Amazon COSMO Algorithm: What It Is and How Authors Can Use It

The Amazon COSMO algorithm maps why people buy. Here is what it is, what Amazon's own paper shows, and how to use it for your book, in plain English.

By Gundi Gabrielle

13 min read

Start Publishing With AI

🚀 AI-Powered Publishing Studio✍️ Full Writing App🎧 Audiobook & Translation Generators🤖 AI Publishing Team🔍 Viral Topic Gap Finder📢 Ads Generator🎯 AI Marketing Automation

Alrighty, let's talk about the Amazon COSMO algorithm... because it's one of the few times Amazon has told us, in writing, how it thinks about the reasons people buy things.

And the COSMO algorithm matters for your book more than most authors realize.

I've read the whole paper (yes, all 13 pages :)) and I'll give you the plain-English version of how the COSMO algorithm works. It's part of my Amazon KDP algorithm guide.

What is the Amazon COSMO algorithm?

The Amazon COSMO algorithm is a giant knowledge graph (a web of connected facts) that links products to the reasons people buy them: who they're for, what they're used for, and when.

COSMO is short for Common Sense Knowledge Generation and Serving System. Fancy name, simple idea.

Here's Amazon's own example. Someone searches "shoes for pregnant women." Amazon's science team says the system should work out that pregnant women want slip-resistant shoes. Nobody had to write "for pregnant women" on the shoe listing. Amazon learned the link by watching what people search for and buy.

Pretty smart, right?

Now swap the shoes for your book. Who is it for? What problem does it solve? When does the reader pick it up? COSMO is built to answer exactly those questions.

👑 Wondering what your book looks like to COSMO? Golden Bestseller Empire maps it for you. → See how it works

Is the COSMO algorithm the Amazon ranking algorithm?

No. The COSMO algorithm is the layer that figures out what a shopper means. The ranking itself is still done by Amazon's A9/A10 system.

I want you to get this one right, because I see it wrong everywhere.

Think of COSMO as the translator. A reader types "calm book for a tired working mom." COSMO works out the intent behind it and hands Amazon a list of books that fit. Then A9/A10 decides who shows up first, based on sales, conversion and relevance.

Amazon says the same thing about search in general. KDP's help page lists these factors: "Past sales history," "Availability," "Length of time the item has been listed on our website" and "Popularity."

So COSMO gets you into the right conversation. Sales and conversion win it. You need both. I go deeper on that in my A10 algorithm guide.

Why did Amazon build the COSMO algorithm?

Old product databases only knew what a product is, not why someone wants it, so they couldn't connect a shopper's words to a shopper's need.

Amazon's paper says it plainly: existing e-commerce knowledge graphs "fail to discover user intentions." A jacket database knows the size and the fabric. It doesn't know the buyer is trying to stay warm on a cold school run.

Same problem with books. A listing that says "a guide to wellness" tells Amazon nothing about who needs it or why.

That gap, between what a person types and what they actually want, is the whole reason the COSMO algorithm exists. It's also the gap your book page has to close.

👑 Want to be the book COSMO picks? Golden Bestseller Empire plans it from day one. → See the knowledge graph

How does the COSMO algorithm learn why people buy?

The COSMO algorithm starts with real shopper behavior, asks a large language model to guess the reason, and then filters the guesses with human reviewers and trained models.

Here's how it goes, step by step:

1. 1Collect behavior. Two kinds: a search followed by a purchase, and two products bought together. Odd pairs get tossed out, like two items from totally different categories.
2. 2Ask an AI why. A language model explains each pair using a named relation, like "what is this product capable of?"
3. 3Check the answers. People score a sample for plausibility (could this be true?) and typicality (is this the usual reason?). A trained model scores the rest, and anything with a plausibility score under 0.5 is dropped.
4. 4Teach it. The best examples become training instructions.
5. 5Scale up. Amazon fine-tuned LLaMA 7B and 13B models on only 30,000 hand-scored examples, and COSMO-LM grew the graph across 18 product categories.

So the COSMO algorithm is built from what shoppers actually do. AI makes the guesses, people check them. And nobody is writing keyword lists by hand.

How big is the Amazon COSMO algorithm?

According to Amazon's paper, COSMO has 6.3 million nodes, 29 million edges, 15 relation types and 18 product domains.

A node is one thing in the graph, like "pregnant women" or "slip-resistant shoes." An edge is the connection between two of them. A relation is the label on that connection, like "used for" or "used by."

That's big. And it learns from both searches and purchases, so it keeps getting sharper. Amazon refreshes the model daily from new shopper behavior.

What did Amazon's tests show?

