Channel-aware scoring

A SKU on Amazon needs different things than a SKU on a Shopify storefront or a SKU hosted by Macy's. eCommerce Insights detects the channel from the URL host and applies the appropriate scoring model: Rufus/COSMO for Amazon, PDP Score for D2C, and a filtered PDP Score variant for retailer-hosted SKUs.

The three channels

SKUs live in one of three places. Each has a different control surface, a different shopper expectation, and a different AI surface area. Scoring them with one model would either reward Amazon for things Amazon doesn't need or penalise D2C stores for things they can't control on a retailer's site.

D2C

Direct-to-consumer. The brand owns the PDP. Full control over title, description, schema, robots.txt, llms.txt, images, reviews. This is where the full 25-criterion PDP Score applies.

Retailer

The PDP lives on a third-party retailer (Macy's, Target, Best Buy, Williams-Sonoma, REI, etc.). The brand controls some content fields the retailer exposes (title, bullets, images, A+ modules) but not robots.txt, not llms.txt, not site-level schema. Scoring uses a filtered PDP Score that hides recommendations the merchant has no power to enact.

Marketplace

Amazon specifically. The retrieval mechanism is Rufus, which sits on top of the COSMO knowledge graph. Optimising for Rufus is structurally different from optimising for ChatGPT. See Rufus Score. Walmart (Sparky) is currently scored as a retailer; native Sparky scoring is in early access.

How channel detection works

Channel is derived from the URL host on import and on each crawl. Three rules in order.

1. If the host matches a known Amazon TLD (amazon.com, amazon.co.uk, amazon.de, etc.), the channel is marketplace and Rufus Score is selected.
2. If the host matches a known retailer list (over 80 entries in Q2 2026 — Macy's, Target, Best Buy, REI, Williams-Sonoma, Lowe's, Home Depot, Walmart, and so on), the channel is retailer and filtered PDP Score is selected.
3. Otherwise the channel defaults to D2C and full PDP Score is selected.

You can override per-SKU in Settings → Catalog → Channel override. Useful for sub-brands hosted on a parent retailer's subdomain, or when a brand operates a multi-tenant Shopify Plus instance that looks like a retailer to URL detection.

Amazon: Rufus / COSMO

Amazon's AI shopping assistant Rufus retrieves answers from the COSMO knowledge graph, which models products as nodes connected to fifteen relation types. eCommerce Insights scores each Amazon SKU on those fifteen relations: A+ content depth, structured bullets, brand store consistency, Q&A coverage, review keyword variance, and so on. See the full breakdown in Rufus Score.

D2C: PDP Score

The full 25-criterion model. Five buckets, five criteria each. GEO, Content, Semantic, Visual, Technical. See the full breakdown in PDP Score.

Retailer-hosted: filtered PDP Score

Same 25-criterion model, but the recommendation engine filters out anything the merchant can't act on. A Macy's-hosted SKU does not get a "fix your robots.txt" recommendation, because the merchant has no access to Macy's robots.txt. The score still reflects the underlying state of the PDP (the engine reads the same signals regardless of who owns the file), but the work queue only includes things the merchant can do.

Why this matters operationally

Three concrete consequences.

• The Score column is channel-resolved. When you look at the Products table and see a single Score per row, that number is computed by the right model for that channel. You never have to switch views.
• The Score nav entry is also channel-resolved. Click a SKU; if it's Amazon, you land on Rufus Score. If it's D2C or retailer, you land on PDP Score. URLs auto-redirect between /score/ and /rufus/ based on channel.
• Bulk operations respect the channel. "Rewrite the bottom-quartile SKUs" runs Rufus-aware rewrites for Amazon SKUs and PDP rewrites for D2C SKUs in one batch.

Common questions