Short version: "The Amazon algorithm" isn't one system — it's three, stacked on top of each other. A9/A10 is the ranking engine that indexes listings and is still fully active. COSMO is an AI knowledge-graph layer built on top of A9/A10 that matches shopper intent, not a replacement for it. Alexa for Shopping is the conversational layer that answers questions directly — it absorbed Rufus, Amazon's earlier chatbot, when Amazon retired it on May 13, 2026. None of the three replaced the one below it.

If you've read conflicting things about "Amazon's new algorithm," you're not alone — a lot of seller-blog content flattens three separate systems into one story, and some of it still talks about Rufus as if it's current. This is the explainer version: what each layer actually is, in what order they were introduced, and what that means for how you optimize a listing.

The stack, in the order it was built

Each layer was added on top of the last — none of them turned the previous one off:

  • A9 (and the "A10" era) — Amazon's original, foundational search and ranking engine. It's the system that decides which searches your listing is even eligible to appear in, and it's still running today.
  • COSMO — an AI commonsense-knowledge layer, published as a research system at the 2024 SIGMOD conference, that sits on top of A9/A10 and re-ranks results by matching a shopper's underlying intent, not just their literal words.
  • Alexa for Shopping (formerly Rufus) — the conversational layer. Rufus launched in 2024 as a separate chat assistant; Amazon retired it on May 13, 2026 and folded its product knowledge into Alexa for Shopping, which now answers questions directly inside the search bar, search results, and product pages.

The practical implication: a listing has to clear A9/A10's indexing and ranking first. COSMO and Alexa for Shopping can only surface a listing they can already find.

Layer 1: A9/A10 — still the foundation

A9 is Amazon's original keyword-matching and ranking engine. It reads your title, bullets, backend search terms, and attributes to decide which queries you're indexed for, then ranks the resulting candidate pool using performance signals — click-through rate, conversion rate, sales velocity, and account health chief among them.

"A10" is not a name Amazon has ever officially used. It's shorthand the seller community adopted to describe how A9's ranking behavior has evolved — weighting conversion and external traffic more heavily than it used to. Useful as a mental model; not a confirmed, separate algorithm.

Either way, this layer hasn't gone anywhere. Every keyword you place in a listing is still doing its original job: getting you into the pool A9/A10 controls.

Layer 2: COSMO — the intent layer, not a replacement

COSMO is a large-scale commonsense knowledge graph Amazon built using LLMs, described in a paper presented at SIGMOD 2024 (Amazon Science). Instead of just matching words, it captures relationships — what problems a product solves, what it's typically used with, what a shopper searching one phrase actually means.

That's why a listing can rank for a search like "something to keep coffee hot on a commute" without ever using that exact phrase — COSMO connects the intent behind the query to products that satisfy it. But COSMO doesn't decide whether your listing is in the running in the first place; A9/A10 still does that. Think of COSMO as smarter re-ranking and expansion on top of a system that never stopped running underneath it.

Layer 3: Alexa for Shopping — what happened to Rufus

Rufus was Amazon's standalone generative-AI shopping chatbot, launched in 2024 and used by an estimated 300 million customers by 2025. On May 13, 2026, Amazon retired Rufus as a separate product and replaced it with Alexa for Shopping — combining Rufus's product knowledge with the personalization of Alexa+.

The functional difference matters for sellers: Rufus lived in its own chat window. Alexa for Shopping is embedded directly in the main search bar, the search results page, and product detail pages, and no sign-up or Echo device is required — it's on by default for signed-in U.S. shoppers. It answers questions, builds comparisons, tracks prices, and can complete purchases through an agentic "Buy for Me" feature.

For sellers, the source material didn't change: Alexa for Shopping still draws on your listing copy, attributes, reviews, and Q&A. What changed is that those answers now show up inline in ordinary search, not in a separate assistant tab — so more shoppers see them without opening anything extra. See what Alexa for Shopping is and how it answers shopping questions for the deeper dive.

How one search touches all three layers

It helps to trace an actual query through the stack rather than think about the layers in isolation.

