Can I make AI assistants change what they say?

Not directly. Fix your pages and feeds; assistants read those.

Does claiming my ARC store card change my score?

No. Claims never change measured scores, ranks or dataset rows.

Whose view is the read-back?

One AI model's view, checked against your pages. It isn't a consensus of all assistants. ARC does not name that model on this page.

ARC · Answers

What do AI agents get wrong about my products, and how do I fix it?

Agents most often get prices, stock, variants and policies wrong when a store's pages, feeds and structured data disagree or can't be read. ARC compares one AI model's read-back with your actual store pages and flags mismatches on your free store page. Fix the source data first, then retest.

Short answer

The usual disagreements are price, stock, variants and policies, when the page, the feed and the structured data do not match or cannot be read. Fix the source, then retest. ARC does not name the model behind the read-back.

Common mismatch types

A quoted mismatch is a genuine disagreement with the store's own page. A price range that contains the store's price, a few cents of rounding, or a different currency is not counted as a mismatch on the public hub.

See your store's mismatches on ARC

The store page lists quoted mismatches for that host. The public list is /what-agents-get-wrong. Typed facts are on /agent-facts.

Mismatch rate across stores

ARC fact-check hub, stored read-backs on Index stores. The open dataset CSV does not include these rows.

Index stores with a stored fact check: 89MeasuredA store with no stored fact check is not in this denominator, and it is not counted as a mismatch.

Mismatch among facts that could be judged: 50 of 179 (27.9%)MeasuredMatch and mismatch only. Facts the model didn't know, and facts that could not be verified, stay out of this rate. They are not mismatches.

Stores with at least one quoted mismatch: 45 of 89 (50.6%)MeasuredQuoted means the store's own page was stored next to the disagreement.

Facts that matched the page: 129Measured

Facts the model didn't know: 167Measured

Facts that could not be verified: 99Measured

Fix the source, not the AI

You cannot edit what an assistant says. You can make the page, the feed and the structured data say the same thing, then run the check again. The fields to start with are on what product data agents need.

Claim your store card to add a labelled statement

Claiming is free, at /store/{host}/claim. A claim can add a labelled statement and a contact. It does not change measured scores, ranks or dataset rows.

Frequently asked questions

What do AI agents get wrong about my products, and how do I fix it?: Agents most often get prices, stock, variants and policies wrong when a store's pages, feeds and structured data disagree or can't be read. ARC compares one AI model's read-back with your actual store pages and flags mismatches on your free store page. Fix the source data first, then retest.
Can I make AI assistants change what they say?: Not directly. Fix your pages and feeds; assistants read those.
Does claiming my ARC store card change my score?: No. Claims never change measured scores, ranks or dataset rows.
Whose view is the read-back?: One AI model's view, checked against your pages. It isn't a consensus of all assistants. ARC does not name that model on this page.

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.