In Amazon's tests, adding COSMO knowledge made search results about 60% better on one score, and a live test on about 10% of U.S. traffic lifted sales 0.7%.

Here's the paper's own list:

• ✦Search relevance, frozen models: 60.06% higher Macro F1 and 29.26% higher Micro F1 than the best comparison model.
• ✦Search relevance, trainable models: 27.81% higher Macro F1 and 22.30% higher Micro F1.
• ✦Live A/B test: about 10% of Amazon's U.S. traffic, over several months, with "a notable 0.7% relative increase in product sales."
• ✦Navigation engagement: up 8% in the same test.

Amazon says that sales lift is worth hundreds of millions of dollars a year. Not bad for a "common sense" system... :)

Amazon also confirms it: "COSMO has been deployed in Amazon search applications such as search navigation." Search navigation is how shoppers move from "camping" to "winter camping" to "winter boots."

👑 Your plan holds your reader's intent. Golden Bestseller Empire maps it for you. → Meet the Book Planner

What does the COSMO algorithm look for in a book?

Amazon's paper lists 15 relations, and I add 3 book-specific ones (Author, Series and Format), so you map 18 for a book.

The 15 cover who the book is for, what it does, when and where it's used, what the reader wants, and what it goes with. The extra three are the author as a person Amazon knows, the series the book belongs to, and the format (Kindle, paperback or audiobook).

I break down every single one in my COSMO relations guide. And this is the Coverage Map Golden Bestseller Empire builds for your book, with sample phrases:

Don't let 18 scare you. You write one clear book, and the same few ideas show up in the right places.

What does a COSMO algorithm knowledge graph look like for a book?

It looks like your book at the center, with the 18 relations branching out from it, each one filled with phrases in your reader's own words. Golden Bestseller Empire's Knowledge Graph Generator builds it from your Book Planner blueprint. Here is a real output for a sample book:

Six groups: Audience, Classification, Functional, Context, Interest and Desire, and Book Specific. And here's the same graph as a visual map:

See the Core Entity in the middle? That's the one idea everything else connects to. More on that next.

Why does one clear idea beat five fuzzy ones?

Because COSMO rewards the typical reason people search, not every possible reason, so a book with one precise focus beats a book that tries to cover everything.

This part of the paper is the one I tell every student about. Human reviewers scored the AI's guesses. For search-and-buy pairs, "more than one-third" were typical (35.0%). For two products bought together, only 9.0% were.

The lesson is simple. Ask yourself: what do most buyers in my niche actually want? Build around that.

Let's say you write for working moms. "Working Mom Guilt Reset" is one clear idea. "Everything about moms, work, health, money and sleep" is five fuzzy ones. COSMO can place the first one. The second just confuses it.

I call that one clear idea your Core Entity. It's the single concept Amazon files your book under. Here's the formula:

Reader + Primary Gap → Transformation = Core Entity

Here it is with a real example:

Working Mom + Guilt → Reset = "Working Mom Guilt Reset"

One book, one main idea. Another book in your series gets the next neighboring idea.

Does Amazon read the words inside your book?

Yes. Amazon has confirmed it: your manuscript is read, and it matters for where your book shows up.

The KDP Read Sample page says: "When customers search for books on Amazon.com, we use the actual words that match inside the book to return the best possible selection of books."

That's huge, because most authors only polish the cover and the description.

Your introduction, your table of contents and your chapter subheaders are all indexed. Amazon's preview shows roughly the first 10% of your book, so the intro does a lot of the work. Write those pages in the exact words your reader uses, name the real daily moments (the 5pm chaos, the school drop-off), and you've given COSMO plenty to match.

👑 Ready to try it on your book? Golden Bestseller Empire is AI-assisted. You write it. → Start with the Book Planner

How can you test your book page against the COSMO algorithm?

Read your title, subtitle and description out loud and ask whether a stranger could name your reader, their problem and the moment they'd use the book.

This is a 10-minute check you can do today, with no tools:

1. 1Cover the cover. Look only at the words. Can you tell who it's for?
2. 2Underline the reader's words. Are they the same words a real person would type into the Amazon search bar?
3. 3Count the repeats. Your one main idea should show up in the title, subtitle, description and your first pages.
4. 4Ask a friend. Hand them the page for 10 seconds and ask, "Who is this for?" If they hesitate, the COSMO algorithm will hesitate too.

If you pass, great. If not, fix the description first. It's the quickest win.

Where does the COSMO algorithm fit with Alexa for Shopping?

The COSMO algorithm works behind the scenes in search, while Alexa for Shopping is the chat assistant that replaced Rufus on May 13, 2026.

