Amazon A9 Algorithm: How Amazon Search Ranks Products

Before your garlic press can compete for a spot near the top, Amazon first has to find it as a matching result. These two jobs are what sellers and agencies call the “A9 algorithm”. Amazon itself says “Amazon search” and now describes quite openly how it finds and sorts products.

Using the search “stainless steel garlic press”, we go through both steps: first with Amazon’s own statements, then with what its researchers have published. From there, we show where sellers and vendors can start working on their listings.

What is the Amazon A9 algorithm?

“A9” is the name of a former Amazon subsidiary. According to an Amazon press release from September 15, 2004, A9.com, Inc. opened an office in Palo Alto in October 2003 to develop search technology. In 2004, the subsidiary launched its own web search at a9.com, which Amazon presented as “a search engine with a memory”. Since then, many sellers have used “A9” as the name for Amazon’s search algorithm.

In Seller Central help, Amazon calls this area “Amazon search”, not A9. There is also no “A10 algorithm” in any Amazon source: the name comes from agencies and tool vendors.

What matters for your listing is how search selects and sorts products. Most people who buy on Amazon start with a search query. That makes Amazon above all a search engine where you can buy. Like Google, it has to filter the matching “documents” out of a huge number of them. On Google these are web pages, on Amazon products.

How Amazon search works: matching and ranking

In March 2026, Amazon described for the EU how its search works, in Annex 1 of its DMA compliance report. Search runs in two steps:

1. Matching: Amazon puts together the products that best match the customer’s intent, the so-called “match-set”.
2. Ranking: Ranking models give each of these products a score. The score sets the order. Amazon uses different models for different categories.

The same split already appears in a 2019 Amazon research paper on semantic product search:

product search typically operates in two stages: matching and ranking

How Amazon search works

1. 1

   Search query

   stainless steel garlic press

   Amazon corrects typos itself (“Showing results for”).

3. 2

   Step 1: Matching

   Amazon finds the products that best match the customer’s intent: the “match set”.

   Word matching

   Title, item highlights, bullet points, description, attributes, search terms

   Semantic search

   Meaning of query and product, not only the words

   Only with a buyable offer, a browse node and no block for quality issues.

5. 3

   Step 2: Ranking

   Ranking models give every product in the match set a score, then it is sorted.

   Past customer actions, e.g. purchasesPriceAvailabilityProduct info such as title and description

   Different models per category. According to Amazon no effect: Amazon or seller, FBA, own brand.

7. 4

   Results page

   The position also depends on location, customer account and device.

   According to Amazon, the same logic applies to Rufus product recommendations.

Step 1: Which products Amazon finds for a search

If your garlic press is missing from the matching results, no fine-tuning of the ranking will help. Only what Amazon finds in the first step can get a place in the second.

Direct matching: title, item highlights, attributes and search terms

The 2019 research paper says: “Products that contain words in the query (Qi) are the primary candidates.” Products whose text contains the words of the query are the first candidates. Amazon compares the fields of your listing for this: title, item highlights, bullet points, description and attributes. According to its help, Amazon uses the backend search terms for the lexical matching of an ASIN to a query, which is exactly this word comparison.

For the KNOVELDA garlic press, this means: both words from “stainless steel garlic press” should appear in the listing fields. Each product also needs the right category, the browse node. Our Amazon category tree shows all categories.

Limits of word matching

Matching words do not mean that a product fits the search. Amazon describes the limits of word matching in the same paper:

Pure lexical matching falls short in this respect due to several factors: a) lack of understanding of hypernyms, synonyms, and antonyms, b) fragility to morphological variants (e.g. “woman” vs. “women”), and c) sensitivity to spelling errors.

Hypernyms are broader terms (“kitchen gadget” for garlic press). Synonyms are words with a similar meaning (“wallet” and “billfold”). Antonyms are opposites. If someone searches for “latex free gloves”, a pure word search can also show latex gloves, because two of three words match. Search engines do not always understand words like “free” or “without” correctly.

Word order alone can change the meaning: “business suit” is about a suit, “sock suit” is more likely about socks. The shared word “suit” is not enough to tell the intent.

Stemming helps with singular and plural: search reduces words to their stem. That is why, according to Amazon, either the singular or the plural is enough in search terms.

Spelling mistakes and corrections

Amazon corrects typos itself. The results page then shows, for example, “Showing results for” with the corrected query. The suggestions while typing (Amazon Suggest) also steer customers to common spellings. So you can leave misspelled variants out of your search terms.

Semantic search

This is where Amazon’s semantic search comes in: a model compares the meaning of the query and the product, not just the words. The 2019 paper describes this. In 2024, Amazon’s researchers presented COSMO, a knowledge graph of customer intents and use cases. According to the paper, it is used in search navigation.

Amazon reviews are also a research topic. A 2020 research paper derives from reviews what customers use a product for and who uses it, for example “for running”. Amazon nowhere shows that review text feeds directly into ranking.

Why a product does not appear at all

According to Amazon’s help page on search optimization, a product cannot be found if it has no buyable offer, has no browse node, is suppressed because of quality issues or has a launch date in the future. Seller Central has a tool that determines why an ASIN is not discoverable.

Step 2: Amazon ranking factors

Your garlic press is among the matching results. Now it is about the order. In Seller Central help, Amazon says that factors such as price, availability, selection and sales history determine where your product appears in search results. Products with more sales generally appear higher.

