Section 01

A9 Is a Relevance Engine With a Memory, Not a Search Bar

The first thing to get straight is what A9 is trying to do. It is not trying to show the shopper the best product. It is trying to show the shopper the product most likely to sell from the exact words they typed. Those are different goals, and the gap between them is where most listings lose.

A9 is the legacy core of Amazon’s ranking system, and despite everything layered on top of it since, it is still the foundation. It reads your listing for relevance, which keyword you are indexed for, and it reads your history for proof, whether shoppers who searched that keyword actually bought your product. Relevance gets you into the running for a search. Proof decides your position in it.

The proof part is the piece operators underrate. Amazon remembers. When a shopper types a query and buys your listing, that purchase is attached to that query, and it raises how relevant Amazon thinks you are for it. This is why a brand new listing with a perfect title still starts near the bottom. It has relevance but no proof. The relevance is a claim. The sales are the evidence. A9 ranks the evidence, not the claim.

What to do this week:

• Stop treating your listing as a brochure and start treating it as a claim Amazon has to verify. Every keyword you place is a claim that you convert for it. If you do not have the sales to back the claim, the placement alone will not hold a top position.
• Separate your relevance problems from your proof problems. If you are not showing up for a search at all, that is relevance. If you show up but sit low, that is proof. They need completely different fixes.

Section 02

The Inputs Operators Can Actually Confirm

Here is where it pays to stay strict about your sources. There is a documented set of inputs A9 reads, and there is a much larger set of invented weightings floating around agency content. We only build on the first set.

The inputs Amazon’s system uses are well established in direction. Keyword indexation, whether the term is in your title, bullets, or backend at all. Click through rate from the search results page. Conversion rate on the listing. Sales velocity, both organic and paid. Price relative to your category. In stock status, because going out of stock tanks rank fast. And reviews, both count and average rating, with more recent reviews carrying more weight. On the listing itself, the title is the most heavily weighted field, bullets sit below it, and the backend Search Terms field is the catch net for everything relevance still needs but the front of the listing cannot fit.

What you will not find in that list is a number. Amazon has never said title is worth X and conversion is worth Y. When you see a post that assigns exact percentages to these inputs, treat it as one operator’s read, not a rule. The direction is real and you can act on it. The precise math is not published, and building a strategy on a fabricated weighting is how teams end up optimizing the wrong thing.

The practical version of this is simple. You cannot control the weights, so you control the inputs you know are real, in the order that a listing needs them. Get indexed. Earn the click. Win the conversion. Feed the sales velocity. Everything else is noise until those four are working.

What to do this week:

• Audit which of the confirmed inputs you are actually losing on. Pull your listing and grade it honestly on indexation, click through rate, conversion rate, and in stock status. One of them is usually the bottleneck.
• Ignore any advice built on a specific weight number. If a tactic only makes sense because someone claims conversion is thirty percent of the algorithm, the tactic is built on sand.

Section 03

A9 Did Not Retire, It Got Two New Layers on Top

You will hear operators talk about A10, and more recently about Cosmo and Amazon’s AI shopping assistant, as if each one replaced the last. That is not how it works. Amazon’s ranking system in 2026 is three layers stacked, and the classic A9 relevance and sales signals sit at the bottom of the stack still doing the heavy lifting.

The first layer is A9 and A10, the legacy core we have been describing. The second layer is Cosmo, a semantic retrieval system Amazon deployed in 2024 that maps what a shopper means to products using behavioral patterns, not just literal keyword matches. Cosmo is real and documented in Amazon Science. What is not real is the Cosmo weighting formula some agencies sell. Any specific weight claim about Cosmo is interpretation, exactly like the A9 weights. The third layer is Amazon’s AI shopping assistant, which in the US market Amazon merged into Alexa for Shopping in May 2026. It answers shopper questions and recommends products by reading your listing copy, your reviews, your questions and answers, and your A+ Content text.