Example — a shopper searches "mug that won't leak in my bag"

A9/A10: builds the candidate pool from listings indexed with related terms — "leak-proof travel mug," "spill proof tumbler" — pulled from titles, bullets, and backend search terms.

COSMO: recognizes the underlying intent ("won't leak in my bag") even for listings that never used that exact phrase, and re-ranks the pool by how well each product actually satisfies that need.

Alexa for Shopping: if the shopper asks a follow-up like "does it fit in a backpack side pocket?", it pulls the answer directly from that listing's attributes, bullets, or Q&A — if the data is there to extract.

A listing that's missing at any one layer loses visibility at every layer after it. Weak keyword coverage means A9/A10 never puts you in the pool; vague copy means COSMO can't match intent even if you are; incomplete attributes and Q&A mean Alexa for Shopping has nothing to answer with.

Checklist: is your listing built for all three layers?

  • ☐ Primary and secondary keywords placed in title, bullets, and backend terms (A9/A10 indexing)
  • ☐ Title within the 75-character limit (in force since July 27, 2026), with overflow moved to Item Highlights
  • ☐ Copy written in clear, natural, problem-solving language rather than stuffed keyword lists (COSMO intent matching)
  • ☐ Product attributes fully filled in — none left blank
  • ☐ Reviews and Q&A monitored and seeded with real shopper questions (Alexa for Shopping's source material)
  • ☐ A+ Content claims also present as indexable text, not only baked into images

Common mistakes when reasoning about "the algorithm"

  1. Assuming COSMO or Alexa for Shopping replaced A9/A10. They didn't — they're layers on top of a system that's still fully active.
  2. Optimizing for Rufus. Rufus was retired May 13, 2026; any advice still framed around a separate Rufus chat window is out of date.
  3. Treating "A10" as an official, documented algorithm. It's community shorthand for how A9 has evolved — useful context, not a confirmed system to chase specific "hacks" for.
  4. Writing for one layer and ignoring the others. A listing that's keyword-rich but has blank attributes and no Q&A will index fine and still go unanswered by Alexa for Shopping.

Check how your listing performs across all three layers

OptimalCentral's AI Listing Audit scores your keyword indexing, intent-match copy, and AI-answer readiness in one pass — free, in under a minute.

Start free trial →

For the tactical, field-by-field version of this guide, see Amazon SEO: the complete 2026 guide. To go deeper on one layer, see Amazon keyword research for A9/A10, or Alexa for Shopping vs. Amazon search for the conversational layer.

Frequently asked questions

Did COSMO replace A9 or A10?

No. COSMO is an AI intent layer that sits on top of A9/A10, which is still the active engine that indexes listings and builds the candidate pool for a search. COSMO re-ranks and expands that pool by matching shopper intent; it doesn't replace the underlying system.

What happened to Amazon Rufus?

Amazon retired Rufus, its standalone AI shopping chatbot, on May 13, 2026. Its capabilities were folded into Alexa for Shopping, which combines Rufus's product knowledge with Alexa+'s personalization and is now embedded directly in the search bar, search results, and product pages rather than a separate chat window.

Is Amazon A10 a real, official algorithm?

Amazon has never officially confirmed an algorithm called A10. It's a label the seller community uses to describe how A9's ranking behavior has shifted over time — weighting conversion rate, external traffic, and seller performance more heavily. Treat it as informal shorthand, not a confirmed separate system.

What is Amazon COSMO in simple terms?

COSMO is Amazon's commonsense knowledge graph, built using large language models and published as a research system at SIGMOD 2024. It lets Amazon search connect a shopper's query to what they actually need, even when their words don't match a listing's exact keywords, by reasoning over relationships between products, uses, and problems.

Do I need to optimize differently for each layer of Amazon search?

You optimize the same listing for all three, but each layer reads it differently. A9/A10 needs your keywords in the right fields to index and rank you. COSMO needs clear, intent-matched language, not stuffing. Alexa for Shopping needs complete, structured data — attributes, Q&A, reviews — since it pulls direct answers from your listing rather than just ranking it.