I have spent a large part of my career on the other side of this problem. Not selling on marketplaces, but running SEO for them: building and optimising search visibility for large marketplace platforms, including one operating across four markets with millions of monthly impressions and catalogues in the hundreds of thousands of listings. That vantage point changes how you read the Amazon SEO advice that circulates, because you have watched the same mechanics from inside the machine rather than inferring them from outside it.

What follows is the version I wish existed when I started: honest about the naming, clear about the mechanics, current on the AI layers that most guides have not caught up with, and explicit about which of your hard-won Google SEO instincts will help you here and which will quietly sabotage you.

First, the Naming Problem

Almost every Amazon SEO article opens by confidently naming the algorithm, and a good share of them are wrong. Here is the accurate position, because starting from a false premise makes everything downstream shakier than it needs to be.

A9 is real. It takes its name from A9.com, the Amazon subsidiary that built the product search technology, and for years it was the correct term for the ranking system behind Amazon’s search results. If you use it, you are not wrong, you are being slightly historical.

A10 is not real, in the sense that matters. It was coined by the seller and agency community to label a set of observed behavioural changes, chiefly that external traffic seemed to count for more and that raw sales volume seemed to count for less than conversion quality. Those observations may well be accurate. The name is not Amazon’s, has never appeared in Amazon documentation, and there is no A11 or A12 waiting in the wings. When a guide talks about the A10 update as though Amazon announced it, you are reading someone repeating industry folklore.

Meanwhile, the systems Amazon genuinely does publish about have unglamorous names that most articles skip: COSMO, an intent-understanding layer described in peer-reviewed research, and Rufus, the conversational assistant now live in front of shoppers. Those are where the interesting change actually is, and we will come to both.

The Mental Model: Why Marketplace Search Is Not Google

This is the section I would keep if I had to delete the rest, because almost every expensive mistake I have seen an SEO make on a marketplace traces back to importing Google assumptions wholesale.

Google is, in the loosest sense, trying to answer a question well. Its incentives point toward satisfying the searcher with the most useful, trustworthy document available across the entire web, which is why authority, links, expertise and content depth carry the weight they do.

A marketplace is trying to sell something. Its incentives point toward showing the listing most likely to convert this particular shopper into a completed order, from a catalogue it owns, where it takes a cut of every transaction. It does not care whether your listing is the most authoritative document on the topic. It cares whether people who see it buy it.

Once you internalise that, the ranking factors stop looking arbitrary. Conversion rate matters because conversion is the objective. Sales velocity matters because it predicts future conversion. Price and availability matter because an out-of-stock or overpriced listing wastes an impression. Reviews matter because they raise conversion. Every one of those signals is downstream of a single question the marketplace is always asking: if I show this, will it sell?

The Two-Stage Machine: Matching, Then Ranking

Underneath the naming debates, the architecture has been consistent and is well understood. Amazon’s search works in two stages, and confusing them is the source of endless wasted effort.

Stage One in Practice: Where a Marketplace Reads Your Words

If matching decides eligibility, then keyword coverage is an inventory problem: have you claimed every term a relevant shopper might use, somewhere the algorithm reads? Marketplaces give you more surfaces for this than Google does, including fields the shopper never sees.

The title carries the most weight and deserves the most thought: the primary term, the brand, and the distinguishing attributes a buyer filters on, written so a human can still read it. Bullets carry both keywords and persuasion, and should cover use cases and problems rather than restating specifications the attribute fields already hold. The description or enhanced content layer is where longer-tail and secondary phrasing lives.

The back-end fields are where the discipline shows. These exist precisely so you can claim terms that would make the visible listing worse: misspellings, synonyms, regional phrasings, alternate use cases, competitor-adjacent language where permitted. The single most common error I see is filling these with repetitions of terms already in the title, which adds nothing, when their entire value is covering ground the front end cannot.

And a point most guides underplay: reviews and customer questions are indexed text too. The words your buyers naturally use to describe the product become part of what the listing can match against, which is one more reason review generation is not merely a conversion tactic.

Stage Two in Practice: The Flywheel That Decides Everything

The ranking stage is where marketplace SEO becomes genuinely different from anything in organic search, because the primary signals are outcomes rather than attributes. This creates a self-reinforcing loop that explains both why established listings are hard to displace and why new ones struggle so badly.

