We Re-Ran Our Agent Commerce Study on 147 Brands. Two of Our Numbers Were Wrong.

Checking Our Own Work in Public

In August the PlatformDTC team published a measurement of how many direct-to-consumer brands serve a live agent-commerce discovery document. The sample was 58 domains — small enough that any striking result deserves a second look before anyone builds on it.

So the study was re-run at 147 brands, the collection was turned into a script anyone can execute, and the results were checked against what had already been published.

The headline held. Two of the supporting numbers did not, and correcting them in public is the reason the follow-up exists.

What Held

Seventy of 147 brands serve a live agent-commerce discovery document — 48% of the whole list, and 65% of those that answered a request at all. The finding survived 2.5 times the sample.

The expansion deliberately reached for categories the first list was thin on, and for brands large enough to run their own infrastructure. The rate fell exactly as the platform-default explanation predicts: 74% on the original list, 59% across the expansion, 65% combined.

Every one of the 70 documents reports the same protocol version, and 69 answer a tool-listing request with the same thirteen tools. That uniformity is itself the finding — this is one vendor's rollout, observed from outside, rather than 147 independent integrations.

Correction: Some Brands Do Block Answer Engines

The first study reported that not one brand in the sample disallowed AI answer engines in robots.txt. Across 147 domains, two do, and both shut out the full set — search-facing crawlers, training crawlers, and live agents fetching on behalf of a user. That reads as deliberate policy rather than a copied snippet.

Six brands disallow training crawlers, and the variation between them is the interesting part: some block everything in that class, while others block two named crawlers and leave the rest free. These are choices, not defaults.

The accurate claim is therefore not that nobody blocks. It is that blocking remains rare and highly selective — 2 of 147 for answer engines, 6 of 147 for training. The direction of the earlier finding stands; the absolute has moved off zero, and anyone who quoted the zero should quote this instead.

Correction: Refusal at the Edge Is Far More Common

The first sample had seven of 30 domains refusing or failing to answer. At 147, it is 39 — just over a quarter. The refusals break down as 24 returning 403 Forbidden, 11 not responding at all, and the remainder split across 429, 404 and a non-standard 420.

The limitation that came with the first study applies here with more weight, and is worth restating plainly: the client identifies itself honestly, and no bot vendor recognises the name. A 403 proves a site refuses unfamiliar crawlers. It does not prove the site refuses a crawler that arrives with a name operators know and IP ranges they can verify. Twenty-seven per cent is an upper bound on edge blocking, not a measurement of it.

It is still the number worth knowing, because it is invisible to every audit that reads robots.txt. Nearly all of those 39 domains publish a permissive robots.txt. The file says yes and the edge says no, and only one of the two is ever checked.

Why This Matters to Us

Autonomous commerce is a field with very few measurements and a great many assertions. A number published without a method anyone can re-run is a claim, not a result.

The full study, the corrections, the dataset and the collection script are published on the PlatformDTC blog at platformdtc.com/blog.