How the score is calculated

The NOREN AI readiness score is NOREN AI's own measure of how much of a store's product information AI shopping services can read from its product pages. It is not a rating by any AI service. Current method: 2026-10-09.

What a scan reads

  • The store's robots.txt, its homepage, its sitemap files and up to 20 product pages. The product pages are a fixed sample: chosen from the sitemap by a hash of the store's domain, so a rescan reads the same pages. A store with no product sitemap is sampled by following links from its homepage and category pages.
  • The store's return and shipping policy pages, only to find whether they exist. What they say is never scored or shown.
  • Every request identifies itself as our crawler, Mozilla/5.0 (compatible; AICommerceReadinessBot/0.1; +https://noren-ai.vercel.app/). robots.txt is honoured, including its Crawl-delay; otherwise at most one request per second. A refusal is reported, never retried under another identity.
  • When the store asks for a long Crawl-delay, the scan stops within its time limit and the report says it is partial and why.

What counts as product data

A schema.org Product in the HTML the store sends: JSON-LD or microdata. Open Graph tags are made for social-media previews and do not count today. A price of 0 counts only when the page says the product is free.

The score

For each channel open in the store's market, the score takes the share of sampled products whose product data carries every field that channel requires. The score is the plain average of those shares, so each open channel weighs the same. When the store's market cannot be identified, every channel below is counted, and the report says so.

ChannelFields required per productDocumentation read
UCP 2026-08-25 catalogPrice, Product name, Description2026-09-29
Google Merchant Center product dataPrice, Availability, Product name, Image, Description2026-09-29
OpenAI product feed (discovery)Price, Availability, Product name, Image, Description, Brand2026-09-29

Two more figures sit beside the score:

  • After NOREN AI's fixes: a missing field counts as met when the store already states its value somewhere NOREN AI reads (the page's Open Graph tags, the page itself, or the store's catalog), and the value is found there. Nothing is invented.
  • With the owner's confirmations: fields the store's owner can confirm in one step, such as a brand the sampled products already state. Shown only when one applies.

Findings and fixes

The report lists findings the scan reads directly from public pages: access refused, no product data, invalid product data, a required field missing, a policy page not found, products not discoverable. Each finding shows what the scanner looked for and on which pages. For product findings, the corrected JSON-LD for the example product is built only from values the scan read on that page or in the store's catalog. A field with no such value is left out and listed as still needed.

What the score does not mean

  • It is not a rating, ranking or score from Google, OpenAI or any other AI service, and it does not predict how any of them will show the store.
  • It says nothing about search position, visits, sales or revenue, and it is not a comparison with other stores.
  • Prices and stock are not checked for correctness. Price and stock disagreements are never shown on the public report.
  • It covers the sampled pages only, not the whole catalog, on the date of the scan.
  • It does not judge a store's policies, products or service: only whether machine-readable product data is present.

Method versions

Every scan stores the method it was scored under, with its scan time. A store owner's history marks scans scored under an earlier method, because their scores may not compare with today's.

  • 2026-10-09 (current): A sampled product page that answers with a "not available" notice instead of a product (a product the store lists but does not publish) is left out of the sample and counted on its own, so the score covers only real product pages. Before, such a page counted as a product without product data.
  • 2026-10-08: The first versioned method. Scores count structured product data only (JSON-LD or microdata), averaged over the channels open in the store's market, on a seeded sample of up to 20 product pages. Scans from before 8 October 2026 carry no version and may have been scored under earlier rules.

Questions about a report of your store: manage or remove a report.