If you have been selling on Amazon for a while, you have probably noticed that what worked two or three years ago does not carry the same weight today. Keyword-packed titles, bullet points stuffed with search terms, back-end fields crammed to the character limit, the old playbook is losing ground. The reason has a name: COSMO.

Amazon’s COSMO system represents the biggest shift in how the platform evaluates and ranks products since the A9 algorithm first launched. It is not a tweak to the existing search logic — it is a fundamentally different way of reading what a shopper wants and deciding which products deserve to be in front of them.

This guide breaks down what COSMO actually is, how it works, what it means for your listings, and what you can do right now to stay visible in an increasingly AI-driven marketplace.

What Is the Amazon COSMO Algorithm?

COSMO stands for Common Sense Knowledge Generation and Serving System. It is an AI-powered knowledge graph built by Amazon’s applied science team and formally published in a peer-reviewed paper at SIGMOD-Companion 2024, the flagship conference on data management research. The paper was co-authored by ten researchers from Amazon’s Palo Alto lab.

At its core, COSMO is designed to bridge the gap between how traditional search algorithms understand products: through keywords, attributes, and sales data, and how real people think when they shop. A conventional knowledge graph catalogs facts: brand, colour, dimensions, material. COSMO catalogs context: who buys it, when they buy it, what problem they are solving, and what else they tend to buy alongside it.

The system contains 6.3 million nodes and 29 million knowledge edges across 18 product categories. Those edges are relationships,  the commonsense connections between a query and its implied meaning, and they are what give COSMO the ability to surface products a shopper did not know to name specifically.

Think about the difference between a shopper typing “shoes for a wedding” and a shopper typing “black suede Oxford size 10 formal.” The second query is explicit. The first requires common sense. COSMO was built to handle the first kind — which is, in practice, how most people actually search.

How COSMO Works: The Knowledge Pipeline

Understanding how COSMO builds and applies its knowledge helps explain why certain listing practices are becoming more effective and others are falling flat.

Collecting Seed Knowledge

COSMO starts by processing millions of shopper behaviour signals — search sessions, click sequences, co-purchase patterns, and purchase-to-search flows. These raw signals form the basis of its initial knowledge assertions. A large language model (LLM) then generates commonsense inferences from these patterns.

Importantly, those inferences are not applied directly. They pass through a Human-in-the-Loop (HITL) annotation layer, where human reviewers validate, refine, and filter the generated knowledge. This step is what separates COSMO from a system that simply pattern-matches at scale, the validated knowledge represents genuine understanding, not statistical noise.

Building the Knowledge Graph

Once validated, the knowledge is structured into a graph with 15 distinct commonsense relation types. These include relationships like:

• UsedFor — what purpose or context does this product serve
• UsedBy — who typically buys or uses this product
• HasProperty — what qualities or characteristics define it
• CompatibleWith — what other products or scenarios does it pair with
• RelatedTo — what adjacent concepts surround the product category

Together, these relation types allow COSMO to answer questions that keyword matching simply cannot — like whether a “camping lantern” is relevant to someone searching “off-grid power for weekend trips.”

Serving the Knowledge Downstream

Once the graph is built, it feeds into several parts of Amazon’s product discovery system: search relevance, sponsored ad targeting, personalised recommendations, and search navigation refinement. It is also the backbone of Rufus, Amazon’s AI shopping assistant, which now handles over 274 million daily queries and has driven billions in incremental sales.

Whether COSMO and Rufus operate as a single system or in parallel is not publicly confirmed by Amazon. What is confirmed is that the practical optimization guidance for both is identical: structure your content around intent signals, not keyword repetition.

COSMO vs. A9 and A10: What Actually Changed

Amazon sellers have been optimizing for the A9 algorithm since the early days of the platform, and many have adjusted to A10’s greater emphasis on external traffic and seller authority. COSMO does not replace these systems outright — it sits alongside them, shifting the weight of what matters.

Here is the clearest way to understand the difference:

• A9 asked: Does this listing contain the words the customer typed?
• A10 asked: Does this listing have strong seller authority, good conversion, and external traffic signals?
• COSMO asks: Does this product solve the problem the customer described?

That shift matters because it changes the unit of optimization. With A9, you were optimizing for a search engine. With COSMO, you are optimizing for a reader who understands context.

Keyword stuffing is not just less effective in this environment; it actively damages readability without delivering any ranking advantage in return. Listings built with natural language that explain what a product does, who it is for, and when someone would use it are now better positioned than those engineered purely for term frequency.

Amazon also updated its title requirements in early 2025, capping titles at 200 characters and restricting special characters. Non-compliant titles are now auto-updated by Amazon, often in ways the seller would not choose. Getting ahead of this means writing titles that are accurate, readable, and intent-aligned — which is precisely what COSMO rewards anyway.

What COSMO Means for Your Amazon Listings

Titles: Intent Over Index

Your title is still one of the highest-weight fields in Amazon’s algorithm. The guidance has not changed in that regard. What has changed is how you should construct it. The most effective titles in a COSMO-influenced environment lead with the core use case or intent, include the primary keyword naturally, and describe the product accurately for a real person, not a crawler.

“Stainless Steel Water Bottle Insulated 32oz BPA Free Leak Proof Gym” is a keyword list masquerading as a product description. “32oz Insulated Stainless Steel Water Bottle, Stays Cold 24 Hours, Leak-Proof Lid” tells COSMO what the product does and who it serves.

