How Amazon Organic Ranking Works: What Is Actually Documented

What Amazon states, what its published research describes, and what sellers only observe about organic search ranking, and how to measure it.

Most explanations of Amazon search ranking mix three kinds of statements without saying which is which: what Amazon has officially said, what Amazon's own researchers have published, and what sellers believe after watching their listings move. This guide keeps them apart. Each section says where a claim comes from, so you can decide how much weight it deserves.

Amazon does not publish its ranking formula, and research papers describe systems at a point in time, not today's production code. Anyone quoting exact weights is guessing.

"A9" and "A10": names, not documentation

"A9" was a real Amazon subsidiary, A9.com, set up in the early 2000s to build search and search advertising technology (the name is a numeronym for "algorithms"). Amazon's 2016 paper on product ranking came from researchers at A9.com. In 2019 the A9.com website was taken down.

"A10" is different. We found no Amazon announcement, help page, or paper that uses the term. It is a seller-community label for "the algorithm after some change." Treat "A10 ranking factors" as practitioner opinion, not documentation.

What Amazon states about how results are ordered

Amazon's most direct public statement is its customer help page on searching. The default sort is "Featured," and Amazon writes: "We consider many factors when choosing featured results." The page then lists these examples:

• Customer actions, such as how often an item was bought
• Information about the item, such as title, price and description
• Availability
• Delivery speed
• Costs, such as shipping costs
• Whether Amazon thinks the item will be of interest, such as new items
• Customer preferences, such as items you previously purchased or clicked on

The list (identical on Amazon's US and UK help pages) gives no weights and says "such as," so it is not exhaustive.

Amazon's seller-facing guidance adds a few points. Its Amazon SEO guide says listing quality and account health "can contribute to search rankings," that competitive pricing "can improve your product ranking," that back-end keywords can help a product rank for specific searches, and that keyword stuffing can hurt rankings.

Step one: matching (getting into the result set)

Before anything is ranked, Amazon has to decide which products match a query at all. Amazon's 2016 paper describes ranking as ordering the products "in a match set," so a product that does not match never gets ranked.

What Amazon states. Sellers supply text that search uses: the title, bullet points, description, product attributes, and back-end search terms. Amazon's guidance recommends putting primary keywords and details like brand, product type, and size in the title. It describes back-end search terms as keywords customers do not see that can still help a product rank. The search terms field has a strict length limit, so check current Seller Central guidance for the exact rules.

What Amazon's research describes. Matching is more than exact word overlap:

• Product type understanding. The 2016 paper describes detecting whether a query word is a product type or a modifier. For example, "dress" is the product type in "casual dress" but a modifier in "dress shoes." Products are labeled with the product types that describe them, and these labels become ranking features.
• Semantic matching. A KDD 2019 paper by Amazon researchers says pure keyword matching falls short: it misses synonyms, trips over word forms like "woman" vs. "women," and breaks on misspellings. The authors trained a neural matching model on customer behavior data and tested it online.
• Graded relevance. Amazon's public Shopping Queries Dataset labels each query-product pair as Exact, Substitute, Complement, or Irrelevant. Relevance is graded, not just yes or no.

What this means in practice. A missing attribute or a vague title can keep a product out of the match set for queries it should serve. But keyword coverage only buys eligibility. It does not decide order.

Step two: ranking, where behavior dominates

Once a product is in the match set, Amazon's published research is consistent: customer behavior carries the most weight.

The 2016 SIGIR paper by Sorokina and Cantú-Paz describes:

• Separate machine-learned ranking models, usually one per category and marketplace (their example is Books in Japan)
• Training labels taken from customer actions: purchases, add-to-basket, and clicks
• Gradient boosted trees as the model type, evaluated in A/B tests on converting customers, units purchased, and revenue
• Features covering the product (for example sales and reviews), the query and context (for example query specificity and customer status), and text similarity

The paper's key observation is that many products can look equally relevant by text, so popularity has to separate them. In its words, "behavioral features drive the rankings in Amazon Search" much more than in web search. They typically account for most of the model's variance reduction. The authors also describe correcting for position bias, because shoppers click top results more often regardless of quality.

