ARC Benchmarks
Open, versioned benchmarks of agentic commerce. Each one answers one question, on a fixed method, with a dated figure and its denominator. Data is CC-BY-4.0.
There is no single combined ARC score. The benchmarks measure different things (agents, stores, protocols, channels, claims), and a blend would hide the gaps that matter. A planned benchmark shows no figure.
Versioning and reproducibility rules → · Open dataset · Submit an agent run
betaasi-v0.1 · Weekly (planned for v1)
ARC Agent Shopping Index
Which shopping agents can get a product into the cart on real stores, and at what cost?
Attempt: one agent, one store, one fixed product URL. Reach-checkout rate with a 95% interval, plus cost and steps per success.
57.6%95% CI 52.8–62.3%
239 of 415 browser cart attempts reached the cart, across 83 stores with a browser test.
Snapshot 2026-11-06 (as of 2026-11-06) · store_answers.csv · browser_cart_test
Five ARC browser shoppers on the stores with a browser test. A first per-model pilot is published: 3 models, 35 non-Shopify stores, two runs, with 95% intervals. The full per-model panel is not published yet.
Pilot, November 2026: can AI models reach checkout on their own?
Methodology · Data · Leaderboard
betasri-v0.1 · About weekly, with each open snapshot
ARC Store Readiness Index
How ready is each store, and each platform, category and market, for AI shopping agents?
Store. Pass, fail or not measured on each readiness measure, grouped into find, understand and reach checkout.
79.0%95% CI 78.1–79.9%
6,600 of 8,352 tested stores give agents a readable price and stock on the product page.
Snapshot 2026-11-06 (as of 2026-11-06) · store_answers.csv · can_agents_read_price_and_stock
The snapshot carries the find, price-and-stock and checkout answers. Most scorecard measures are not in the snapshot yet.
Methodology · Data · Leaderboard
livepci-v1.0 · With each open snapshot
ARC Protocol Conformance Index
For UCP, ACP, MCP, WebMCP and x402: how many stores declare it, how many are valid, and at how many does an agent get a checkout link back?
Store × protocol × stage (declared, valid, usable, transacts), each with an evidence label.
67.5%95% CI 66.6–68.4%
7,220 of 10,698 completed stores in the 6 Nov 2026 snapshot where UCP create_cart returned a checkout link (ARC stopped before payment; no order was placed).
Snapshot 2026-11-06 (as of 2026-11-06) · Canonical snapshot figure · store_answers.csv · can_agents_get_checkout_link
Published in every snapshot as protocol_matrix.csv. ACP, PAP and agent payments are not measured yet. x402 is not in the matrix yet.
Methodology · Data · Leaderboard
plannedcri-v0 · Census weekly, panel monthly (planned)
ARC Channel Reach Index
Through which agent channels (Google AI Mode, Copilot, Perplexity, Amazon, ChatGPT) can merchants be reached, and does in-chat checkout work?
A public channel signal per store (census), plus a monthly sampled panel of shopping queries per assistant (directional).
Planned. No figure until the first run is frozen and published.
Research done. No census file or panel run is published, so there is no figure.
Methodology · Data: not published · Leaderboard: not yet
betapti-v0.1 · Weekly with the Agent Shopping Index (planned)
ARC Product Truth Index
When an AI agent tells a shopper the price, stock or variants, does it match the store's own page?
Claim: one price, stock, variant or policy statement, scored match, mismatch, partial or couldn't verify.
Not in the open snapshot yet, so no figure.
The fact-check runs on scans today (one model, via API). Results are not in the open snapshot yet, so there is no figure here.
Methodology · Data: not published · Leaderboard