Amazon search has been described the same way for years: match the query to indexed terms, then rank the matches mostly on the likelihood of a sale. That description is now incomplete. Amazon Search has published research on retrieval systems that reason about why someone is searching, not just what they typed — and the practical effect is showing up in listings that rank for queries they were never explicitly optimised for.
What actually changed
Classic Amazon retrieval is lexical. A shopper types "running shoes for flat feet", the engine finds products indexed for those tokens, and behavioural signals sort them. It works well when the query contains the product noun and the product page contains it back.
It works badly for everything else — and "everything else" is a large share of modern queries. "What do I need for a first apartment", "gift for someone who camps", "will this fit a 2019 Civic". These queries contain intent and context, and almost none of the words that appear on the right product page.
The shift in one line: Amazon is moving from does this listing contain the query to does this product satisfy what the shopper is trying to accomplish. Listings that only answer the first question are now competing on a narrower set of queries than they used to.
What COSMO is
COSMO is the name Amazon researchers gave to a large-scale common-sense knowledge system for e-commerce: mining shopper behaviour and product data to build explicit relationships like this product is used for that activity, this attribute matters to that audience, people buying this usually also need that. Those relationships then feed retrieval, so a query expressing a situation can reach products described in completely different words.
You will not find a COSMO dashboard in Seller Central, and Amazon has not published a ranking checklist for it. What you can observe is the downstream effect: more traffic arriving on long, conversational, situational queries, and detail pages with rich, specific, well-structured content picking up impressions on phrases nobody put in a backend field.
The same underlying capability powers Rufus, Amazon's shopping assistant. Optimising for one is largely optimising for the other, which is convenient, because it means there is one job here, not two.
What it means for listings
Three consequences, in order of how much money they move:
Writing for intent
Practical rewrite rules we apply across managed accounts:
1Name the shopper in the first bullet
Who is this for, in their words. "For side sleepers with shoulder pain" reaches an entire intent cluster that "medium-firm memory foam" does not. Keep the specification — add the person.
2Answer the three questions a human would ask
Pull them from your own returns and messages: does it fit, how long does it last, what is in the box. Put the answer in plain text on the page. These are exactly the questions conversational search surfaces, and an unanswered one is a lost query.
3State compatibility explicitly
Models, sizes, standards, what it does not fit. Compatibility is the single most common reason a shopper abandons a page, and the single most common thing an AI assistant is asked to verify.
4Mirror the language in your reviews
Reviews are the most honest description of your product that exists. If buyers keep calling it "quiet enough for a nursery" and your listing says "low decibel operation", change your listing.
Attributes are ranking data, not admin
Most brands treat the attributes tab as compliance. Treat it as ranking input and fill it like one: material, size, age range, occasion, compatible models, special features, target audience, unit count. A complete attribute set does three jobs at once — it powers filters shoppers use to narrow results, it feeds the structured data that assistants read back, and it removes ambiguity from retrieval.
| Field | Weak | Strong |
|---|---|---|
| Title | Premium Stainless Steel Water Bottle 32oz | Insulated Stainless Steel Water Bottle 32oz — Fits Car Cup Holders, Keeps Cold 24h |
| Bullet 1 | Double-wall vacuum insulation | Keeps drinks cold 24 hours and hot 12 — tested for gym, commute and job-site use |
| Attribute | Material: Steel | Material: 18/8 stainless · Capacity: 32 fl oz · Lid: screw-top · Dishwasher safe: yes |
| A+ module | Brand story, no specifics | Comparison chart with sizes, fit-in-cup-holder diagram, care instructions |
How to tell if it is working
You cannot see COSMO, but you can see its shadow in your own reports:
- New query rows in Search Query Performance that you never targeted — especially longer, situational phrases.
- Impression growth without rank change on your head terms, which usually means you became eligible for adjacent queries.
- Falling reliance on exact-match PPC for discovery, as organic picks up the long tail you used to buy.
None of this makes keyword research obsolete. It makes keyword research the floor rather than the ceiling: get indexed for the terms that describe your product, then write the page for the person who needs it.
Frequently asked questions
Is COSMO replacing the A9 algorithm?
Nothing suggests a clean replacement. The realistic reading is layered retrieval: lexical matching still decides whether you are eligible for a query, and intent and common-sense reasoning increasingly influence which of the eligible products get surfaced and in what order.
Do keywords still matter on Amazon in 2026?
Yes. Keyword relevance is still the entry ticket — a product that is not indexed for a phrase cannot rank for it. What has changed is that stuffing every variant into the backend no longer buys you much, while clearly expressing use case, audience and compatibility now does.
How do I optimise for Amazon Rufus and COSMO at the same time?
They reward the same things: complete structured attributes, plain-language answers to real shopper questions in bullets and A+ content, and consistency between your title, attributes and reviews. Write the listing so a human assistant could answer a question from it, and both systems benefit.
Get your listings rewritten for intent, not just keywords
Listing SEO is part of our $999/month Amazon management — keyword research, rewrite, attributes and A+ content included.
See Amazon SEO services →