A 2022 CIKM paper by Amazon researchers goes further. It calls behavioral features, built from impressions, clicks, and purchases, the most effective learning-to-rank features. It describes product-query behavioral features as the ones ranking models rely on most. Its reasoning: a product that semantically matches a query is not relevant if the shoppers who search for it don't want to buy it.

Sellers commonly observe that rank follows sales velocity for a keyword, often with a lag. That fits the research, but no public source gives Amazon's exact lookback windows or weights.

Price, availability, and delivery speed

Amazon states that price, availability, delivery speed, and costs such as shipping are among the factors for Featured results. Its seller guidance separately says competitive pricing can improve ranking.

Amazon's research describes price and delivery speed as examples of non-behavioral product features fed into ranking models, alongside category, brand, and text.

Regulators allege a harder edge. In its 2023 antitrust complaint, the FTC alleged that Amazon "can bury discounting sellers so far down in Amazon's search results that they become effectively invisible" when they offer lower prices elsewhere. That is an allegation, Amazon disputes the case, and it has not been decided. Still, it describes price as a ranking lever, not just a conversion lever.

Location matters because delivery speed depends on where the shopper is. The same product can show a different delivery promise at another ZIP code, so results seen from one address don't carry over to another.

The Featured Offer (Buy Box) and its link to search

On a product detail page, Amazon features the most competitive offers in the Offer Display, "sometimes called the 'Buy Box.'" Amazon states that Featured Offer prices are commonly at or below the lowest-priced alternatives. Offers with fast, free shipping and more certain delivery dates are more likely to be featured. An out-of-stock offer cannot be featured, and a price that is too high or suspiciously low can make an offer ineligible.

Amazon does not say how the Featured Offer feeds search ranking. The link is indirect: the price and delivery promise on a search result generally come from the featured offer, those are stated ranking factors, and the clicks and purchases that follow depend on that offer.

Sellers commonly observe that a listing that loses its Featured Offer, or shows no featured offer at all, loses search visibility. Amazon has not documented this.

Reviews and ratings

Amazon states that "Avg. Customer Review" is a separate sort option based on star average and review count. Ratings are also a stated input to Amazon's Choice (see below).

Amazon's research describes reviews as an example of a product feature in ranking models. The 2022 paper describes a "trust bias": shoppers rely on reviews and engage less with products that lack them, which feeds back into the behavioral signals.

So ratings can act as a direct feature and through their effect on clicks and conversion. The research does not say which effect is larger.

Personalization: no two shoppers see exactly the same page

Amazon states that customer preferences, "for example, items you previously purchased or clicked on," are a factor. The 2016 paper says ranking context can include "some user features." Together with delivery-dependent factors, this means a search result page is not one fixed list. It varies with the shopper, the address, and the time. To compare pages, keep the address, signed-in state, and time window fixed.

Sponsored results are a separate system

Sponsored Products are cost-per-click ads. Amazon Ads says they may appear "at the top of, alongside, or within shopping results" and on product pages. They are labeled as sponsored, and they are not organic positions.

Two points are often confused:

• Ads do not buy organic rank directly. Amazon's seller guide says ads can help offers "rise through search results," but no Amazon source names ad spend as an organic ranking factor.
• Ads can influence organic rank indirectly. Ad-driven clicks and sales are still customer behavior. The 2022 Amazon paper names sponsored results and promotions as tools sellers use to push new products into the "search feedback wheel."

The FTC complaint also alleges that Amazon replaced relevant organic results with paid ads. Whatever the outcome, organic listings share the page with ads, so positions that mix the two are misleading.

Badges: what Amazon says they mean

Amazon's Choice. Amazon's FAQ says these products are "highly rated, well-priced and available to ship immediately." The badge considers ratings, price, popularity, product availability, and fast delivery. Brands cannot pay or apply for it, and third-party products are eligible. Sellers report that the label shown on search results has changed over time, for example to "Overall Pick."

