A9

A9 is the name of Amazon's product-search ranking algorithm — the engine behind every results page a shopper sees when they type a query into the Amazon search bar. It was originally developed by A9.com, a Palo Alto subsidiary Amazon spun up in 2003 specifically to own product discovery, and although Amazon has consolidated those teams over the years, the algorithm itself still answers to that name internally and across the seller ecosystem.

A9 decides two things for every query:

1. Which ASINs are eligible to appear — the candidate set is everything indexed for at least one token in the query.
2. In what order they rank — the candidates are scored, sorted, and sliced into the page-1, page-2, and rest-of-search slots a shopper actually scrolls through.

The output is the SERP: a mix of organic results, sponsored placements, and various badges (Amazon's Choice, Best Seller, Prime). Organic rank on that SERP is the direct deliverable of A9; everything else is downstream of where A9 places the ASIN.

What A9 optimises for

Unlike Google's web search, which optimises for answering the question, A9 optimises for revenue per shopper session. The optimisation target is unambiguous: of all the listings that could plausibly satisfy this query, which one will Amazon make the most money from if it shows it in the top slot? That single objective rolls up dozens of contributing signals, but every signal is in service of that question.

This is why A9 behaves so differently from Google. It does not care about backlinks, domain authority, or external citations. It cares about whether shoppers who see this listing click it, whether clickers add to cart, whether cart-adds become orders, and whether orders stay orders (returns reverse the signal). The faster a listing converts a query into revenue, the higher A9 ranks it for that query — and conversely, a listing that gets impressions but no conversions is demoted, often within days.

The two layers A9 blends

Every ranking decision A9 makes is a weighted blend of two layers:

Relevance — can this listing answer the query?

• Indexing. Does the listing contain the search tokens at all, in any indexed field (title, bullets, backend search terms, A+ alt text, brand, product name, category attributes)? A keyword that appears nowhere on the listing is invisible to A9 for that query — no rank, no impressions, no clicks. Indexing is binary and is the most fixable layer.
• Field weighting. A keyword in the title carries far more weight than the same keyword in the bullets, which in turn weighs more than the description or backend search terms. Early position in the field matters too — the first 60–80 characters of the title carry the strongest signal.
• Category match. A listing in the dominant browse node for a query out-ranks an equivalent listing in a wrong category. Category is itself a relevance signal.

Performance — does this listing convert the query?

• Click-through rate from the SERP tile (image, title, price, star rating, review count).
• Conversion rate on the detail page — sessions that end in an order divided by total sessions.
• Sales velocity — units per day, especially in the trailing 7 and 30 days.
• Review count and average rating — both an input to A9 and an input to shopper CTR.
• Buy Box ownership, Prime eligibility, FBA status, and price competitiveness vs the SERP cohort.
• Return rate and account-health metrics — late shipments, defects, and high returns demote the listing.

The two layers compound. A perfectly indexed listing with weak CVR drops; a marginally indexed listing with stellar CVR climbs past category leaders. Most serious Amazon SEO work is about feeding both layers in parallel.

A9 and "A10" — there is no A10

For roughly the last five years, parts of the seller community have called the modern algorithm "A10," framing it as a successor that re-weights toward organic off-Amazon traffic and devalues keyword stuffing. Amazon has never released or named an algorithm called A10. What people are describing as "A10" is the normal, ongoing re-tuning of A9 — the same way Google has re-tuned PageRank hundreds of times without ever shipping "PageRank 2."

The practical implication: ignore the folklore. The levers that move rank in 2026 are the same levers documented when A9 was first described — title structure, bullet keyword placement, backend search terms, indexed category, and the performance loop (CTR, CVR, velocity). What has actually changed is the weighting: today's A9 weights performance signals more heavily relative to relevance signals than it did in 2015, and it weights pure keyword density less. None of that requires a new name.

Why PPC moves A9 rank

Paid clicks count as a performance signal on the search term they served against. A converting Sponsored Products click at €1.20 CPC feeds the same A9 ledger as an organic click — A9 sees that the listing earned a sale from this query and rewards the listing's organic rank for that query over the following days and weeks. This is the rank-by-spend loop that underpins serious launch strategy.

Two important caveats:

• The signal is by search term, not by bid keyword. A broad-match keyword that converts on a long-tail search term feeds organic rank for the long-tail term, not the seed. This is why keyword harvesting → exact-match is the canonical loop.
• Returns reverse the signal. High-velocity sales followed by high-velocity returns pull rank down, sometimes harder than the rise.

Common mistakes

• Treating "A10" as a different algorithm. It is not. Tactics built on the A10 myth (off-Amazon traffic at all costs, abandoning backend search terms) waste budget on signals A9 has always rewarded weakly.
• Optimising only one layer. A keyword-stuffed title with no CVR work won't rank. A high-CVR listing with no keyword indexing won't show up in the first place. Both layers must move together.
• Reading rank daily. A9 has measurable noise; daily position changes of 3–8 slots are usually noise, not signal. Smooth to a 7-day rolling rank before acting on a trend.
• Optimising for keywords you don't convert on. Page-1 rank for a long-tail term that produces two sales a month is not a meaningful win — A9 will eventually demote you anyway when CVR doesn't justify the slot.