In every public Amazon channel we checked on 2026-08-19 — its seller-facing content, its research site, and its own forums — Amazon has not announced, named, or documented a search algorithm called “A10.” The term was created by the seller and agency community, not by Amazon. Every page that ranks for the phrase is a third-party interpretation of ranking behavior — useful in places, but not an official specification, and not something you can look up in Amazon’s own documentation.
That does not make the underlying question fake. Search ranking on Amazon really does change, and sellers really do need a model of it. This page separates three things that usually get blended together: what Amazon itself has published, where the “A9 / A10” naming came from, and which ranking inputs actually have a documented basis versus which are practitioner inference. For the mechanics of ranking itself, we point to the existing deep dives rather than repeat them: the Amazon SEO guide covers the end-to-end system, and Amazon COSMO covers the intent-search layer Amazon’s own researchers published.
What We Checked Before Saying “Amazon Never Used It”
“Amazon never said that” is an easy claim to make and a lazy one to make from memory, so we went looking for the term in Amazon’s own channels on 2026-08-19.
| Official channel checked (2026-08-19) | What we looked for | Result |
|---|---|---|
sell.amazon.com seller blog, including its own Amazon SEO post (published 2025-09-03) | The strings “A9” and “A10” | Neither term appears anywhere on the page |
amazon.science — Amazon’s public research site, including the Semantic Product Search paper (KDD 2019) and the commonsense knowledge graph post (2024-05-10) | The strings “A9” and “A10”; any named ranking algorithm | Neither term appears; Amazon’s published search work is named by system or paper, never by version letter |
| Seller Central Seller Forums (the public, Amazon-hosted forums) | Threads discussing A10 and any Amazon-side reply | “A10” appears only in posts written by sellers, not in Amazon documentation. The Amazon-side reply says the opposite of a spec |
That last row is the informative one. In a Seller Forums thread titled “Decoding the A10 Algorithm: Has Anyone Found Reliable Documentation?”, an Amazon community manager replied: “Amazon does not disclose information or release official documentation around search/discovery algorithms.” A seller in the same thread put it more bluntly — Amazon “will never release comprehensive documentation about how search/discovery algorithms work.”
One honest limit on this check: pages behind a Seller Central login were not searched, because this site does not operate a Seller Central account. What we can say precisely is that in Amazon’s public seller-facing content, its public research site, and its public forums, the term “A10” exists only in seller-written text.
Where “A9” Came From — a Company, Not an Algorithm Name
“A9” at least has a real referent. A9.com was an Amazon subsidiary founded in 2003 in Palo Alto, working on search and advertising technology; the name is a numeronym for algorithm — the letter A plus its nine remaining letters. Amazon retired the standalone A9.com brand in 2019 and folded its teams into Amazon’s broader search organization (third-party source: Wikipedia’s A9.com entry, checked 2026-08-19).
So sellers who say “A9” are borrowing the name of a business unit and applying it to a ranking system. That is a reasonable shorthand and it caused little harm. The problem starts at the next step: because “A9” looked like a version number, the community incremented it. There was no A9 release, so there could be no A10 release — but the naming convention implied both.
Where “A10” Came From
Third-party accounts that discuss the term’s origin generally place it around 2020, coined by agencies and sellers to label a set of observed changes — external traffic seeming to matter more, conversion quality seeming to matter more than raw volume, seller-level factors seeming to matter at all. These are practitioner observations, dated and attributed as such. They are not an Amazon release note.
Two things then happened that made the label harder to dislodge. First, the phrase is commercially useful: “the new A10 algorithm” is a better headline than “some ranking behavior appears to have shifted.” Second, generative answer engines began synthesizing those blog posts into confident summaries. Claims currently circulating in AI-generated answers include “A10 launched in 2025” and “A9 promoted profitable products while A10 promotes relevant ones” — statements with no Amazon source behind them, produced by summarizing an ecosystem that was already summarizing itself.
Layer 1 — What Amazon Has Actually Published
Amazon publishes a good deal about search. It just never publishes it as a versioned algorithm.
- Semantic Product Search (KDD 2019). Amazon researchers describe the problem as: “We study the problem of semantic matching in product search, that is, given a customer query, retrieve all semantically related products from the catalog.” The paper’s subject is moving beyond lexical inverted-index matching to handle synonyms, hypernyms, morphological variants, and misspellings.
- COSMO, the commonsense knowledge system (published 2024). Amazon’s own write-up (2024-05-10) reports that the knowledge graph can improve downstream task performance by as much as 60% in its evaluations. This is Amazon describing intent understanding — inferring the need behind a query, not just its keywords. Our full breakdown is in the COSMO guide.
- Rufus, Amazon’s shopping assistant. A shipped, publicly announced conversational surface that changes how some shoppers reach products; see Rufus for sellers.
- Seller-facing guidance. Amazon’s own SEO post states that “Listing quality and your account health can contribute to search rankings” (published 2025-09-03) and otherwise sticks to practices — keyword research, titles, descriptions — rather than weights.
Read together, these say something clear: Amazon’s search is a stack of models that changes continuously, and Amazon documents capabilities and practices, never a ranked list of factors with weights. There is no version number to track because Amazon does not ship search as versioned releases.