In its 2026 EU report, Amazon names what the ranking models work with:

• past customer actions with the products, for example how often an item was bought,
• information about the product, for example title, price and description.

According to the report, Amazon may show new ASINs higher “for a very short period of time”, so they can collect customer signals at all.

According to Amazon, these characteristics are not taken into account: the EU report says matching and ranking do not distinguish between Amazon selling itself or a seller selling, whether a product ships via FBA, or whether it is an Amazon private label. This is Amazon’s account to the European Commission.

What research says about customer signals

How much clicks and purchases count is described in a 2020 research paper:

Learning to rank (LTR) models rely on several features to rank documents for a given query. Many LTR features are based on users’ interactions with documents such as impressions, clicks, and purchases. We call these features behavioral features. Ranking models are trained to optimize user engagement, and therefore, such behavioral features tend to be the most important training signals.

A garlic press in position 1 gets more clicks than one on page 3, just because of its position. Such signals have to be corrected. Research therefore compares click rates with the expected click rate for each position. Whether Amazon does exactly this in its live system has not been published.

The category also changes what a click says. In clothing, for example, shoppers often click on expensive designer pieces without buying them. For a garlic press, a click means something else. That is why Amazon uses different models.

Personalization: why rankings fluctuate

Two customers do not necessarily see the same results for the same search. According to Amazon, a product’s position depends on location, the customer account and whether someone searches on desktop or mobile, among other things. Amazon does not guarantee any specific placement.

A 2019 workshop paper by the University of Massachusetts and Amazon Search examines how clicks within a session can change the next result pages:

We reformulate product search as a dynamic ranking problem where we leverage users’ implicit feedback on the presented products as short-term context and refine the ranking of remaining items when the users request the next result pages.

The paper studies an approach. It does not show that Amazon uses it this way. For your own measurement, this remains: if you search for your garlic press once in your own browser, you learn little about its ranking. Track the position regularly and under the same conditions.

AI search and Rufus

Rufus is Amazon’s shopping assistant. In Germany it is called Rufus; in the US it has been called Alexa for Shopping since May 13, 2026. According to Amazon’s help, Rufus was trained on catalog data, reviews, community questions and answers, product detail pages and information from the web. For product questions, it uses the information on your product detail page.

According to the EU report, the same description of matching and ranking applies to Rufus’s product recommendations. If you optimize for search, you also optimize for Rufus. How to use Rufus for your own work is covered in our Rufus guide.

Improving your Amazon ranking: what sellers and vendors can do

Amazon does not publish its ranking formula. Its own statements still give you concrete starting points for your listing:

How to fill each field of your listing under the current rules is shown in our article on Amazon listing optimization. The overall picture of Amazon SEO is in a separate article.

Tracking rankings with Keyword Rankings

One garlic press, three keywords: you can still look that up yourself. With hundreds of ASINs, dozens of keywords and several marketplaces, you can no longer keep up. And because of personalization, your own search stays inaccurate.

Keyword Rankings in the AMALYTIX Amazon keyword tool takes over this search. Every week, AMALYTIX checks the first five result pages and stores the organic rankings of your products. You compare the current position with the one a week ago and four weeks ago. The heatmap shows the trend week by week. If a product drops for an important keyword, you get an alert. This also works for competitor products on your watchlist.

Keyword Rankings shows the position. You have to check the reason for a change yourself: is it the title, the price or the stock? Use your listing and your sales data for that.

Conclusion

For the garlic press as for the rest of your range: first, Amazon needs the right words in the right fields and the right category. Then sales, customer actions, price and availability decide the placement. The principle is known; the formula stays with Amazon. Work on both steps and measure the effect over weeks. A single search is not enough for that.

How Amazon SEO differs from optimizing for Google is explained in our article Amazon SEO vs. Google SEO.

Frequently asked questions about the Amazon A9 algorithm

What is the Amazon A9 algorithm?

It is what sellers and agencies call Amazon’s search algorithm. The name comes from A9.com, a former Amazon subsidiary for search technology. Today, Amazon itself speaks of “Amazon search”.

Is there an Amazon A10 algorithm?

Not according to Amazon. The term “A10” does not appear in any Amazon source. It comes from agencies and tool vendors.

What is the name of Amazon’s search engine?

Amazon calls it “Amazon search”. The name A9 goes back to A9.com, Inc., an Amazon subsidiary that opened an office in Palo Alto in 2003 and launched its own web search in 2004.

Which factors determine Amazon rankings?

Amazon names price, availability, selection and sales history. Then there are past customer actions, such as how often a product was bought, and product information such as title, price and description. Amazon uses different ranking models for different categories.

Does Amazon favor FBA or its own brands in search?

Amazon says no. According to its 2026 EU compliance report, matching and ranking do not distinguish whether a product ships via FBA, whether Amazon sells it itself or whether it is an Amazon private label.

How important are reviews for ranking?

Amazon has made no statement on this. In our assessment, ratings work indirectly: they influence whether customers click and buy, and Amazon names purchases as a factor. That review text feeds directly into ranking has not been shown.

How long does it take for changes to show in search?

According to Amazon, up to 72 hours for changes that affect search. A changed title is visible after 24 to 48 hours.