Understanding Amazon ranking: retrieval, ordering and what Amazon actually documents
- You know the two stages of Amazon search and which levers act on which stage.
- You recognize invented ranking numbers (“A10”, “30 % conversion weight”) and know what Amazon has actually published.
- You check indexing with a word list and the ASIN test and keep a log that shows problems BEFORE the sales dip.
- You measure your position within the organic results and test whether your ads replace sales instead of adding them.
Almost every guide explains “the Amazon algorithm” to you — with percentages, factor lists and a name like A9 or A10. The problem: Amazon has published none of it. This lesson cleans up. You learn what is documented (search runs in two stages), what is guesswork (every weighting) and how to find out with a word list, the ASIN test and a log at which stage your listing is currently failing. Once you know the split, you stop spending weeks on the title while the problem is the price — or the other way round.
1What Amazon publishes — and what it doesn't
Amazon publishes no documentation of its search and discovery algorithms. The clearest publicly readable statement on this comes from a community manager in the official seller forum — a forum post, not a help page, but unambiguous. What is documented are the fields and their rules, not their weighting. Also documented: Amazon describes its own product search in research publications as an interplay of several methods for matching and ranking — not as a single, versioned algorithm.
| Claim | Evidence | What you do with it |
|---|---|---|
| “Amazon ranks by A9 / A10” | A9.com was an Amazon subsidiary for search technology (press release 2004). There is no Amazon source for “A10” — the name only appears at service providers. | Call observations observations, not algorithm rules. |
| “Conversion rate makes up 30 % of the ranking” | Amazon publishes no weightings. None of these percentages traces back to a named study. | Treat every weighting percentage as an estimate. |
| “Search runs in two stages: retrieval, then ordering” | Covered by Amazon's own research publication (SIGIR 2016, “Amazon Search: The Joy of Ranking Products” — as of 2016, no statement about today's setup). | The working model of this lesson. |
| “External traffic is rewarded directly” | No Amazon document names it as a ranking factor; only the indirect path via sessions and sales is plausible. | External only with an attribution link and only if the page converts (Growth, lesson 8). |
Amazon search is a black box with two doors. At the first door stands a bouncer with a word list: if you're not on it, you don't get in. Behind the door the audience decides who stands up front. Anyone who tells you they know the percentages used to weigh the bouncer and the audience has guessed them.
You read in an agency blog: “Conversion rate is the most important ranking factor at 30 %.” Check in three steps: (1) Source in the post? None — “Amazon's internal data” without a link. (2) Is there an Amazon publication on weightings? No, Amazon explicitly says there is none. (3) Does a named study give the number? No. Result: the 30 % is an estimate copied from blog to blog. What remains is the direction: conversion rate acts on the ordering. That's what you work with — without the invented decimal.
2Stage 1: retrieval — is the word in your data?
First Amazon builds the set of all catalog products that match the query. Your catalog data counts for this: title, Item Highlights, bullet points, backend search terms, attributes and the category assignment. Three rules from Amazon's own guidance define what “match” means:
- No partial matching: Amazon's search glossary says search cannot perform partial matches. A term must appear as a whole word somewhere — a fragment isn't enough. For German compound nouns that means “Gewürzmühlenset” doesn't reliably cover “gewürzmühle set”.
- Amazon recognizes word forms itself: singular or plural is enough, you don't need both. Entering both only costs bytes.
- Amazon does NOT recognize synonyms automatically: Amazon says you should enter those yourself. “Bath towel” won't find you if only “towel” is in your data. How far search maps semantically beyond that is undocumented — don't rely on it.
So the practical question is not “How often is my main keyword in there?” but “Which term am I still missing somewhere?” The word-list check in five steps:
- Copy every field into a text file: title, highlights, bullets, backend, visible attributes.
- Put your keyword map next to it (from Getting started, lesson 6).
- For every term check: does every word of the query appear at least once as a whole word?
- Add missing words where they fit naturally — otherwise in the backend.
- Wait 48 hours, then run the ASIN test (next chapter).
Repeating a keyword across several fields to “reinforce” it. Amazon writes on its seller page: a word combination you've used once you usually don't need a second time — work with variants. A density multiplier like in web SEO is documented nowhere. The repetition adds no coverage, it uses up space you'll miss for a synonym.
A customer searches “towel xxl cotton”. Title, bullets and search terms everywhere say only “Towel XXL” — the word “cotton” appears nowhere because “you can see it in the photo”. For stage 1 you're missing a third of the query. Fix: “100 % cotton” into the Item Highlights, and “cotton” is in the catalog once. Whether you write “towel” or “towels” doesn't matter — Amazon recognizes both forms. What Amazon doesn't recognize: that “bath sheet” means the same. You enter that in the backend yourself.
3Stage 2: ordering — what buyers do with your result
Once your product is in the set, Amazon sorts. This order is driven mainly by how customers react to your offer — how strongly, nobody documents. What can be observed and is consistent with Amazon's statements:
- Clicks on your result: title, main image, price and star rating decide in the search tile. That's why you test title and main image by sessions, not by sales (lesson 1, expert chapter).
- Purchases per session: bullets, gallery, A+, price and reviews on the product page. That's the metric you use to judge product-page changes.
- Price relative to the field: not your price in absolute terms but relative to the offers next to you — as the landed price including shipping.
- Availability: a sold-out offer breaks its sales history, and in many sellers' experience recovery takes longer than the gap.
- Separately per search term: sales via “spice grinder set” improve your position for exactly that term. How to use that deliberately is in Growth, lesson 3.
One caveat belongs here: in two-stage search systems, text-relevance features regularly feed into the ordering as well. How strongly that is the case at Amazon is unknown. That's why the defensible version stands here: copy decides whether you get to play; the order is driven mainly by buyer behavior.
Two spice grinder sets are both in the index for “spice grinder ceramic” — stage 1 is done for both. Set A has a main image with both grinders cleanly cut out, 4.4 stars from 310 reviews and €24.99 with Prime. Set B shows a dark photo, 3.9 stars from 41 reviews and €22.90 plus €3.99 shipping. On the same position, set A gets clicked more often and bought more often — and drifts upward over time, although B's title is denser in keywords. Anyone now tweaking B's title is working on the stage that was passed long ago.
4Checking indexing: the ASIN test and a log
Whether a keyword is really in the index is answered not by a statistic but by search itself: ASIN and keyword together in the search bar, for example “B0XYZ12345 ceramic grinder”. The ASIN is an indexed term itself and narrows the results to your product. If it appears, you're in the index for both terms. Amazon documents the test nowhere — it's practitioner knowledge, and it measures pure indexing without competing against anyone.
| Rule | How you do it |
|---|---|
| When | After every listing change, the full list once per quarter — and immediately when Amazon has shortened or rewritten a title: keywords have dropped out then, and only the test tells you which. |
| What | The 10 to 20 most important terms per ASIN. Ten of them go into the weekly walk-through from lesson 11, the full list into the quarter. |
| Limits | The test doesn't say which field the word sits in. It's unreliable for very generic single words and for German compound nouns. Index-checker tools also give a yes/no answer — the search-bar test is free. |
| Log | One table per ASIN: keyword · date · ASIN test (yes/no) · field it sits in · note. Five columns that show you in three months when a term disappeared. |
With Brand Registry you add the Search Query Performance report — impressions, clicks, cart adds and purchases per query, each also as your share. It narrows the diagnosis without proving it: no impressions can be an indexing problem OR a position so far back that nobody sees you. About position the report says nothing. Only the ASIN test separates the two. How to read the report line by line is in the expert chapter of lesson 8 and in Growth, lesson 4.
Amazon gradually rewrites titles over 75 characters to its AI recommendation. Whether your ranking term disappeared in the process you check not on the title but in the index: five main keywords via the ASIN test. If the word still sits in a bullet or the backend, the index is often retained — then the new title is a click problem, not a retrieval problem, and you rewrite it at your leisure.
5Page 1 is not page 1: subtracting the ad slots
Part of page 1 is ad space and cannot be reached organically. How big that share is, is not an algorithmic constant but a business decision by Amazon. In its lawsuit of September 26, 2023, the US Federal Trade Commission together with 17 states additionally accuses Amazon of deliberately showing worse-matching ads; Amazon disputes it, these are allegations and they concern the US market. For you, regardless of the outcome: measure your position within the organic results, not by the slot on the screen.
- How to count: open the results page in an incognito window, skip every tile marked “Sponsored”, number the rest. Your number is the organic position.
- How to log: always the same time of day, the same device (phone!), the same marketplace — results pages differ by device and location.
- How to read movement: a jump from organic 12 to 6 is a success even if you still stand “on tile 15” on the screen because Amazon inserted more ads.
The opposite direction is the cannibalization test: if you rank in the organic top 3 for a keyword and still bid on it via PPC, lower the bids there for two to three weeks as a test and track the ASIN's total sales. If it stays almost the same, the paid click has mostly replaced a free one. That's no proof — season, competitor prices and stock change in the same period. Hence: note the confounders and repeat the test in a second period before changing course. The trade articles clearly contradict each other here; there is no Amazon source on the relationship.
For “spice grinder ceramic” the results page on the phone shows 48 tiles. 14 of them carry the word “Sponsored” — including tiles 1 to 4 and a block in the middle. Your AURELO set sits on tile 9. Two of the eight tiles before it are ads: your organic position is 7, not 9. Four weeks later you're on tile 9 — but now six of the eight tiles before you are ads: organic position 3. Anyone counting only the tile would have logged “no change” and rolled back the title change that brought the jump.
6The ranking diagnosis in ten minutes
“I don't rank for X” has five possible causes, and they sit on different stages. This order keeps you from working on the wrong one:
- Is the word in my data? Word-list check — as a whole word, at least once.