Layer 2 — The Circulating A10 Claims, Checked
| Claim in circulation | Verdict | Basis (checked 2026-08-19) |
|---|---|---|
| “A10 launched in 2020 / 2025 as Amazon’s new algorithm” | No support | No Amazon announcement, help page, blog post, or research publication uses the name |
| “A9 favored profitable products; A10 favors relevance” | No support | Traces to a September 2019 Wall Street Journal report that Amazon adjusted search to favor profitable listings — a report Amazon publicly disputed, with a spokesperson quoted as saying “We have not changed the criteria we use to rank search results to include profitability” (CNBC, 2019-09-16). No version name was involved either way |
| “Off-Amazon traffic now carries more weight” | Partly supported, indirectly | Amazon runs the Brand Referral Bonus, an official program that pays brands a bonus on sales driven from outside Amazon. That documents commercial interest in external traffic; it does not document a ranking weight |
| “PPC influence on organic rank was reduced” | No support | Amazon documents sponsored placements as advertising, distinct from organic results, and publishes no statement about organic weighting in either direction |
| “Seller authority and account health matter” | Partly supported | Amazon’s own SEO post says listing quality and account health “can contribute to search rankings” (2025-09-03). The word is contribute, not a weight — see account health rating |
| “Conversion rate and click-through drive rank” | Practitioner inference | Amazon’s Search Query Performance dashboard reports impressions, clicks, and purchases per query, which shows Amazon measures the funnel. Measurement is not a published ranking input |
The pattern worth noticing: the claims that survive contact with evidence are the vague ones, and the claims that are specific enough to act on are the ones with nothing behind them.
Layer 3 — Ranking Inputs Sorted by Evidence Type
Drop the naming argument and the practical question remains: what should you actually optimize? Sort inputs by how well they are evidenced, and treat the tiers differently.
| Input | Evidence type | What the evidence is |
|---|---|---|
| Keyword coverage and relevance matching | Officially documented | Amazon’s seller guidance instructs keyword research and keyword placement in listing fields; semantic matching is the subject of Amazon’s own published research |
| Intent behind the query, not just terms | Officially documented | COSMO and related Amazon Science publications (2024) |
| Listing quality (title, images, description completeness) | Officially documented | Amazon seller blog, 2025-09-03; also the subject of listing optimization |
| Account health | Officially documented as a contributor | Same source, same date; no weighting given |
| Conversion rate, click-through rate, sales velocity | Practitioner inference | Widely reported by agencies and tool vendors from correlation studies; Amazon reports these metrics to sellers but has never named them as ranking inputs |
| External traffic as a rank input | Practitioner inference | Agency observation since roughly 2020; Amazon’s documented position is the Brand Referral Bonus, a payout program |
| Price competitiveness | Practitioner inference | Frequently asserted; Amazon’s public statements tie price to the featured offer, which is a separate system from search ranking |
The useful discipline is this: spend your build time on tier-one inputs, because they are documented and stable. Treat tier-two inputs as hypotheses you test on your own catalog, with your own rank tracking and your own before-and-after dates — not as rules you inherited from a blog post that inherited them from another blog post.
Why the Name Won’t Die
Sellers need a mental model, Amazon supplies none, and a version number is the most satisfying shape a model can take. It implies discrete releases, a knowable current state, and — most appealingly — that someone out there has the changelog.
The Seller Forums thread quoted above is the honest picture of that gap: a seller asking whether anyone has found reliable documentation, another seller saying it will never exist, and an Amazon community manager confirming that Amazon does not publish it. A vacuum that specific gets filled by whoever writes the most confident page. Recognizing that is more useful than memorizing anyone’s “eight A10 ranking factors” list, because next year the same vacuum will be filled with a different number.
Frequently Asked Questions
Is the Amazon A10 algorithm real?
There is no algorithm that Amazon calls A10 in any public channel we checked. The ranking behavior people describe under that name is real and does change; the name is industry shorthand. As of a check of Amazon’s public seller content, research site, and forums on 2026-08-19, the term appears only in seller-written posts.
Did Amazon ever call its search algorithm A9?
Not as an algorithm name. A9.com was a real Amazon subsidiary (founded 2003, wound down as a standalone brand in 2019) that worked on search technology. Sellers borrowed the company name for the ranking system; Amazon did not publish it as a version label.
Is there an A11 algorithm?
No. The same incrementing logic that produced “A10” has already produced “A11” in some forum posts. Since we found no official A9 or A10 release in any public Amazon channel we checked, there is nothing for an A11 to follow.
Does external traffic improve Amazon rankings?
Amazon has not documented external traffic as a ranking input. It does run the Brand Referral Bonus, which pays brands a bonus on externally driven sales, so external traffic has a documented commercial benefit regardless of its ranking effect. Treat rank impact as a hypothesis to test, not an established rule.
If Amazon publishes no specification, what should sellers optimize?
Work the documented tier first — relevance and keyword coverage, intent-aligned copy, listing completeness, account health — then test everything else on your own listings. Start with keyword research and the Amazon SEO guide.
Conclusion
The honest answer to “what is the A10 algorithm” is that it is a name the seller community gave to its own observations, and one that did not appear in any Amazon channel we checked on 2026-08-19. That is not a reason to ignore ranking — it is a reason to grade your sources. Amazon’s published research tells you the direction of travel: from lexical matching toward intent understanding. Amazon’s seller guidance tells you what it will state plainly: listing quality and account health contribute. Everything past that line is inference, and inference is worth having as long as it is labeled.