Amazon COSMO: what the "COSMO algorithm" really is and what it means for your listings
Amazon Science presented COSMO at SIGMOD 2024 to improve shopping knowledge; Amazon has not confirmed that the system powers Alexa for Shopping.
What COSMO is
COSMO comes from a research paper by Amazon scientists, presented at ACM SIGMOD 2024: "COSMO: A large-scale e-commerce common sense knowledge generation and serving system at Amazon". The Amazon Science blog describes it as "a framework that uses large language models (LLMs) to discern the commonsense relationships implicit in customer interaction data from the Amazon Store".
In plain words: when many people search for one thing and buy another, there is a reason that nobody wrote down. COSMO tries to write it down. Amazon's own example links "slip-resistant shoes" to "pregnant women" through a relation called used_for_audience. The result is a knowledge graph: products connected to uses, audiences, activities and needs.
What Amazon has said, and what it has not
What Amazon has said:
- COSMO mines common-sense knowledge from customer behaviour in the Amazon Store, using large language models, and stores it as a knowledge graph.
- The paper page says COSMO has been deployed in Amazon search applications, such as search navigation.
- The Amazon Science blog reports that, in its experiments, adding this knowledge improved results on downstream tasks "by as much as 60%". That figure comes from a research benchmark, not from sales.
What Amazon has not said:
- That COSMO is "the Amazon algorithm" or that it replaced Amazon search ranking.
- That COSMO powers Alexa for Shopping. Many articles connect the two, but Amazon has not confirmed it.
- Any list of ranking factors or weights.
So be careful with advice that starts with "COSMO rewards…". Nobody outside Amazon knows the weights.
What it means for your listings anyway
You do not need to know the weights to act on the idea. If Amazon is building knowledge about who a product is for and what it is used for, a listing that says those things clearly is easier to understand, for COSMO-style systems, for Amazon search, for Alexa for Shopping and, above all, for shoppers.
A practical checklist for one ASIN. These are our questions, shaped by the way Amazon describes its knowledge graph, not an Amazon rule:
| Question | Where to answer it |
|---|---|
| What is it? | Title and first bullet |
| Who is it for? | Bullets, A+ content |
| What is it for, and when? | Bullets, description |
| What can it do, what can it not do? | Bullets, Item Highlights |
| What does it work with? | Attributes, bullets |
| Which variant fits which need? | Variation names, comparison chart |
Keep the keywords. Amazon SEO still starts with the words shoppers type into the search bar; the use, audience and context sit on top of them, in plain sentences. For the field-by-field version, see How to get an ASIN ready for Alexa for Shopping.
How to check whether it works
You cannot see COSMO, but you can see results. For each important ASIN, ask Alexa for Shopping the questions your buyers ask, with use and audience in them ("running shoes for flat feet", "a gift for a new mum"), and note which products it names. Repeat over several days, because answers change. An Alexa for Shopping optimization platform does this per ASIN and saves every answer, so you can see whether a listing change moved anything.
Sources
- Amazon Science, COSMO: A large-scale e-commerce common sense knowledge generation and serving system at Amazon, SIGMOD 2024.
- Amazon Science, Building commonsense knowledge graphs to aid product recommendation, 10 May 2024.
- Amazon, Amazon Rufus: Amazon's AI shopping assistant gets smarter and more personal, 18 November 2025.
- Last checked: 13 November 2026.
Frequently asked questions
What is the Amazon COSMO algorithm?
COSMO is an Amazon Science system that uses large language models to turn customer behaviour into common-sense knowledge, such as who a product is for and what it is used for, stored as a knowledge graph. Amazon says it is deployed in Amazon search applications such as search navigation.
Does COSMO power Alexa for Shopping?
Amazon has not said so. Many articles connect COSMO with Rufus, now Alexa for Shopping, but that is an industry reading, not an Amazon statement.
Did COSMO replace Amazon's A9 or A10 ranking?
Amazon has not said that. "A9" and "A10" are names the industry uses for Amazon search ranking; Amazon has not published how its ranking works or how COSMO fits into it.
How should I write a listing for COSMO?
State plainly what the product is, who it is for, what it is used for and when, what it works with and which variant fits which need, on top of the keywords shoppers search. That helps every system that reads your listing, and the shopper most of all.