A listing that ranks well receives impressions. Impressions produce sales. Sales produce conversion data and velocity, which are exactly the signals that justify ranking it well. The wheel turns, and each turn makes the next easier. This is why incumbency on a marketplace is so durable, and why the honest answer to how do I rank a brand-new listing is that you must generate the first turns of the wheel by other means, because organically you have no history to rank on.

That is the structural reason advertising and marketplace SEO are entangled in a way that Google ads and organic search are not. Paid placement buys impressions, impressions produce sales, and those sales feed the organic ranking signals. Sellers who treat the two as separate budgets consistently underperform those who treat early advertising as the ignition for an organic flywheel.

The AI Layer: COSMO and What It Changes

Here is where most Amazon SEO content is now out of date, and where the practical advice genuinely shifts. Amazon has published research on a system called COSMO, an intent-understanding layer that evaluates whether a product actually solves the problem expressed in a query, rather than merely containing its words.

The practical consequence is significant. Under pure keyword matching, a shopper searching for something to keep coffee hot on a long commute would only surface listings containing those literal words, which is why sellers historically crammed every possible phrasing into their listings. Under an intent layer, the system can reason that the problem involves insulation, portability and spill resistance, and surface a travel flask that never used the word commute anywhere.

So the optimisation advice inverts in an important way. Describing what your product is for, who it suits, and which problem it solves, in natural language, becomes a ranking asset rather than fluffy copy. Keyword coverage still governs eligibility and still matters, but the marginal value of the twentieth repetition of a term has collapsed, while the value of clearly articulating use cases has risen.

Rufus, the conversational assistant, sits alongside this as the layer shoppers actually interact with, answering questions and making recommendations in dialogue rather than returning a grid of results. For sellers, the implication is similar in direction: listings that clearly answer the questions a shopper would ask, in the language they would ask them, are better positioned to be surfaced and recommended by a system whose job is to explain and justify choices.

What Transfers From Google SEO, and What Actively Misleads

For anyone arriving here from organic search, this is the section that saves the most wasted effort. Some of your instincts are directly valuable. Others are worse than useless, because they feel productive.

Your keyword research discipline transfers almost intact, though the tools change: marketplace keyword data comes from marketplace-specific sources rather than general search tools, because the query language differs sharply from Google. People search marketplaces in shorter, more product-shaped, more transactional phrasing.

Your intent instinct transfers and arguably matters more, especially now. Understanding that a query implies a problem rather than a string is precisely the muscle COSMO rewards.

Your structured data instinct transfers in disguise. Attribute fields are the marketplace equivalent of schema markup: they are how you tell the system unambiguously what a product is, and completing them thoroughly is one of the highest-return, least-glamorous tasks available.

What does not transfer is the entire authority apparatus. There are no backlinks to your listing that build domain authority in the Google sense, because you are ranking inside someone else’s domain. External traffic to a listing appears to carry some weight, and the community has argued for years about how much, but it is not a link graph and treating it like one leads to spending money on the wrong things.

Content depth for its own sake also fails to transfer. On Google, a 3,000-word guide can outrank a 600-word one on quality and comprehensiveness. On a marketplace, a beautifully written listing that converts at two percent will be outranked by a plainer one that converts at eight, indefinitely. The document does not win; the outcome does.

The Myths Worth Discarding

A Working Checklist

If you want the operational version of everything above, this is the order I would work in, and the reasoning behind the order is the two-stage model: eligibility first, because nothing else matters if you are not in the candidate set, then conversion, because that is what the ranking stage actually rewards.

The Honest Summary

Amazon search is a commercial machine wearing the costume of a search engine. It is called A9, the A10 you have read about is folklore, and the genuinely current story is an AI layer that reads queries as problems rather than strings. The two-stage structure has not changed: words make you eligible, performance decides whether you win, and the flywheel between ranking and sales explains almost every counterintuitive thing you will observe.

If you come from organic search, keep your keyword discipline, your intent thinking, and your instinct for structured completeness. Leave behind the authority mindset and the belief that a better document wins. And if you take one operational idea, take this: on a marketplace, improving conversion rate is simultaneously a conversion tactic and a ranking tactic, which makes it the single most efficient work available to you. That dual payoff has no real equivalent in Google SEO, and it is why the sellers who quietly win tend to be the ones obsessing over their images and their first two lines rather than their keyword density.

