Amazon SEO
Amazon product listings are increasingly evaluated not just by the legacy A9/A10 keyword-matching algorithm but by Rufus (the AI shopping assistant interface) and Cosmo (the underlying intent/relevance engine), which ask whether a product solves the problem a shopper described rather than whether the listing contains the exact words typed — so sellers should keep doing core keyword research but also write listings around who the product is for, what it does, what problem it solves, and what benefit each feature delivers.
Amazon Rufus is a generative AI shopping assistant (desktop and mobile app) that lets shoppers ask natural-language questions instead of typing keywords and scrolling results.
Rufus generates answers using Amazon's product catalogs, reviews, Q&A content, and structured listing data.
Amazon Cosmo is an AI-driven relevance system that decides which products are eligible to appear for a given search or shopping need, using machine learning models and knowledge graphs to connect queries to outcomes based on context, use case, and relevance rather than exact word matches.
Rufus is the interface shoppers talk to; Cosmo is the background system deciding product relevance; both sit on top of, not instead of, the original A9/A10 algorithm.
A9/A10 still factors in keyword presence plus performance signals (click-through rate, conversion rate, sales velocity) and has powered product search for over a decade.
The classic framing: A9 asks 'Does the listing contain the words the customer typed?'; Cosmo asks 'Does this product solve the problem the customer described?'
Example from Cosmo research: for the query 'shoes during pregnancy,' A9 just matches 'shoes,' while Cosmo understands the underlying need (slip-resistant, arch support, low heel) and surfaces listings that communicate those capabilities even without those exact words.
Cosmo does not replace A9 — keywords still matter for indexing and building the initial candidate pool; Cosmo then changes how that pool is evaluated and ranked.
Amazon rules prohibit using the same keyword more than twice in a title; the presenter has long advised writing titles for humans as well as for keywords, since bullets and descriptions are less read by humans but are now also parsed by Rufus/Cosmo.
Listings should avoid being too narrow/specific in case that alienates customers the product doesn't perfectly fit.
A simple copywriting framework: whenever a feature is mentioned, spell out the benefit/use case it delivers to the customer (e.g., what does '304 stainless steel' actually give the customer?).
Practical research technique: treat Rufus like ChatGPT or Gemini and ask it natural-language shopping questions to surface sellable features/benefits (e.g., asking for a picnic blanket suitable for the beach surfaces 'waterproof and sand-proof' as key attributes).
Rufus also surfaces suggested prompts on the search results page and on product detail pages, which the presenter believes signal where Amazon is heading long-term, partly because these prompts are starting to appear in PPC data.
Product detail page prompts include an AI-generated 'What are customers saying?' review summary distinct from the standard star-rating summary, which reveals what to prioritize in listing copy.
The left-hand filter/tick-box attributes on the search results page reflect the structured attribute data Cosmo considers; sellers should treat them as a checklist of attributes to clearly state in the listing.
Only claim features/benefits the product genuinely has — mismatched suggestions lead to unhappy customers, returns, and Amazon/Rufus 'learning' from that, ultimately suppressing future suggestions for those terms.
Amazon launched Sponsored Products Prompts and Sponsored Brands Prompts, an AI-powered ad enhancement that surfaces relevant product info in shopping results and product detail pages, opening either a Rufus dialogue or an on-page response when clicked; per the referenced article, at the time of publication (start of March) this was US-only and not yet in the UK.
Rationale for prompts: the Amazon product detail page is not a great browsing experience and customers struggle to find exact information, so prompts function as a '24/7 virtual product expert' that answers questions before they're asked, building shopper confidence and purchase likelihood.
Current recommendation: keep core keyword research as the foundation, don't keyword-stuff, and layer in customer-intent writing (problems solved, use cases, features, and benefits) on top.
Amazon Rufus — A generative AI shopping assistant embedded in the Amazon app that lets shoppers ask natural-language questions instead of typing keywords, generating answers from Amazon's catalogs, reviews, Q&A, and listing data. Apply: Open Rufus and role-play as a shopper asking natural-language questions about your product category to surface the features and benefits shoppers actually care about.
Amazon Cosmo — An AI-driven relevance system that determines which products are eligible to appear for a search by using machine learning and knowledge graphs to match shopper intent to products by context and use case rather than exact keywords. Apply: Write listing copy that communicates who the product is for, what it does, and what problem it solves, not just the literal search keywords, so Cosmo can match it to relevant intents.
Amazon A9/A10 — The legacy Amazon search algorithm that ranks products by keyword match in the listing combined with performance metrics like click-through rate, conversion rate, and sales velocity. Apply: Continue core keyword research and keyword placement in listings as the indexing foundation, since Cosmo builds on top of A9's candidate pool rather than replacing it.
Feature-to-benefit rewriting — A copywriting method where every stated product feature is paired with an explicit explanation of the benefit or use case it delivers to the customer. Apply: For each feature mentioned in a listing (e.g., '304 stainless steel' or '8-mil high-density foam'), add a line stating what real-world benefit that feature gives the customer.
Two-use title keyword rule — An Amazon title rule restricting sellers from using the same keyword more than twice within a product title. Apply: When drafting or auditing titles, check that no single keyword phrase appears more than twice, and use remaining space for human-readable, benefit-oriented language.
Rufus natural-language query research — A research technique of treating Rufus like a general AI chat agent (comparable to ChatGPT or Gemini) and asking it the kind of question a real shopper would ask. Apply: Type a natural shopper question into Rufus (e.g., 'suggest a picnic blanket suitable for the beach') and mine the response for sellable features and benefits to include in your listing.
Rufus suggested-prompt mining (search results page) — A feature where Rufus, opened from the search results page, surfaces a set of suggested prompts related to the searched product category. Apply: Open Rufus from the search results page for your product's keyword and review the suggested prompts to identify which materials, features, or use cases to spell out in your bullets and description.
Rufus product-detail-page prompts and AI review summary — Rufus prompts available on individual product detail pages, including a 'What are customers saying?' prompt that triggers an AI-generated summary of review themes distinct from the standard star-rating breakdown. Apply: Check the AI review summary and prompt questions on your (or competitors') product detail pages to identify praised attributes and unresolved customer questions to address directly in your copy.
Filter/attribute checklist mining — Using the left-hand filter tick-boxes on the Amazon search results page, which represent the structured attribute data Amazon evaluates across all competing listings. