CanlahAI/agent-readiness-2026

Agent-Readiness of 50 Cross-Border DTC Brands (2026)

Open dataset · CC BY 4.0 · published by Canlah AI (CANLAH AI PTE. LTD., Singapore)

Canonical citation — cite the DOI: https://doi.org/10.5281/zenodo.22103177

⚠️ v1.0.2 (2026-09-01) withdraws two claims from earlier versions — that three manifests had gone dark, and a ~12% churn rate derived from them. Both were an artifact of this repo's verification script probing the wrong path. The headline finding (25/25 platform-issued) survives and is now better evidenced. Full account: ERRATUM.md. Repository: https://github.com/Canlah-AI/agent-readiness-2026 Mirror: https://huggingface.co/datasets/CanlahAI/agent-readiness-2026 Mirror (Hugging Face): https://huggingface.co/datasets/CanlahAI/agent-readiness-2026 Archive (Software Heritage): swh:1:dir:d14734936dec4add414c537398d7d5d7b133c1ac Sample frame: 50 Chinese-origin cross-border DTC brands with US-facing storefronts (47 reachable). Purposive sample; do not project onto DTC brands generally.

An audit of whether 50 cross-border DTC storefronts expose the files an AI shopping agent looks for: a UCP manifest, an agent-payments endpoint, llms.txt, an MCP endpoint, and product schema.

The finding

Of 47 reachable storefronts, 25 serve a UCP manifest (53.2%).

All 25 of those manifests were issued by Shopify. Not one was authored by the brand.

Agent-readiness, at this point in 2026, is something a storefront platform switches on — not something a brand has decided.

25 / 47	reachable storefronts serving a UCP manifest (53.2%)
25 / 25	of those issued by Shopify, not the brand
0 / 47	exposing an agent-payments endpoint
1 / 47	exposing an MCP endpoint
9 / 47	with Product schema on the sampled product page
6 / 47	naming any AI crawler in robots.txt at all (2 of them block one)

How the issuer column was established

The original collection could only label a manifest "Shopify" when the host hint leaked a *.myshopify.com address — 3 of 25, which under-reports badly, because a Shopify store on a custom domain leaks nothing.

On 2026-09-01 every one of the 25 manifests was re-requested at the same path the collection pipeline used, and the manifest body was read rather than a response header: 25 of 25 are still served, and all 25 cite a *.myshopify.com address inside the manifest itself. That is 25 of 25, now evidenced by live content rather than by a collection-time host hint.

⚠️ This paragraph previously said 3 of them (Anker, Soundcore, eufy) "no longer answer at that path at all," and a related note reported ~12% churn. Both were wrong — an artifact of this repository's verification script probing /.well-known/ucp.json while the collection pipeline probed /.well-known/ucp. Shopify's edge answers both; a self-hosted front end answers only the latter. See ERRATUM.md.

verify_ucp_issuer.sh in this repo re-runs that check. It needs nothing but curl.

./verify_ucp_issuer.sh dataset.csv

Files

File
dataset.csv	50 rows × 22 columns. ucp_issuer_evidence records how each attribution was established.
verify_ucp_issuer.sh	Re-runs the issuer check against the live web.
LICENSE	CC BY 4.0

Method

• Unauthenticated HTTP requests only. No login, no crawling behind a session — nothing a brand could not reproduce against its own storefront.
• One sampled product page per storefront for the PDP schema columns, not a full catalogue scan.
• Percentages use the 47 reachable storefronts as the denominator, not 50. Three did not answer.

Limits — read these before citing

• n = 50, purposively sampled from cross-border DTC brands selling into the US. Not a random sample. Do not project it onto DTC generally.
• A single point in time. We previously reported ~12% manifest churn here; that figure has been withdrawn (see ERRATUM.md). Re-checked 2026-09-01, manifest presence has not churned at all: 25 of 25 are still served. What did change is the manifest version — 22 of 25 moved to 2026-08-25 while 3 remain on 2026-04-08. Boolean presence columns cannot see that, which is a limitation of this dataset's shape, not evidence of stability.
• Presence of a file is not quality of a file. We did not evaluate whether the manifests describe the catalogue well.
• This measures what a storefront exposes. It does not measure whether any AI agent actually used it — nobody outside the engines can measure that today, and we will not claim otherwise.

Citation

Pang, H. (2026). Agent-Readiness of 50 Cross-Border DTC Brands. Dataset.
https://doi.org/10.5281/zenodo.22103178

Attribute to Canlah AI (full name). Found an error? Open an issue or write to admin@canlah.ai — corrections get a dated changelog entry, not a silent edit.

Cite

DOI: https://doi.org/10.5281/zenodo.22103178 (Zenodo, CC BY 4.0)

@misc{canlah_ai_2026_agent_readiness,
  title  = {Agent-Readiness of 50 Cross-Border DTC Brands},
  author = {Pang, Haoyang and {Canlah AI}},
  year   = {2026},
  note   = {Dataset, v1.0.0, CC BY 4.0},
  doi    = {10.5281/zenodo.22103178},
  url    = {https://doi.org/10.5281/zenodo.22103178}
}