State of agent checkout, week of 12 Oct 2026: 650 of 1,331 US stores returned a UCP checkout link

In ARC's 12 October 2026 snapshot, 650 of 1,331 US stores (48.8%) returned a UCP checkout link to an AI agent: 629 of 715 Shopify stores (88.0%) and 21 of 616 other stores (3.4%).

This is the first weekly State of agent checkout from ARC (Agentic Report Card). Each week we take the newest immutable version of the ARC dataset and report three things. How many stores let an AI agent reach checkout over UCP, the Universal Commerce Protocol. How many stores block agents, by category. And, once they exist, what agents report back. Every number has its n, its date and the dataset version, so you can recompute it.

This week is the baseline

There is no week-over-week change in this edition, on purpose. This edition uses dataset version 2026-10-12.r2. That version supersedes 2026-10-12, exported earlier the same day. The later export adds whether a store publishes a UCP file. It does not change the checkout, profile, bot-block or robots.txt counts below, and it does not re-check any store. The previous sample, 2026-10-09.r3, uses the same method (arc-method-2026.10) but is only three days older. None of the 1,261 US stores whose check completed in both 2026-10-09.r3 and 2026-10-12.r2 was re-checked between them. The methodology changelog records a sample expansion on 12 October. The changelog does not give a count. 76 is the number of US stores in version 2026-10-12.r2 that are not in version 2026-10-09.r3. Any difference would come from which stores are in the sample, not from stores changing. From next week we compare only stores that appear and were re-checked in both versions.

UCP agent checkout, US sample

ARC dataset version 2026-10-12.r2. n = 1,331 US stores whose HTTP check completed (11 more could not be checked). Those completed checks ran 7–12 October 2026.

Group	Stores	Valid UCP profile	Checkout link returned
Shopify	715	681 (95.2%)	629 (88.0%)
Not Shopify	616	56 (9.1%)	21 (3.4%)
All US	1,331	737 (55.4%)	650 (48.8%)

"Not Shopify" covers 293 stores on a platform ARC could not detect, 205 whose homepage could not be read, 60 on Salesforce Commerce Cloud, 24 on BigCommerce, 17 on WooCommerce, 7 on Adobe Commerce / Magento, 6 on Squarespace and 4 on Wix. "Valid UCP profile" means the store's /.well-known/ucp profile passed ARC's validation. "Checkout link returned" means a UCP create_cart call returned a checkout link. ARC stops there, before payment.

Declared profiles (/.well-known/ucp answered with a body) are counted for the whole sample but not per store. Across all markets, 1,694 of 3,497 stores (48.4%) declared a UCP profile, 1,677 had a valid one and 1,369 (39.1%) returned a checkout link. See the Protocol Matrix, version 2026-10-12.r2.

The split is the story. On Shopify a checkout link is the default outcome. Off Shopify it is rare: 21 of 616 US stores.

Which store categories block agents

ARC dataset version 2026-10-12.r2, all markets, n = 3,497 completed stores. Every labelled category has at least 20 stores, so none of those rows is suppressed. (The rule is to suppress any category with n < 20.) Two uncategorised stores appear only in the All total, because the category rows sum to 3,495, not 3,497.

Category	Stores (n)	Bot protection blocked ARC's check	robots.txt disallows an AI crawler	robots.txt disallows an AI answer agent
Electronics	304	55 (18.1%)	18 of 257 (7.0%)	12 of 257 (4.7%)
Pets	264	35 (13.3%)	18 of 245 (7.3%)	14 of 245 (5.7%)
Apparel	359	44 (12.3%)	20 of 315 (6.3%)	17 of 315 (5.4%)
Outdoor	289	31 (10.7%)	15 of 262 (5.7%)	8 of 262 (3.1%)
Beauty	306	31 (10.1%)	16 of 276 (5.8%)	15 of 276 (5.4%)
Home	375	33 (8.8%)	19 of 331 (5.7%)	11 of 331 (3.3%)
Footwear	288	25 (8.7%)	8 of 260 (3.1%)	3 of 260 (1.2%)
Food & drink	443	38 (8.6%)	14 of 410 (3.4%)	7 of 410 (1.7%)
Jewelry	298	25 (8.4%)	6 of 278 (2.2%)	1 of 278 (0.4%)
Supplements	287	18 (6.3%)	11 of 268 (4.1%)	9 of 268 (3.4%)
Kids	282	17 (6.0%)	15 of 262 (5.7%)	10 of 262 (3.8%)
All	3,497	352 (10.1%)	160 of 3,166 (5.1%)	107 of 3,166 (3.4%)

How to read it:

• Bot protection blocked ARC's check: the store returned a block response to ARC's HTTP check, so ARC could not see its products. This is an observed block.
• robots.txt columns: a rule in the store's robots.txt, not an observed block. The denominator is the 3,166 stores whose robots.txt ARC could read.
• Human-check: ARC only sees a human-check (a CAPTCHA or "are you human" page) in its browser cart test, which runs on the ARC Index. That covered 83 stores, scanned 5–6 October 2026. Three stopped at a human-check and two blocked automated agents at the door. Apparel has 24 scored stores, and one of those stopped at a human-check. Every other category is under 20 stores, so we don't publish a per-category human-check table.

Electronics stands out. Its block rate (18.1%) is nearly twice the sample's 10.1%.

Agent outcome reports

Coming once reports arrive. Agents can report a checkout outcome for any store. A report stays private until three independent contributors agree within 14 days, and ARC does not yet publish a total. This section will count them once there is something to count.

Why this matters this week: browser agents and blocking

On the 12 October 2026 episode of Cheeky Pint (#35), John Collison asked Mark Zuckerberg about businesses that might try to block agents. Zuckerberg said of Meta's Muse: "we trained into the models very good browser use, right? So that way we want to make it so that no one No one has to do anything to integrate with Muse." He added, "But we have, we have way more businesses that are calling us asking to integrate deeply, and very few really who have tried to block it." (episode and transcript)

ARC's data shows what any browser-based agent meets on the way in. In the 2026-10-12.r2 version, 352 of 3,497 stores (10.1%) blocked ARC's automated check with bot protection. In electronics it was 55 of 304 (18.1%). ARC did not test Muse or any other named agent here. These are ARC's own checks, and a store that blocks one automated visitor may treat another differently. A UCP checkout link is the route that skips the storefront page altogether. In the US it is common on Shopify and rare everywhere else.

Method and data

• Data: ARC dataset version 2026-10-12.r2 (store_answers.csv, rails_checks.csv, stores.csv, protocol_matrix.csv), CC-BY-4.0. The sha256 checksums matched the manifest.
• Method: arc-method-2026.10. See methodology and the changelog.
• Market: a curated label in the dataset, not a measurement. "US" means market = United States.
• Live pages: the adoption page keeps counting as new stores get their first check, so its live numbers will differ from this frozen version.
• Recompute: every number in this post can be regenerated from the version above. Next week's edition follows the same recipe.