizing charts → Early Commerce layer — protocol cannot observe this
• User decides to buy; AI agent initiates checkout → ACP observable
• AI agent requests payment authorization → AP2 observable
• Order completes → ACP/AP2 observable

The problem is steps 3–5: the entire span of human behavior between AI discovery and AI checkout — recommendation, arrival, browsing, comparison — is completely invisible to the protocols.

This is not a design flaw in the protocols. Protocols solve machine-to-machine interoperability: how to let agents safely query catalogs, execute transactions, and verify authorization. They do not concern themselves with (and should not need to concern themselves with) what humans do on websites.

But from a commercial measurement perspective, this gap is fatal.

If you only have UCP query data and ACP transaction data, you know "AI queried the catalog 100 times" and "AI completed 3 orders." You do not know what happened between 100 queries and 3 orders: How many queries produced user-visible recommendations? How many recommendations were clicked? How many clicks generated meaningful browsing? How many browsing sessions reached checkout?

Without data from these intermediate layers, you cannot optimize anything. You do not know if the problem is recommendation quality (AI recommended the wrong products), landing page experience (users arrived but did not convert), or price competitiveness (users compared and chose a competitor).

AIAA Is the Solution to the Evidence Gap

This is precisely why the five-layer AIAA framework exists. AIAA does not attempt to replace any protocol — it fills the measurement gap between protocols:

AIAA's five layers — Answer, Request, Visit, Commerce, Attribution — precisely fill the measurement vacuum between UCP's product queries and ACP's transaction execution.

Why This Gap Matters

In traditional e-commerce, an analogous measurement chain is mature: Google Ads click → landing page → product page view → add-to-cart → checkout → order. Every step has well-established tracking tools (GA4, GTM, Shopify Analytics, Facebook CAPI). Advertisers can precisely measure the efficiency of each dollar spent at every funnel stage.

In the AI commerce path, this chain does not yet exist. Protocols define both endpoints (discovery and checkout), but the measurement layer in between — what AI recommended, how users responded, what happened on-site — is blank.

This blank is producing real commercial consequences:

• Brands cannot attribute AI revenue. You know Shopify says AI orders grew 13x, but you do not know your own AI-attributed revenue — because your analytics tools cannot see whether users arriving via AI recommendation completed purchases.

• You cannot optimize the AI channel. Traditional channels have clear optimization levers: ad copy, bidding strategy, landing page design, checkout flow. The AI channel's optimization levers are unclear — partly because you do not know where the funnel leaks.

• You cannot make AI vs non-AI ROI comparisons. If you cannot measure each step's efficiency in the AI channel, you cannot compare it on equal terms with Google Ads, Meta Ads, or Email. AI will lose in the resource competition — not because it is actually less efficient, but because it lacks measurement infrastructure.

The "Commercial Evidence Layer" for Agentic Commerce

I said in the first article of this series: the future of e-commerce is an AI discovery, AI recommendation, AI arrival, AI cart, AI transaction, and AI attribution commercial evidence layer.

Protocols (UCP/ACP/AP2) provide the foundational pipes. But pipes are not measurement. A highway enables cars to travel from A to B, but you still need a traffic monitoring system to know how many cars are on the road, where they came from, where they are going, and how many arrived at their destinations.

AIAA is the traffic monitoring system for the AI commerce highway.

Protocol Interoperability Reality

It is worth noting that these three protocols are currently in early stages with limited interoperability:

• UCP and ACP are led by different companies (Google/Shopify vs OpenAI/Stripe) with data formats and integration approaches that are not fully compatible
• AP2 is still at the proposal stage with no large-scale commercial deployment
• No protocol provides a cross-protocol unified measurement interface

This means that for the foreseeable future, measuring AI commercial impact will not be automatically solved by protocols. You need an independent, cross-protocol measurement framework. AIAA is designed to complement protocols, not depend on them:

• If you have integrated UCP, AIAA's Request and Visit layers supplement post-query user behavior data
• If you have integrated ACP, AIAA's Attribution layer can join ACP transaction data with AI-sourced sessions to produce complete AI-attributed revenue
• If you have integrated neither, AIAA still works independently — it is based on your own website logs, Edge data, and first-party tracking

MCP and Measurement Infrastructure

There is another perspective worth discussing: the relationship between MCP (Model Context Protocol) and measurement infrastructure.

MCP is not a commerce protocol — it is a communication standard for AI agents calling external tools. When ChatGPT connects to your Shopify backend via MCP to read product data, it uses MCP's tool-calling interface. Meta Ads MCP lets agents read ad account data. Shopify MCP lets agents read product and order data.

