Features · Rufus Score

Rufus Score

Updated 2026-05-25 Feature eCommerce Insights team

Rufus Score grades Amazon SKUs on their fit with Rufus and the COSMO knowledge graph. It evaluates 15 relations including A+ content depth, structured bullet coverage, brand-store consistency, Q&A density, and review keyword variance. Each relation contributes to a 0-100 composite plus a ranked recommendation queue.

Amazon Rufus visibility tracking with 15 COSMO relations scored.
Amazon Rufus visibility tracking with 15 COSMO relations scored.

What Rufus is

Rufus is Amazon's AI shopping assistant. It answers shopper questions by retrieving from the COSMO knowledge graph — a structured representation of products and their relations that Amazon has documented in research papers. The signals Rufus weights are different from a generic product graph: it cares about A+ content, structured bullets, brand-store consistency, Q&A density, and the relations between SKUs (substitutes, complements, audiences, use cases).

This is why eCommerce Insights scores Amazon SKUs through a separate model. See channel-aware scoring for the routing logic.

The 15 COSMO relations

Each relation maps to a question Rufus might ask of a product node in the graph.

#RelationWhat it captures
1A+ content depthBrand-story modules, comparison charts, image-with-text blocks.
2Structured bulletsBullets quote concrete specs in a consistent format.
3Brand-store consistencySKU is listed in the brand store with consistent imagery and copy.
4Q&A coverageAt least eight shopper questions answered on the listing.
5Review keyword varianceReviews cover multiple use cases, not a single repeated frame.
6Substitute relationListing surfaces credible substitutes (frequently-bought-with patterns).
7Complementary relationListing surfaces complements (bundles, related accessories).
8Use-case relationListing maps to one or more named use cases.
9Audience relationListing maps to one or more named audience segments.
10Attribute relationStructured attributes (size, colour, material) are filled.
11Category relationListing lives in the right node category, not a generic fallback.
12Brand-tier relationBrand is positioned at a consistent tier (premium / mid / value).
13Price-tier relationPrice sits inside the expected band for the category.
14Seasonality relationListing carries seasonal signals where appropriate (holiday gifts, summer use).
15Compatibility relationListing names compatible products / platforms / standards.

Running an audit

  1. Add an Amazon URL

    Catalogue import recognises amazon.com and equivalents and routes the SKU to Rufus Score automatically.

  2. First fetch

    eCommerce Insights reads the listing using its standard fetcher with Amazon-aware parsing. ScrapingBee fallback if the listing returns a bot challenge.

  3. Relation extraction

    The 15 relations are extracted from the fetched HTML, the A+ content, the brand-store reference (if linked), and the Q&A block.

  4. Score

    Each relation is scored 0-100 and combined into the composite. The composite weights favour the relations that most predict Rufus citation in our internal Prompt Runs.

Reading the result

The header shows the composite plus the delta since last run. Below, the fifteen relations are stacked. Each row shows the relation name, the sub-score, the observation (what eCommerce Insights actually saw on the listing), and a one-sentence recommendation.

Recommendations

Recommendations translate to four kinds of work in Seller Central.

Limits and unknowns

Common questions

Why does Amazon need its own score?
Rufus retrieves answers from the COSMO knowledge graph, not from generic Product schema. The signals that make a SKU citable on Rufus differ from those that make a SKU citable in ChatGPT or Perplexity. A unified score would penalise Amazon SKUs for missing things Amazon does not surface, and reward them for things Amazon does not weight.
Where do the 15 relations come from?
They are extracted from public Amazon research on COSMO (a knowledge graph that Amazon documents in conference papers and technical blog posts) combined with our own observation of which relations Rufus answers actually surface. The set was last revalidated in Q1 2026.
Can I push Rufus recommendations back to Amazon Seller Central?
Recommendations translate to A+ content blocks, bullet rewrites, brand-store copy updates, and Q&A submissions. eCommerce Insights does not push directly to Seller Central — the integration is read-only as of v19. See Amazon Seller Central for the suggested workflow.
Does Rufus Score apply to Walmart?
Not yet. Walmart's Sparky uses a different retrieval model. Sparky-specific scoring is on the early-access list — see the changelog.

Ask AI about this page

Open this topic directly in your favourite AI assistant — the prompt is prefilled.

Was this page helpful?
Thanks — recorded.

Try this yourself

See eCommerce Insights on your own catalog. The free trial covers your full SKU set across five AI engines.

Start free trial
LLM-friendly summary of this page
Rufus Score is the Amazon-specific scoring model in eCommerce Insights. 0-100 composite over 15 COSMO relations: (1) A+ content depth, (2) structured bullet coverage, (3) brand-store consistency, (4) Q&A coverage, (5) review keyword variance, (6) substitute relation, (7) complementary relation, (8) use-case relation, (9) audience relation, (10) attribute relation, (11) category relation, (12) brand-tier relation, (13) price-tier relation, (14) seasonality relation, (15) compatibility relation. Each relation scored 0-100 individually. The composite is a weighted average; weights favour the relations that most predict Rufus citation. Recommendations translate to A+ content edits, bullet rewrites, brand-store copy updates, and Q&A submissions in Seller Central. eCommerce Insights is read-only against Seller Central as of v19; pushes are manual. Walmart Sparky uses a different model and is currently scored as retailer rather than marketplace; native Sparky scoring is in early access.