Agentic readiness of 1,580 online stores: a scan of UCP and ACP support.

What the sample includes

The 1,580 stores are the domains someone chose to run through the free scan at /free-tools/agentic-readiness: prospects, demo targets, competitors, and stores their own teams checked. The public leaderboard keeps one row per domain, the latest scan only, and that is what we read: 1,580 distinct domains, 1,580 of them with a score.

Scans ran from 14 May to 9 October 2026. 1,547 of the 1,580 (97.9%) were run in May 2026 and 33 since June. Protocol adoption moves quickly, so treat the headline figures as a May 2026 picture with a small later top-up, not an October one.

The scanner labels each store with the platform it can fingerprint. Among the 1,580 stores we scanned, 14 May to 9 October 2026, 1,188 were labelled Custom / Headless, 363 Shopify, and 29 everything else combined (WooCommerce, BigCommerce, Magento). Shopify is over-represented relative to its share of stores that exist, and Magento badly under-represented, because the sample is whoever got scanned.

What the scan checks

The scanner runs up to 147 checks across nine dimensions, grouped into three rollups: UCP Compliance, ACP Readiness, Infrastructure. The number of checks a store actually receives depends on what it exposes. A store with a UCP manifest gets the full battery; a store without one is not tested for the manifest's contents and transport, so fewer UCP checks run. The two stores below are the top and bottom ranked rows on 10 October 2026.

The leaderboard row for each store carries the score, grade, platform, a UCP flag, an ACP flag and the scan date. It does not carry per-check results, so this page cannot say which individual checks fail most often. That needs the backend audit records, and we have not run that query.

Readiness is bimodal, not a gradient

Among the 1,580 stores we scanned, 14 May to 9 October 2026, 78.9% grade Very Poor and 19.7% grade Excellent. The three grades in between hold 23 stores together, 1.5% of the sample. There is almost no middle: a store either exposes the protocol surface agents look for, or it does not.

Protocol support

UCP is the Universal Commerce Protocol; a store supports it by publishing a manifest an agent can read. ACP is the Agentic Commerce Protocol; the scanner records whether it can detect an ACP checkout endpoint. Among the 1,580 stores we scanned, 14 May to 9 October 2026, 326 (20.6%) publish a UCP manifest, 323 (20.4%) expose ACP, 323 (20.4%) do both, and 1,254 (79.4%) do neither. Every store with ACP in this sample also has a UCP manifest.

Score by platform

Only two platform groups are large enough to compare. Among the 1,580 stores we scanned, 14 May to 9 October 2026, the 1,188 Custom / Headless stores average 11.6 (median 13.3) with 0.3% carrying a UCP manifest; the 363 Shopify stores average 84.1 (median 94.4) with 87.3% carrying one. WooCommerce, BigCommerce and Magento are shown with their counts and should not be read as platform verdicts; below 30 stores we report the number with the n attached, or not at all.

The platform is not the lever. Within Shopify alone, among the 1,580 stores we scanned, 14 May to 9 October 2026, the 317 stores with a UCP manifest average 93.4 (median 95.0) and the 46 without average 19.8 (median 21.4). Same platform, a 73.6-point gap. A Shopify store without a manifest scores about where a Custom / Headless build does. Whether the manifest is on decides most of the score; see how we enable agentic checkout for what turning it on involves.

When the stores were scanned

The leaderboard keeps the latest scan per domain. Grouping the 1,580 rows by scan month shows how concentrated the sample is in May 2026.

What this data cannot show

It cannot describe the web. These are stores someone chose to scan, not a random sample. Every figure here supports “among the stores we scanned” and never “of all stores”. Shopify is over-represented and Magento under-represented relative to the stores that exist.

It may understate Custom / Headless stores. Custom / Headless is the label the scanner assigns when it cannot identify a platform. A site it cannot fingerprint may also be one it cannot fetch cleanly, which would depress the score for reasons unrelated to readiness. The spread argues against a pure artefact (among the 1,580 stores we scanned, 14 May to 9 October 2026, the 1,188 Custom / Headless scores run from 0.8 at the minimum through 13.3 at the median to 76.3 at the maximum, and 4 carry a UCP manifest), but we have not hand-checked the group for fetch failures.

It is mostly a May 2026 picture. 1,547 of 1,580 scans date from May 2026. A May UCP share is not an October one. Re-scanning the sample would make the numbers current; until then they describe adoption in May with a small later top-up.

It has no revenue or conversion data. The scanner measures what a storefront exposes to agents. It does not observe whether an agent visited, recommended, or bought from the store, so nothing here shows that readiness causes sales. For what agents actually do with that surface, read how AI shopping agents work; for who AI engines cite, see our study of AI citations.

It has no per-check detail. The leaderboard exposes a score, a grade, a platform label and two protocol flags per store. Which of the 147 checks fail most often is not in this data. Our methodology page explains how we separate observed coverage from business outcomes.