Amazon search and product discovery used to be simple: type in a keyword and the product listings appear. Today, that system has now evolved into a layered AI ecosystem powered by COSMO (Amazon’s backend intelligence layer) and Rufus (the customer-facing AI shopping assistant).
Together, Amazon COSMO and Rufus represent Amazon’s shift from keyword search to intent-based shopping intelligence.
The Old System: A9/A10 (Keyword-Based Ranking)
Before AI-driven search, Amazon relied on the A9 algorithm, which was later refined to A10. Here’s how it worked:
- Keyword Precision: Amazon provided product listings with keywords that perfectly mirrored what the customer typed in their search bar.
- Performance Metrics: It prioritized sales velocity and conversion rates. Essentially, if people clicked it and bought it, the A9 algorithm places your listing at the top.
- Literal Interpretation: If you searched for “lightweight running shoes,” the system looked for exactly those words.
The New Layer: COSMO (Amazon’s AI Knowledge Graph Engine)
COSMO (literally Amazon’s ‘common sense’ model) is the intelligence layer built on LLMs that replaces pure keyword matching logic with semantic understanding of products and shopper intent. Here’s how COSMO works:
Creates a massive web of connections between products and real-world situations and human needs.
For example, a search pattern like “portable charger,” “low battery,” and “airport” could
signal a traveler navigating long layovers or delays, where staying connected is urgent. COSMO would interpret this as a travel reliability moment, prioritizing high-capacity, fast-charging power banks suited for mobility.
COSMO isn’t just matching words—it’s interpreting the situation behind the search, and surfacing products that fit the moment, not just the query.
Looks into context over keywords by reading between the lines.
COSMO moves beyond keywords by reading between the lines and closing the ‘query-product gap’—the difference between how customers naturally describe what they need and how products are labeled by manufacturers.
For example, a ‘gaming headset’ product may not explicitly mention work-from-home use, but have been consistently described in reviews as “great for Zoom meetings”. Reviews like this will result in COSMO linking the product to ‘remote work communication needs’, allowing it to surface in searches like ‘headset for online meetings’.
Rufus: The AI Shopping Assistant Layer
Compared to COSMO that works on the backend, Rufus is the front-end conversational AI that customers can interact with inside Amazon. Rufus behaves like a personal shopping assistant that answers shopper questions in natural language, provides instant recommendations based on use cases, and helps discover products without the traditional search. Here’s how it works:
- Natural Language Interaction. Customers can ask Rufus full, context-rich questions like: “what’s the best laptop for video editing under $1000?” “What to pack for a beach trip with three young kids?”
- Deep Intent Interpretation. Rufus breaks down vague or broad queries into structured needs. For example, a customer may search ‘running shoes for beginners’, Rufus will recommend shoes that offer comfort, injury prevention, and affordability, rather than technical running shoes.
- Context Expansion and Proactive Recommendations. Rufus doesn’t just respond to queries, but also anticipates adjacent needs. For example, a customer looking for a camping tent may also receive recommendations from Rufus for items such as sleeping bags, portable lights, insect repellent, and more.
How to Rank for COSMO and Rufus
Old A9 listings focused heavily on features and specifications. COSMO and Rufus, however, interpret listings more like a human assistant—matching products to real-life situations and customer intent.
Structure bullet points around ‘intent and use cases’.
Bullet points used to only list technical features. This time, switch to a ‘Benefit > Feature > Use case’ model.
- A9-Style bullet: ‘Made of 100% cotton”
- COSMO/Rufus-style bullet: Made of 100% breathable cotton to prevent night sweats, ideal for hot sleepers during warm summer nights. (This answers questions like ‘bed sheets for hot sleepers’ or ‘how to stop sweating while sleeping’).
Rewrite your title for intent clarity.
Your title should go beyond keywords and clearly signal who the product is for, what problem it solves, and what outcome it delivers.
- A9-Style Title: Cooling Bed Sheets Queen Size 100% Cotton Breathable Soft Lightweight Bedding Set
- COSMO / Rufus-style title: Cooling Queen Size Bed Sheets for Hot Sleepers — 100% Breathable Cotton Designed to Reduce Night Sweats and Improve Sleep Comfort
Fully utilize Backend Search Terms and Attributes
COSMO and Rufus don’t just read your visible listing, they also build a structured understanding of your product through hidden data. This can be done by filling out all optional backend attributes, including variations of intent and not just synonyms, and covering alternative ways customers can describe the same problem.
- Weak backend keywords: cotton sheets, soft sheets, queen bedding, bed set
- COSMO/ Rufus-optimized backend keywords:
- “Cooling sheets for hot sleepers”
- “Bedding for night sweats”
- “Summer bed sheets breathable fabric”
- “Soft sheets for sensitive skin”
Proactively build your Q&A section
Your Q&A section no longer just acts as customer support, but also as training data for COSMO and Rufus on how your products can be interpreted and recommended. Your Q&A section should anticipate real buyer questions and answer them in a natural, helpful tone.
Example:
Q: Is this bedsheet good for hot sleepers?
A: Yes, the breathable cotton fabric is designed to help regulate body temperature and reduce heat buildup, making it ideal for people who tend to overheat at night.
The shift from A9/A10 to COSMO and Rufus isn’t just an algorithm update—it’s a fundamental change in how Amazon understands products and its users.
In this new landscape, your brand success depends on how well your product can be understood, not just how well it’s described. Brands that structure their listings around intent, lifestyle scenarios, and conversational language will be the ones that consistently surface in recommendations.
Are you looking to stay ahead of Amazon’s shift to COSMO and Rufus? Our team at Omni Channel Solutions can help fully optimize your brand’s Amazon presence end-to-end.
From listing optimization built around customer intent, to backend structuring, A+ content, SEO strategy, and ongoing catalog management, we make sure your products are not just listed, but understood and recommended by Amazon’s AI systems.
For more information, reach us at info@omnichannelsol.com.