Bullet Points: Use Cases and Audiences

Bullet points should function as a structured answer to the question: who is this for, and what problem does it solve? Each bullet is an opportunity to introduce a specific use case, audience segment, or contextual scenario that COSMO can map to real shopper queries.

Instead of listing features in isolation, pair each feature with its practical benefit and the situation where it matters. “Reinforced non-slip sole” tells a search engine something. “Reinforced non-slip sole, designed for wet kitchen floors and outdoor decking” tells COSMO something it can use to match your product to shoppers who did not know to search for “non-slip kitchen shoes.”

Backend Keywords: Fill the Gaps, Do Not Repeat

Backend keyword fields remain a legitimate optimization lever, but their role has shifted. They are best used for synonyms, regional variants, alternative phrasings, and long-tail queries that do not fit naturally in your visible copy. Repeating terms already in your title and bullets wastes the limited character count and adds nothing.

A+ Content: Context That AI Can Read

Amazon has not confirmed that A+ Content modules are directly parsed by the COSMO knowledge graph, but Rufus, which draws on product detail page data, does read A+ Content. Modules that include detailed use-case scenarios, audience descriptions, contextual imagery, and comparison tables give Amazon’s AI systems more structured information to work with.

Think of A+ Content less as a design exercise and more as a structured brief about your product’s place in a customer’s life.

Discovery Attributes: Do Not Skip Them

The backend fields in your product template, target audience, intended use, and subject matter now feed directly into how COSMO categorises and surfaces your listing. These fields were often overlooked when they felt optional. They are no longer optional in any meaningful sense. Fill them accurately and specifically.

COSMO and Amazon Advertising: The Spillover Effect

COSMO’s influence is not confined to organic search. Amazon’s advertising systems are showing signs of the same intent-based shift.

Broad match targeting has become significantly more permissive. A campaign targeting “stainless steel water bottle” may now serve ads on queries like “insulated metal bottle for hiking,” matching on intent and context rather than exact keyword overlap. This behaviour is consistent with an advertising layer informed by the same kind of semantic understanding that COSMO applies to organic results.

Amazon also launched AI-powered Sponsored Products prompts in 2025, which surface product suggestions during conversational browsing sessions with Rufus. Listings that are well-structured for intent-based discovery, not just keyword-indexed, have a better chance of appearing in these AI-assisted touchpoints.

The practical implication: your listing quality and your ad performance are now more tightly linked than they have ever been. A listing that communicates context clearly benefits both organic ranking and ad matching simultaneously.

How to Audit Your Listings for the COSMO Era

The following questions are a practical starting point for evaluating whether your current listings are positioned for the way Amazon’s search is working today:

• Does your title describe what the product does and who it is for, or does it read like a list of keywords?
• Do your bullet points cover specific use cases, contexts, and audience types — or do they list features in isolation?
• Have you filled in all backend discovery attributes accurately, including target audience and intended use?
• Does your A+ Content include scenario-based copy that places the product in a real customer situation?
• Are your backend search terms used for coverage gaps, not repetition of copy already visible on the page?
• Does your listing answer the question “why would someone need this” in plain, natural language?

If you are reviewing an existing catalogue, prioritise your top-revenue ASINs first. The listings already generating strong conversion data are the ones most worth protecting as the algorithm continues to evolve.

The Bigger Picture: Amazon Search Is Becoming Conversational

COSMO is part of a longer shift at Amazon toward conversational, intent-aware product discovery. More than 300 million customers used Rufus in 2025, with nearly $12 billion in incremental annualised sales attributed to AI-assisted shopping, according to Amazon’s Q4 2025 earnings report. Over 60% higher purchase completion rates were observed among shoppers who engaged with Rufus before buying. These numbers are not trivial, they represent a structural change in how Amazon customers find products.

Sellers who built their strategies entirely around keyword volume are finding diminishing returns. Sellers who understand that their listings now need to communicate context, not just contain terms, are the ones holding ground and gaining visibility.

The sellers gaining ground in 2026 are mostly doing fundamentals well across the board. Clean listings. Accurate attributes. Natural copy. Strong external signals. None of it is complicated. Most of it is just a different way of thinking about who you are optimizing for.

How Meliora Marketing Helps Brands Navigate Algorithm Changes

Keeping up with changes like COSMO while running a business is not easy. The listing audit process, the copy revisions, the backend field review — it takes time that most founder-led brands do not have. That is where a focused Amazon partner makes the difference.

At Meliora, we work with a small number of brands at a time so that every account gets the attention it actually needs. Our Amazon account management services cover listing strategy, catalogue audits, and ad management — all built around what is actually driving performance today, not two years ago.

If your brand is growing past the point where you can manage it alone, our Ascend growth and profitability service is built around margin-driven decisions, efficient ad spend, and the kind of sustained visibility that algorithm changes like COSMO reward over time.

Final Thoughts

The Amazon COSMO algorithm is not a reason to abandon keyword research or stop tracking your BSR. It is a reason to stop thinking about those signals as the end goal. Keywords are still the language shoppers use to find things. COSMO is the interpreter that reads what they actually mean.

The sellers who will do well in this environment are the ones who write listings that serve a real person, one with a specific problem, a particular context, and a reason they need this product and not a competitor’s. That has always been good practice. COSMO just makes it a requirement.

If you want to talk through how your catalogue is positioned for the current state of Amazon search, get in touch with the Meliora team. We are happy to take a look.