External Traffic: The Signal Everyone Argues About

No topic in marketplace SEO generates more confident disagreement than external traffic, meaning visitors you send to a listing from outside the marketplace: social, email, your own site, influencers, paid media. The community consensus, and the observation behind the A10 folklore, is that it carries real weight. Amazon has never published a coefficient, so anyone quoting you a precise percentage is estimating.

What is defensible is the mechanism rather than the multiplier. External traffic that converts does two things the algorithm demonstrably cares about: it produces sales, feeding velocity, and it improves the listing’s conversion record if those visitors were well qualified. That alone explains most of the observed effect without needing to assume a special external-traffic bonus. It also explains the failure mode, which is sending large volumes of poorly qualified traffic to a listing and watching rankings fall rather than rise, because you have just taught the system that people who see this listing do not buy it.

The practical rule I would give: send external traffic when it is genuinely likely to convert, and treat it as a way to seed the flywheel rather than as a ranking lever you can pull independently of commercial reality. Sending traffic that browses and leaves is worse than sending none.

Multi-Marketplace and International: Where It Gets Genuinely Hard

This is the part I know best from the platform side, because running search for a marketplace operating across four markets teaches you that the naive assumption, that a listing which works in one market will work in another, is almost always wrong. Query language differs, not just in translation but in structure: what shoppers call a product, which attributes they filter on, and which phrasing signals buying intent all vary by market.

Sellers expanding across marketplaces routinely translate a listing and expect the ranking to follow, then find themselves invisible. The listing is eligible for the translated terms and for nothing else, while local shoppers use words the translation never produced. The fix is the same as it is in international SEO on the open web: research each market natively rather than translating a keyword list, and accept that the winning listing in one market may be structured differently from the winning listing in another.

The other cross-market lesson is that the flywheel does not travel. Sales history, reviews and velocity are largely market-specific, so a category leader in one marketplace begins near zero in the next. Expansion plans that assume inherited authority consistently underestimate the ramp, which is the marketplace equivalent of a mistake I have seen many times in multi-market SEO on the open web.

How to Measure Marketplace SEO Honestly

Measurement on a marketplace is unusually clean in one respect and unusually treacherous in another. Clean, because the ranking signal and the business result are literally the same number: conversion. Treacherous, because paid and organic placement blend into a single sales figure that flatters whichever story you want to tell.

Separate organic from advertised placement before drawing any conclusion, because the entanglement that makes advertising useful for igniting the flywheel also makes blended reporting meaningless. Track rank position for a defined set of terms over time rather than spot-checking, since marketplace positions fluctuate far more than Google positions and a single check tells you almost nothing. Watch conversion rate per listing as the leading indicator it genuinely is, and watch velocity as the thing that decays when attention lapses.

The vanity trap here mirrors the one in e-commerce SEO reporting: impressions and keyword counts feel like progress and can rise while sales stall. On a marketplace the discipline is simpler than on the open web, because there is one number that is simultaneously the ranking signal, the revenue driver and the honest scorecard. Optimise conversion and almost everything else follows.

Where This Is Heading

Two directions look clear enough to plan around. The first is that intent understanding keeps deepening. Every year the gap narrows between what a shopper means and what they type, which steadily reduces the return on mechanical keyword coverage and raises the return on describing a product honestly and specifically. Sellers who have built their advantage on cramming term variants into back-end fields are standing on ground that is slowly eroding, while those who have built genuinely clear, well-attributed, problem-oriented listings are standing on ground that keeps appreciating.

The second is that conversational and assistant-led discovery grows. When a shopper asks an assistant to recommend something for a specific situation, the winning listing is the one the assistant can confidently justify: complete attributes, clear use cases, credible reviews, unambiguous specifications. That is a different optimisation target from a keyword-matched results grid, and it rewards the same qualities that make a listing genuinely good rather than merely well-gamed.

Both trends point the same way, which is unusual and worth noticing. For most of the history of marketplace SEO, the gap between what helped the algorithm and what helped the shopper was wide enough to be exploited profitably. That gap is narrowing. The listing that a modern marketplace wants to rank is increasingly the listing a human would genuinely choose, which is the most encouraging development in this field in years, and the least convenient one for anyone whose method depends on the gap.