You'll find Alexa for Shopping in the Amazon app menu. It isn't the search bar. Amazon announced the rename and says shoppers can find products "based on activity, event, purpose, and other specific topics and use cases," which is exactly the kind of thinking COSMO maps.

Good news: nothing changes in your strategy. Same planning, new name. I explain how the two connect in COSMO and Alexa for Shopping.

What mistakes do authors make with the COSMO algorithm?

The biggest mistake is writing for everyone, because the COSMO algorithm needs one clear reader and one clear reason to place your book.

Here are the ones I see most:

• ✦A fuzzy promise. "A guide to better living" could be for anyone, so it gets matched with no one.
• ✦Different words in different places. Your title says "stress," your description says "burnout," your intro says "overwhelm." Pick the words your reader actually types and stay with them.
• ✦Only polishing the cover. A pretty cover helps conversion, but it tells the COSMO algorithm nothing about the reader.
• ✦Stuffing the seven backend keyword boxes. They help a little. They are not the main event. Your title, subtitle and description are.
• ✦Chasing five audiences. One book, one reader. The next book can serve the next reader.

Sound familiar? Don't worry... all five are easy to fix once you see them :)

How do you use the COSMO algorithm for your book?

Make your book's reader, problem, outcome and moment of use crystal clear, and say them the same way everywhere.

Here's my checklist:

1. 1Who is it for? One specific reader, not "everyone."
2. 2What do they want? The outcome, in their words.
3. 3What does the book do? Its main job.
4. 4When and where do they use it? The moment, the season, the place.
5. 5What goes with it? A journal, a course, a routine.
6. 6Who are you? Your story and credentials. Your author page counts.

Then repeat the same ideas in your title, subtitle and description first, because those main keywords do the heavy lifting. After that, use A+ content, your intro and your subheaders inside the book. The seven backend keyword boxes in KDP are a smaller, supporting help (up to seven phrases of two or three words that a reader would really type). Not stuffed. Natural.

If your description can't answer most of those in plain words, that's your to-do list.

What does a book built for the COSMO algorithm look like?

It names the reader, the problem and the moment in plain words, again and again, so Amazon can place it with confidence.

Take a reader who types "calm book for a tired working mom." That one search already holds four clues: who (a working mom), what she wants (calm), what she does (winds down) and when (after a long day).

A page that says all four matches the search from every angle. A page that says "a guide to wellness" matches none.

How does Golden Bestseller Empire use the COSMO algorithm for you?

It turns your Book Planner answers into an algorithm-ready outline plus a knowledge graph built on the 18 relations, then uses that graph across your whole Amazon listing.

You answer guided questions about your reader, their biggest gap and the change you promise. The app builds the blueprint, and you can see it as text, a visual map or a coverage table. That one graph then builds your book description, A+ content, backend keywords, translations and Amazon ads keywords.

I don't know of another app that does this.

You write the book yourself (or dictate it). It's AI-assisted, not AI-written. I used it for AI for Beginners, a Blissful Kittens Explain book, and it hit #1 within 3 days, a lot faster and easier than doing it by hand. See the book here.

Curious where this goes next? Read my guide to what a knowledge graph is and why it beats keywords alone.

Related reading

Want to go deeper? These guides pair well with this one:

• Amazon KDP algorithm: the full picture
• COSMO relations for books
• COSMO and Alexa for Shopping

What's the one thing to remember about the Amazon COSMO algorithm?

The COSMO algorithm maps what a shopper means to the books that fit, and the clearer your book is about its reader and purpose, the easier it is to match.

Pick one precise idea. Say it plainly. Repeat it in the right places.

Ready to rock the Amazon COSMO algorithm with your next book?

Created By

Gundi Gabrielle

Top 100 Business Author · Carnegie Hall Conductor · Founder of Golden Bestseller Empire

Hey, I'm Gundi, and for over a decade, I've helped thousands of everyday people turn what they know into a bestselling book.

Along the way, I sold over 350,000 books, and was listed in Amazon's Top 100 Business Author Charts, ahead of authors like Simon Sinek, Brené Brown, Tim Ferriss, Robert Kiyosaki, or Brian Tracy.

No traditional publisher. No massive budget. Just understanding Amazon's algorithm and researching what readers actually want.

In my earlier years, I was a classical conductor & organist, touring the world, incl. Carnegie Hall & a performance for the Pope on St. Peter's Square.

These days, I love traveling as a Digital Nomad and orchestrating busy AI Agents 😎. Out of that came Golden Bestseller Empire — a complete AI-Powered Book Publishing Studio now helping authors around the world.

Can't wait to see you inside & help you publish your dream:

Start Publishing With AI →

Featured In: