Amazon's ranking algorithm does not care about your brand story or how nice your gallery looks. Many factors feed organic rank, but the primary one is how likely your product is to produce a sale for the shopper who typed this particular query. That likelihood is what we measure as purchase rate: of everyone who sees you on this search, how many buy. Amazon does not publish the mechanism. What follows is the operating model this site runs on, and it earns its place by predicting what actually moves rank, not by claiming to read Amazon's code.
Two things follow. First, purchase rate is keyword-specific. The same ASIN can sit at #2 for one query and #47 for another, because it earns a different purchase rate on each. Amazon sells a mountain of toilet paper, and none of it appears for “usb c cable,” because its purchase rate on that query is zero. Second, purchase rate bakes the click and the conversion into one signal. Clicks without purchases do not count. Which means the whole game reduces to two moments you can engineer: getting chosen on the shelf, and closing the sale on the page.
- 01The number Amazon actually ranks you by: purchase rate, per keyword, with CTR and CVR baked into one signal.
- 02The complete input inventory: the seven elements of the search tile and the six factor groups of the page, including every gallery slide's job.
- 03Why the unit of analysis is the shelf, not your listing: you are never chosen in a vacuum.
- 04Conventions vs. white space: high adoption among winners is an assignment; absence is an opening.
- 05Levers vs. context: what you can change tomorrow, and what you have to strategize around instead.
- 06The overhaul-then-test split that resolves the "you changed everything at once" problem.
- 07How to measure creative honestly, at any traffic level, without waiting for Amazon's permission.
Two moments, one number
The click is won on the search shelf, at thumbnail size. The purchase is won on the product page. Both are merchandising decisions, and Amazon grades you on the product of the two.
Most of the buying decision happens earlier than brands think: in the search row, at thumbnail size, before price or reviews get a full vote. So before any method, know the inputs. Everything that moves purchase rate lives in one of the two moments, and each moment has a complete, finite anatomy.
Everything the tile says before they tap
On the search shelf your entire listing collapses into one tile, and every element on it is either winning or leaking the click:
Whey
3Brand Whey Protein Powder, 25g Protein, Chocolate, Low Sugar, 2 lb
★★★★☆4.64(1,238)5
$29.99($0.94 / Ounce)6
prime· Get it tomorrow7
- 1Main image: the entire first impression at thumbnail size: pack large, hero number readable, contrast against Amazon's white
- 2Badge: borrowed authority (Best Seller, Amazon's Choice, deal flags): you influence them, you don't choose them
- 3Title: front-load the words they searched; the first ~70 characters do almost all the work
- 4Star average: judged relative to the row: a 4.3 sitting next to 4.7s leaks clicks even though 4.3 is "good"
- 5Rating count: order-of-magnitude social proof: 40 vs 5,000 changes which strategy you can run
- 6Price + unit price: read against the neighbors, not in absolute; unit price is the quiet tie-breaker on size-varied shelves
- 7Delivery promise: tomorrow beats Thursday; the Prime badge is table stakes, the date is the differentiator
Click factors are relative. Your star average, your price, your image size and contrast are all judged against the tiles beside yours, never in isolation. The same 4.3 stars is a strength on one shelf and a leak on another. This is why every diagnosis starts with the row, not the listing.
Of the seven, main image and title carry the most upside for most brands, for a simple reason: they are the two you can change tomorrow. Stars, rating count, and badges are earned over months; price is a P&L decision; delivery is an operations decision. The image and the title are pure merchandising, and they are usually the furthest from the shelf's standard.
Everything the page must close
Win the click and the product page inherits a skeptic holding a thumb over the back button. Conversion resolves into six groups of factors, and one broken group caps all the others:
- Content qualitythe gallery, bullets, A+ content, video, answered questions: the selling you control most directly
- Price presentationprice vs. the category's expectation, unit price, the visual of a deal (strikethrough, coupon badge), Subscribe & Save
- Social proofstar average (absolute here, not relative), rating count and recency, what the top reviews actually say
- Availability & fulfillmentin stock, Prime, delivery speed, and the Buy Box, whose loss is a severe penalty
- Variation experiencethe right default variant, a clear picker, the preferred size and flavor in stock
- Competitive presencethe compare widget and competitor ads sitting on your own page, ready to poach the sale
Within content quality, the gallery is where most listings leave the most on the table, because most brands treat it as a photo album instead of a sales sequence. Every slide has a distinct job:
This inventory is the same tree the retail formula decomposes in full. What this article adds is the method for acting on it: which of these factors to change, in what order, and how to know it worked. That method starts with a shift in what you look at.
You are never chosen in a vacuum
Here is the mistake underneath most weak listings: they were designed in a brand deck, not against a shelf. The shopper never sees your listing alone. They see it in a row, next to eight competitors, and every choice is relative. So the work starts with reading the live shelf for your keyword: decompose the winning listings into their working parts. The hero stat and where it sits. The badges and certifications. The claim language. The staged props and texture cues. The gallery's slide types: ingredient spotlight, proof, before-and-after, how-to.
Held against that shelf, a listing's diagnosis stops being a matter of taste and becomes concrete: no hero number anywhere on the tile. The benefit invisible at thumbnail size. The product rendered 30% smaller than the row's winners. Low contrast against Amazon's white. A gallery with six angles of the bottle and zero proof. Each of those is a specific, fixable reason the click goes elsewhere.
Conventions and white space
Once the shelf is decomposed, every element sorts into one of two piles by a single question: how many of the winners do it?
- Pure white background, product largeconvention: match it10/10everyone does it: table stakes
- Hero stat readable at thumbnailconvention: match it8/10the winners do it: match it
- Trust badge or certificationconvention: match it6/10common enough to be expected
- Staged props / texture cueslever: differentiate2/10rare: doing it well stands out
- Clinical-proof gallery slidelever: differentiate0/10absent: white space to own
High adoption is an assignment. If nine of the top ten print a hero stat on the tile, that is not a trend to consider; it is table stakes the shelf already validated, and you match it just to compete. Low or zero adoption is the opposite: an opening. Never read “the winners don't do X” as “don't do X.” Absence means nobody has claimed that move, and doing it well is how a newcomer stands out on a shelf where everyone else has more reviews.
Levers and context
Not everything on the shelf is a creative decision. Main image, title, gallery, bullets, and A+ are levers: you can change them this week. Price, review count, and fulfillment speed are context: they shape the outcome, but they are not merchandising choices. The discipline is to optimize the levers while strategizing around the context, not to pretend the context away.
The context that changes strategy most is reviews. A 40-review ASIN running a parity strategy against a 5,000-review incumbent loses by default: match the leader move for move and the shopper breaks the tie on social proof every time. Low-review listings need a differentiation strategy, not a conformity strategy. That is exactly what the white-space column is for.
Overhaul first, then test
The classic objection to rebuilding a listing is attribution: “if we change everything at once, we won't know what worked.” The objection dissolves once you split the work into its two actual jobs.
No test needed. The shelf already ran that experiment for you, at a scale you could never afford. Ship it as one move.
Genuinely unknown territory: your differentiators, the white space. This is where measurement earns its keep.
Closing a gap to a proven shelf norm does not need your A/B test, because the shelf already ran it. Ten winners printing a hero stat is an experiment with more traffic behind it than you will see in a year; treat the result as settled and ship the whole overhaul as one move. Save the measurement budget for what is genuinely unknown: the differentiators, the white-space moves, the things no winner has tried. One variable at a time, where a control still means something.
Prioritize from data, not a mood board
When the diagnosis produces fifteen possible changes, the order is not a matter of opinion. Score each one on three things and work the list:
- Impacthow big is the gap, weighted by the demand on the keywords where it shows? A missing hero stat on your highest-volume shelf outranks a weak seventh slide.
- Confidencewhat share of the page-1 winners run this play? Adoption is evidence. A move nine winners make is near-certain; a hunch is a hunch.
- Easewhat does the fix actually cost? A text overlay swap ships today; new photography does not. Cheap and fast moves up the queue.
