Why do AI shopping agents fail at checkout?
In the Measured Index, stops follow the dataset order. The Measured stops, most common first, are Couldn't add it to the cart (32), then Blocked at the size picker (8), then Stopped by a human-check (3), then Store blocks automated agents at the door (2). Stores where 0 of 5 shoppers added to cart can still have returned a checkout link. That figure covers Measured scores only, snapshot 2026-10-12.
Short answer
The Measured stops, most common first, are Couldn't add it to the cart (32), then Blocked at the size picker (8), then Stopped by a human-check (3), then Store blocks automated agents at the door (2). A store with no stop recorded is not in the failure count.
The failure ranking
ARC open dataset snapshot 2026-10-12. Index browser tests stop at add-to-cart. Nothing is bought.
Measured scored stores: 77MeasuredA published score whose dataset row is evidence=measured. Rates below use only these stores.
Inferred scores: 6InferredKept out of the rates. An inferred score is not treated as a failure or a success.
No published score: 6Not measuredRetest pending or not finished. A missing score is not a zero, and it is not a mark against the store.
5 of 5 shoppers added to cart: 32 of 77Measured
0 of 5 shoppers added to cart: 18 of 77Measured
0 of 5, but a checkout link still came back: 13 of 18MeasuredMeasured scores only, snapshot 2026-10-12. Inferred scores and stores with no published score are outside this count. The protocol check said yes. The browser test did not get the product into the cart.
Median score: 83Measured
Highest score in this snapshot: 86 of 100MeasuredIndex tests stop at add-to-cart, so a clean cart pass does not score 100. The formula is on the methodology page.
Published stop reason where at least one shopper missed add-to-cart (45 of 77 Measured stores):
| Stop reason | Stores |
|---|---|
| Couldn't add it to the cart | 32 |
| Blocked at the size picker | 8 |
| Stopped by a human-check | 3 |
| Store blocks automated agents at the door | 2 |
| Shoppers that added to cart | Measured stores |
|---|---|
| 0 of 5 | 18 |
| 1 of 5 | 6 |
| 2 of 5 | 1 |
| 3 of 5 | 2 |
| 4 of 5 | 18 |
| 5 of 5 | 32 |
Protocol yes, storefront no
13 of 18 still returned an agent checkout link. That is the only checkout-link figure in this section. It covers Measured scores only, snapshot 2026-10-12: stores where 0 of 5 shoppers added to cart. Inferred scores and stores with no published score are not in it. A dated report can use a different sample, so its count can differ.
Fixes per failure type
Add-to-cart: the control should change the cart in the page. Picker: in-stock sizes and colours should be real options, with labels. Human check and door block: decide on purpose whether agents are allowed. None of these is a judgement of the brand.
Retest
Run /agent-checkout again after the change. The October write-up, with its own sample, is State of agent buyability, October 2026. This page uses the current dataset, which can differ from that report.
Frequently asked questions
- Why do AI shopping agents fail at checkout?
- In the Measured Index, stops follow the dataset order. The Measured stops, most common first, are Couldn't add it to the cart (32), then Blocked at the size picker (8), then Stopped by a human-check (3), then Store blocks automated agents at the door (2). Stores where 0 of 5 shoppers added to cart can still have returned a checkout link. That figure covers Measured scores only, snapshot 2026-10-12.
- What's the most common failure?
- The add-to-cart step, in ARC's Index. This page shows the current counts.
- Can a store pass the protocol check but fail in the browser?
- Yes. This page counts how many Measured Index stores did that in the current dataset.
On ARC
Cite this: ARC, arcreport.ai, accessed 2026-10-13, CC-BY-4.0. Protocol figures: ARC saved checks, accessed 2026-10-13 (13 October 2026). Index browser figures: ARC open dataset snapshot 2026-10-12. Platform figures: ARC saved checks, accessed 2026-10-13.
Every figure is labelled Measured, Inferred or Not measured. Not measured is not zero. A missing result is not a mark against a store. ARC stops before payment.