Best Seller. Amazon's seller guidance says the badge appears on detail pages for products that meet an ordering threshold and rank #1 in a category with at least 100 products. It is based on Best Sellers Rank, which reflects sales volume (with recent sales weighted more) and is not influenced by page views or reviews. Amazon also states that a top BSR does not always mean a high organic search ranking.

Amazon does not say badges are ranking inputs. They draw on the same kinds of signals (sales, price, availability, delivery), so they tend to move with rank: correlation, not a documented boost.

New listings and the "honeymoon" question

Many sellers describe a "honeymoon period" in which new listings rank unusually well for a while. Amazon has never confirmed this.

Amazon's research shows why new listings behave differently. The 2022 cold-start paper describes the core problem: "no behavioral data causes poor ranking," which in turn means fewer chances to collect behavioral data. It says new items usually break through on very specific queries with small result sets, and only later appear on broader queries. The authors tested an exploration method that predicts engagement for new products from non-behavioral features, then updates as real feedback arrives. In a large A/B test, new-product impressions and purchases rose.

That documents a mechanism for giving new products temporary exposure. It does not prove Amazon.com runs this method today, or that any "honeymoon" lasts a fixed time. Treat early rank for a new listing as provisional.

Page layouts: why positions don't compare across keywords

Amazon renders search results in different layouts. Many queries show a multi-column grid, while some categories show a single-column list with larger tiles. The two layouts show different numbers of results on the first page, and ads are interleaved differently.

This comes from observing the pages, not from Amazon documentation. The consequence: "position 12" on a grid keyword and on a list keyword are not the same visibility, so averaging raw positions across layouts produces a number that means little.

Evidence at a glance

What this means for tracking

If behavior drives ranking, and behavior varies by keyword, by shopper, and over time, a single rank check tells you little. Three rules follow from the evidence above:

1. Measure across many keywords. Behavioral signals are collected per query, and new products surface first on specific queries. One keyword is a sample of one. Track head terms and the long-tail queries where movement often starts.
2. Measure the whole page, not just your own spot. Your position is relative. If a competitor wins the Featured Offer, cuts price, delivers faster, or earns a badge, you can drop without changing anything. Only the rest of the page tells those cases apart.
3. Measure over time, under fixed conditions. Behavioral signals accumulate, cold-start effects fade, and pages vary by address and time. Use one delivery address, a consistent capture window, organic positions separate from ads, and enough history to see a trend instead of noise.

How JustRanks approaches this

JustRanks records the whole organic first page of Amazon.com search results for more than 200,000 keywords over time, from a fixed US delivery address, overnight US time. Ads never count toward positions. Keywords saved in any user's market are captured every day, all others every 48h, so history usually already exists when you look a keyword up. For every product on the page it stores position, price, list price, rating, review count, "bought in past month," badges (Best Seller, Amazon's Choice, deals), and the delivery promise. Variants are grouped together, and no Seller Central access is needed.

Sources and further reading

• Amazon Help: Searching and Browsing for Items (sort options and Featured factors)
• Sorokina & Cantú-Paz, "Amazon Search: The Joy of Ranking Products," SIGIR 2016 (PDF)
• Han et al., "Addressing Cold Start in Product Search via Empirical Bayes," CIKM 2022 (Amazon Science)
• Nigam et al., "Semantic Product Search," KDD 2019 (arXiv)
• Reddy et al., "Shopping Queries Dataset: A Large-Scale ESCI Benchmark for Improving Product Search" (arXiv)
• Sell on Amazon: Amazon SEO, 7 ways to improve your product's search rankings
• Sell on Amazon: The Featured Offer (formerly Buy Box)
• Sell on Amazon: Guide to Best Sellers Rank (BSR)
• Amazon: Amazon's Choice FAQs
• Amazon Ads: Sponsored Products
• FTC: FTC Sues Amazon for Illegally Maintaining Monopoly Power (Sept. 26, 2023)