Notice what this replaces: the agency conversation about which concept “feels stronger.” Every score above comes from the shelf data and the production reality, which means two people running the method independently land on roughly the same queue.
Measure like an operator
Creative that is not measured is decoration with better justification. There are three ways to get an honest read, in order of preference:
- Amazon's own experiments (MYE)the cleanest split test, when you qualify: Brand Registry plus a traffic threshold. Use it whenever it is available.
- A single-keyword top-of-search campaignthe closest thing to a purchase-rate instrument you can build yourself. The setup and its limits are below.
- The shelf as control groupfor sequential before/after reads: if the rest of the shelf held flat over the window and you rose, the change is attributable. If the whole shelf rose, it was the season, not your image.
The second one deserves the detail, because Amazon never reports purchase rate and this is one of the few ways to watch your own. Build the campaign so that only one thing can vary:
Run that campaign continuously and creative tests become readable. Change the main image and nothing else, and the CTR line on that one keyword answers you directly. It costs ad spend, but it is spend on traffic you wanted anyway, which is why this is the method most operators can actually afford to keep running.
And the accounting for losses stays honest: a lost test costs nothing but the traffic it ran on, because the control stays. That is the same staircase logic as the ASIN lifecycle's mature phase: wins are permanent steps, losses are flat, and the floor never goes down.
Purchase rate is how rank is made
This is the part that turns creative work from a cost center into the flywheel's crank. Because Amazon ranks by estimated purchase rate, every point of CTR or CVR you win does two jobs: it converts more of today's traffic, and it buys better rank, which brings more traffic tomorrow. The same mechanism runs in reverse: a stockout takes your purchase rate to zero, torches the recency-weighted velocity behind your rank, and typically costs two to four weeks of recovery after restock. The flywheel does not care which direction it spins.
What is a rebuilt main image worth? In prior testing on a brand managed by the people behind this site, a main-image rebuild moved conversion from 0.58% to 1.99% across eight tests. That is a past result on one product, not a promise, and it is exactly why the method insists on measurement: the shelf will tell you what your version of that number is.
Common questions
What is purchase rate on Amazon?
Of everyone who sees your product on a given search, how many buy it. Orders over impressions, per keyword. It bakes the click and the conversion into one number, which is why clicks that do not end in purchases do not help you: the whole game reduces to getting chosen on the shelf and closing on the page.
Does Amazon rank by purchase rate?
Many factors contribute to organic rank. The primary one is how likely your product is to produce a sale for the shopper who typed that particular query, and purchase rate is how we measure that likelihood from the outside. Amazon does not publish the mechanism, so treat this as the operating model behind the method here rather than a description of Amazon's code.
Why is purchase rate keyword-specific?
Because the same ASIN earns a different purchase rate on every query. A product can sit at position two on one term and position forty-seven on another: same listing, different shoppers, different intent, different likelihood of a sale. It is why rank work is done per shelf rather than per listing.
How do you improve purchase rate?
Read the shelf that will judge the listing, match the conventions its winners share, take the white space they have all left open, then overhaul rather than tweak, and test. Conventions are table stakes and their absence costs you; white space is where differentiation is still available. The procedure is the shelf-read method.
Is purchase rate the same as conversion rate?
No. Conversion rate is orders over sessions: it starts counting once someone has already clicked. Purchase rate starts at the impression, so it also prices in whether your search tile earned the click at all. A listing can convert beautifully and still lose the shelf by never being chosen.
The method is decoded from live shelf decompositions in sports nutrition and skincare, applied to real listings. The search-tile and gallery mockups are illustrative wireframes, not real listings, and the adoption figures in the shelf-read board are archetypes, not a specific category; every shelf answers differently, which is why you read yours. The 0.58% → 1.99% conversion figure is prior testing on a single brand, stated as history, not expectation.
Next in the fundamentals: The Shelf-Read Method → (the procedure for engineering the number this page just defined). Or go wider: Amazon Copywriting: The Complete Guide → and the method applied in The Rebuild →
Want your shelf read and your listing graded against its actual winners? → Work with Us