COSMO — Amazon's commonsense engine

What COSMO actually does

Described in Amazon Science's published research ("COSMO: a large-scale e-commerce common sense knowledge generation and serving system"), the framework extracts commonsense relations between queries and products — functional ("used for"), audience ("used by"), occasion, capability, and similar intent relations — and serves that knowledge to ranking systems. eCommerce Insights' channel scoring works against 15 of these commonsense intent relations for Amazon SKUs.

The practical shift: classical Amazon search matched keywords in titles and backend fields; COSMO-era ranking infers why the shopper is asking and rewards listings whose content supports the inferred intent. A listing can match every keyword and still lose to one that matches the intent.

Why it matters for ecommerce

For brands selling on Amazon, COSMO changes what listing optimization means. Stuffing backend keywords addresses the old matcher; the new systems reward listings whose titles, bullets, and A+ Content state use cases, audiences, and occasions explicitly — the relations COSMO mines. Amazon Science's published evaluation reported meaningful improvements in search relevance metrics from intent-aware ranking, which is Amazon's own signal about where listing quality is headed.

It also explains why Rufus answers differ from old search results: the assistant draws on intent relations, not keyword indexes, so a listing that never says who or what the product is for gives Rufus little to work with.

Optimizing against intent relations: an example

A kitchen-brand's chef knife listing ranks for "8 inch chef knife" but never surfaces for "knife for new home cook" (illustrative example). The audit against intent relations finds the gap: nothing in the listing supports the "used by beginners" relation — no mention of forgiving edge geometry, weight balance for learners, or care simplicity. The rewrite adds an explicit audience bullet and an A+ module on first-knife guidance. The listing starts competing on intent queries it previously could not see, because now the content carries the relations the ranker infers.

How it relates to neighboring terms

COSMO is Amazon-channel infrastructure; Rufus is the shopper-facing assistant above it; the general-engine equivalents of this work are GEO and AEO, which target ChatGPT, Perplexity, and Google surfaces instead. The shared lesson across channels: structured, explicit intent coverage beats keyword density everywhere AI does the ranking.

How eCommerce Insights uses it

Channel-aware scoring: Amazon SKUs route to Rufus/COSMO scoring across 15 commonsense intent relations, with rewrite recommendations for titles, bullets, A+ Content, and backend keywords phrased for Seller Central. Shopify and other D2C SKUs route to the six-engine scan instead — same platform, vendor-specific output. See the Amazon solution page and the Rufus score docs.

Related terms

• Rufus (Amazon) — the shopper-facing assistant above COSMO.
• GEO (Generative Engine Optimization) — the general-engine counterpart discipline.
• AEO (Answer Engine Optimization) — answer-surface optimization off Amazon.
• Product schema — the structured-intent layer for D2C channels.
• PDP optimization — the cross-channel practice COSMO reshapes on Amazon.