From a measurement perspective, an MCP tool call is an observable event. OpenTelemetry GenAI Semantic Conventions have already defined execute_tool and invoke_agent as standard trace spans. Every MCP tool call can be recorded as: who called, what tool was invoked, what was the input, what was the output, did it succeed.

This means when Agentic Commerce protocols execute through MCP tool-calling, AIAA's Request layer can directly consume MCP tool call trace data. This is the most likely early integration path between AIAA and the protocol layer.

Series Summary: AIAA's Five Core Arguments

Looking back across all five articles, the AIAA framework distills into five core arguments:

• Measure actions, not agent counts. Agent identity is unstable, incomparable, and does not equal business value. AIAA measures evidence-backed actions. (Article 1)

• Measure in layers, do not blend attribution. AI exposure, machine requests, human arrivals, commercial behavior, and attributable revenue are five independent fact layers. Combining them into one number is false attribution. (Article 2)

• Distinguish real growth from visibility improvement. Every tracking upgrade causes AI metrics to spike. This is observability improvement, not necessarily AI growth. Comparable growth is real growth. (Article 3)

• Upgrade your evidence infrastructure. GA4 sees only the tip of the AI iceberg. The Evidence Ladder from L0 to L3 unlocks new observation layers at each level. Most teams should immediately upgrade from L0 to L1.5. (Article 4)

• AIAA is the missing layer between protocols and commercial evidence. UCP/ACP/AP2 define AI commerce pipes but do not measure what happens inside the pipes. AIAA fills this evidence gap, enabling you to attribute AI's true contribution to revenue. (Article 5)

Next Steps

If you are a CEO or CMO: Add AIAA to your monthly business review. Have your team start reporting AI impact in layers, not as a single ambiguous "AI traffic" number.

If you lead growth: Deploy L1 first-party JS this week and L1.5 Edge Lite within two weeks. You will see the true scale of AI agent activity for the first time.

If you are a data engineer: Start planning the session-to-order join data pipeline. This is the foundation of the Attribution layer and the key to eventually producing an "AI-attributed revenue" number.

If you run GEO optimization: Start Answer-layer SOV sampling. Understanding how AI describes your brand when answering user questions is more important than optimizing one keyword ranking.

AI is not the future — it is already in your server logs today. The question is whether you can see it.

FAQ

Q1: What is the relationship between UCP, ACP, and AP2?

A: They are three complementary protocols in the Agentic Commerce space. UCP (Universal Commerce Protocol, Google/Shopify) handles AI discovery — product queries, catalog searches, price comparisons. ACP (Agentic Commerce Protocol, OpenAI/Stripe) handles execution — checkout, payment, orders. AP2 (Agent Payments Protocol, Google) handles authorization — cryptographic agent delegation and audit chains. They are led by different companies and currently have limited interoperability.

Q2: Does AIAA require integrating with these protocols?

A: No. AIAA is an independent measurement framework based on data you already have (website logs, Edge logs, GA4, e-commerce platform data). It does not depend on UCP/ACP/AP2. But when your business integrates these protocols, AIAA can incorporate protocol-layer data to enhance Commerce and Attribution layer depth.

Q3: What exactly is the "evidence gap"?

A: The gap between AI discovery (UCP product query) and AI checkout (ACP transaction execution) — the entire span of user behavior including whether AI recommended your brand, whether users arrived at your site, what they did on-site, and whether those behaviors led to purchases — is not covered by any protocol. AIAA's Answer, Request, Visit, and Commerce layers precisely fill this gap.

Q4: When will these protocols mature enough for large-scale use?

A: UCP has initial implementations from Google and Shopify. ACP has OpenAI and Stripe backing but remains early stage. AP2 is still at the proposal stage. Full protocol interoperability and large-scale commercial deployment may require 18–36 months. But you do not need to wait for protocol maturity to start measuring AI impact — AIAA can be deployed independently right now.

Q5: What is Gravity's position in the Agentic Commerce protocol ecosystem?

A: Gravity is not a protocol maker. Gravity's position is as the AI commercial evidence layer — through the CitationGraph platform, providing cross-protocol, cross-platform AIAA measurement capability. We help brands answer one core question: "What did AI do for my business?" The answer to that question is not in any single protocol — it is in the measurement gap between protocols.

Q6: If I can only do one thing right now, what should it be?

A: Deploy L1 first-party JS plus L1.5 Edge Lite Bridge. These two steps can be completed within two weeks at minimal total cost, but they will let you see the true scale of AI agent activity for the first time. Jumping from L0 (GA4 baseline) to L1.5 will increase your AI visibility 5–10x. This is the highest-ROI first step.