One loop: measure, approve, push, measure again.
Feedmind reads your catalog the way an assistant reads it, tells you what is missing in the words of the check that failed, drafts or derives the change, and waits for you to approve it. Nothing reaches your store on its own. After a push the affected products are rescored, so the score is also the receipt.
The loop
The score and the four dimensions
Every registered product page is checked for agent accessibility, structured data, content answerability and multi-language parity. You get one number per product and one for the catalog.
Rules first, drafts where rules cannot reach
Deterministic work such as schema, identifiers and markup is derived from data you already have. Where a page needs prose, a model drafts it from that product own attributes.
Your approval is the only write trigger
Changes are shown as before and after on the real product. Approve one, or a batch. Feedmind writes to Shopify only after that, and rescores what it touched.
One catalog, one number, and the products that cost you the most.
The overview is the first thing you see after a scan. The score and the four dimensions sit above the grade distribution, and under it the products ranked by traffic against severity, so the queue starts with what earns rather than what scores worst.
Six things the product does.
The Feedmind Score
One number per product and per catalog, with the four dimensions and every failing check named. Free to see, on any plan or none.
Fix pipeline
A queue of proposed changes with before and after, batch approval, a push to Shopify and an automatic rescore of what changed.
Content rules
Your own conditions and transforms applied across the catalog. Deterministic, previewable against real products, and free to run.
AI drafts
Drafted descriptions, titles and FAQ answers grounded in the product attributes. Credit metered, estimated before generation, never published on their own.
Buyer questions
The purchase questions an assistant cannot answer from your page today, listed per product, with the answer written once and reused.
Multi-language
Parity between locales measured field by field, so a German page is not a thinner translation of the English one.
Scoring and rescoring · schema and JSON-LD · GTIN and identifier mapping · llms.txt and FAQ markup · every content rule you write · the fix queue, approvals and pushes · score history
Only work a language model drafts: product descriptions, restructured titles and written answers to buyer questions. One credit is one product, run once, across all of its fields and every locale you sell in. You see an estimate before generation, and you are charged for drafts produced, not for drafts you reject beforehand.
Impact figures shown in the app next to a proposed fix are labelled EST. They are an estimate of the score movement a change should produce, based on the checks it satisfies. They are not a prediction of traffic, ranking or revenue, and we do not present them as one.
See this on your own catalog.
The free check reads a representative sample of your product pages and returns the score in about a minute. No account, no card.