What product data do AI shopping agents need?
Agents need clear, consistent product data: exact title, price, currency, stock, variants such as size and colour, images, shipping and return terms, and identifiers like GTINs. It should match on your page, your feed and your structured data. ARC scores product data quality per store and shows where an AI read-back disagrees with the page.
Short answer
Agents need the same facts a shopper needs, written so a machine can read them: title, price, currency, stock, variants, images, shipping, returns, and an identifier such as a GTIN. The page, the feed and the structured data should agree.
The fields agents rely on
schema.org Product and Offer, a merchant-center product spec, and a catalog feed all ask for overlapping fields. ARC's product-data check reads structured data on the page. Presence is not proof the text is accurate.
Keep page, feed and structured data in sync
A price on the page that disagrees with the feed is how a read-back goes wrong. The mismatch rate, when ARC has stored fact checks, is on what agents get wrong.
Variants: the common trap
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 |
Size and colour pickers are a Measured stop in the Index when the failure reason says so. The count is in the questions below and in the stop list above. Inferred scores are not in that count.
What ARC's product data score shows
Price and stock from structured data, in the open dataset. Not tested is excluded from the yes rate.
| Platform | Yes | No | Not tested | Yes among tested |
|---|---|---|---|---|
| Shopify | 1,342 | 185 | 99 | 1,342 of 1,527 (87.9%) |
| Not detected | 186 | 106 | 622 | 186 of 292 (63.7%) |
| Salesforce Commerce Cloud | 61 | 26 | 65 | 61 of 87 (70.1%) |
| WooCommerce | 20 | 16 | 25 | 20 of 36 (55.6%) |
| Adobe Commerce / Magento | 12 | 11 | 35 | 12 of 23 (52.2%) |
| Platform not measured | 7 | 6 | 626 | 7 of 13 (53.8%) |
| BigCommerce | 3 | 7 | 40 | 3 of 10 (30%) |
| PrestaShop | 4 | 3 | 6 | 4 of 7 (57.1%) |
| Platform | Measured | Average | Strong |
|---|---|---|---|
| Shopify | 1,687 | 69 | 604 of 1,687 (35.8%) |
| Not detected | 330 | 56.2 | 109 of 330 (33%) |
| Salesforce Commerce Cloud | 91 | 58.8 | 29 of 91 (31.9%) |
| WooCommerce | 47 | 46.6 | 5 of 47 (10.6%) |
| Adobe Commerce / Magento | 28 | 45 | 4 of 28 (14.3%) |
| Squarespace | 6 | 58.3 | 0 of 6 (0%) |
| BigCommerce | 10 | 46 | 1 of 10 (10%) |
| Wix | 14 | 60 | 0 of 14 (0%) |
| PrestaShop | 12 | 53.8 | 1 of 12 (8.3%) |
| Lightspeed | 2 | 67.5 | 0 of 2 (0%) |
Fix list
One product URL with a price and a currency. Stock that matches the page. Variants that can be selected. Shipping and returns in structured data or a policy page. Then retest. Typed facts for one store are on /agent-facts.
Frequently asked questions
- What product data do AI shopping agents need?
- Agents need clear, consistent product data: exact title, price, currency, stock, variants such as size and colour, images, shipping and return terms, and identifiers like GTINs. It should match on your page, your feed and your structured data. ARC scores product data quality per store and shows where an AI read-back disagrees with the page.
- Do I need schema.org markup?
- It helps agents that read pages. Feeds matter for agents that use catalogs. Keep both consistent.
- Why do variants matter?
- Agents often fail to choose a size or colour. In the current Measured Index, ARC recorded "Blocked at the size picker" at 8 stores. Inferred scores are not in that count.
- Can I see what ARC extracted?
- Yes, on the free agent facts page for your store.
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.