What this stack changes for you is less dramatic than the headlines suggest. The AI layers read more of your listing than A9 ever did, which means your A+ Content text, your review language, and your buyer questions now feed into which searches you surface for. A product with reviews that mention great for sleep can surface for a sleep related search even if the word sleep is not in the title. That is a genuine shift and worth writing your content for. But none of it removes the floor. You still have to be indexed, you still have to convert, and you still have to sell. The new layers reward listings that already have the fundamentals, they do not rescue listings that do not.

The layers on top of A9 read more of your listing than ever before. They do not change the floor. You still get indexed, you still convert, you still sell, or none of the clever semantic matching matters.

Section 04

Why PPC and Ranking Are the Same Job

The most useful thing to understand about A9 is that it does not care whether a sale came from an ad or from organic search. A sale from a keyword is a sale from a keyword. When you run Sponsored Products on a target term and that ad produces a purchase, you are not just buying a click. You are handing A9 the exact proof it ranks on, that your listing converts for that search, and that signal feeds your organic position for the same term.

This is why we treat keyword optimization and PPC as one job, not two. The listing work makes the conversion possible, a clean title, images that sell, a price that competes. The ad spend makes the conversion happen at volume, fast enough to move rank while the launch window is still open. Optimize a listing in a vacuum, with no sales feeding it, and it sits indexed and placed perfectly and completely stuck. If you are new to how the paid side works, our explainer on what Amazon PPC is covers the ad types that feed this loop.

We saw this exact loop on a US beauty brand we worked with. The listing was already clean. The wins did not come from broad bidding across every keyword in the category. They came from finding the specific search terms the listing actually converted on, pushing spend behind those, and cutting spend on the terms that clicked but never bought. Concentrating spend on the terms that converted, rather than spraying it, cut that brand’s ACOS from 42% to 18% inside 90 days. The ranking lesson is the same as the efficiency lesson. The keyword you want to own organically is the keyword you fund with ads until the organic position holds on its own.

What to do this week:

• Match your top PPC targets to the exact terms you want to rank for organically. If you want to own a keyword, you should be funding it. Ranking is bought with proof, and PPC is how you buy proof fast.
• Cut spend on terms that click but do not convert. Every one of those is teaching A9 that your listing does not convert for that search, which is the opposite of what you want.

Section 05

The Reports That Tell You A9 Is Working

You do not measure ranking by how many keywords you stuffed into the listing. You measure it by movement on the terms that matter and by the one report that shows you the truth about conversion.

That report is Search Query Performance, inside Brand Analytics. It shows you, per search term, how many shoppers searched it, how many clicked your listing, and how many bought. It is the only place Amazon tells you your real conversion rate on a specific query, which is the input A9 ranks on. If a term shows strong impressions and clicks but weak purchase share, your listing is being shown and losing the sale, which is a conversion problem, an images and price and content problem, not a keyword problem. If a term barely shows impressions, that is an indexing or relevance problem. Same report, two completely different fixes, and Search Query Performance is how you tell them apart. The full field by field rebuild that acts on these signals lives in our 2026 listing optimization playbook, and the pre launch version is our Amazon listing optimization checklist.

Alongside Search Query Performance, track organic rank on your priority terms week over week, not month over month. When the conversion signal is flowing, rank moves in days, and a weekly cadence catches a slide while you can still respond to it. A monthly check tells you what happened after it is too late to react.

What to do this week:

• Pull your top ten terms in Search Query Performance and flag any with strong clicks and weak purchase share. Those are conversion fixes. Fixing keywords on those terms does nothing.
• Set a weekly rank check on your five priority keywords and watch direction, not just position. A term sliding from three to seven matters more than a term sitting steady at nine.

Action Checklist: Five Moves This Week

You cannot see Amazon’s ranking weights, and neither can the person selling you a chart of them. What you can see is which listing sells from which search. Optimize for that and the weights take care of themselves.