- Am I indexed? ASIN test. No → add the word, wait 48 hours, repeat. Yes → continue.
- Am I seen? Search Query Performance (with a brand) or position count in an incognito window. No impressions despite the index → position too far back: stage 2.
- Am I clicked? Click-through rate relative to position. Weak → main image, title, price, stars (lesson 5).
- Do people buy? Purchases per session. Weak → product page (lesson 6, 7) or the Featured Offer (lesson 14).
- Am I in stock and in price? Check inventory and landed price relative to the field before changing anything in the copy.
First check whether you're in the race at all (index), then whether anyone sees you (position), then whether anyone clicks (tile), then whether anyone buys (page). Every stage has its own tools — and a stage-4 tool fixes nothing on stage 1.
7Expert insight: click strength relative to position — Search Query Performance as a normalized click-through rate
An absolute click-through rate says nothing without the position: 3 percent on position 30 is excellent, 3 percent on position 2 is an alarm. Amazon doesn't show you the position — but the Search Query Performance report gives you, per query, your share of impressions and your share of clicks. The quotient of the two is a position-adjusted approximation of your click strength, because both shares refer to the same result set in the same period:
Click strength = click share ÷ impression share. Above 1 means: your tile gets more clicks than its visibility share would suggest — it wins against the neighbors. Below 1 means: you're seen but skipped. The same works one stage on: purchase strength = purchase share ÷ click share. This derivation is practice, not an Amazon rule — the report itself only gives the shares.
AURELO spice grinder set, report for four weeks. Query “spice grinder ceramic”: impression share 4.0 %, click share 6.2 %, purchase share 6.5 %. Click strength 6.2 ÷ 4.0 = 1.55 — your tile clearly wins; purchase strength 6.5 ÷ 6.2 = 1.05 — the page delivers what the tile promises. Diagnosis: the listing is good, you're just seen too rarely → push stage 2 with sales via exactly this term (Growth, lesson 3). Query “pepper mill set”: impression share 9.1 %, click share 5.0 %, purchase share 4.8 %. Click strength 0.55 — you're seen often and clicked rarely. Diagnosis: tile problem. For this query “pepper mill” isn't in the title, the main image shows both grinders the same size, and the price sits above the field. Check title and main image for this query first — not more advertising.
Three rules so the quotient doesn't lie:
- Same query, same period: shares of different queries aren't comparable — the result sets differ. Always compare per query over time.
- Enough volume: below a few hundred of your own impressions per query and period the quotient is noise. Bundle weekly values.
- Think the ads out: the report mixes organic and paid visibility. A campaign launch raises the impression share and depresses click strength — that's not a listing problem. Write campaign changes into the log.
Put your 20 most important queries on a grid: click strength horizontally, purchase strength vertically. Top right (both above 1) are the terms where only visibility is missing — advertising pays there. Bottom right (click strong, purchase weak) are product-page construction sites. Left (click weak) are tile construction sites — title and main image per query. One grid, four actions, no guessing.
The first lessons of every track are open to everyone. From here on you just need a free account — no subscription, no costs.
- 2Stage 1: retrieval — is the word in your data?
- 3Stage 2: ordering — what buyers do with your result
- 4Checking indexing: the ASIN test and a log
- 5Page 1 is not page 1: subtracting the ad slots
- 6The ranking diagnosis in ten minutes
- 7Expert insight: click strength relative to position — Search Query Performance as a normalized click-through rate
- ✓Quiz: 6 questions with instant feedback
- All 41 lessons in 3 tracks — completely free
- Your progress is saved across devices
- Certificate + 100 free Listimo credits per completed track
Check yourself
6 quick questions — one at a time, instant feedback. With a free account your progress is saved.
Frequently asked
Is there evidence that more text relevance also improves the ordering?
In two-stage search systems, text-relevance features regularly feed into the ordering as well — how strongly that's the case at Amazon is unknown. Practically: copy secures participation for sure, and it may influence the order. What demonstrably moves the order is buyer behavior: clicks, purchases, price, availability.
How long until a newly entered word is in the index?
Amazon gives no public deadline. The circulating 72 hours come from a help page behind the login and couldn't be verified. In practice you allow one to three days and then check with the ASIN test — not after six hours, and without making further changes in between.
Does my brand name count as a keyword?
Yes. The brand name sits in the “Brand” field and is captured for search there — which is why Amazon explicitly advises against repeating it in the backend search terms. The same applies to third-party brands: they belong neither in the backend nor in the title, that's a policy violation.
Check your listing for free against the rules from this lesson — or have Listimo build the whole listing from one product photo: images, copy and A+ content.
Go deeper: the complete guide to optimizing Amazon listings →
Everything in this academy comes from day-to-day selling practice — the same playbook behind Listimo, the tool that turns product photos into complete Amazon listings.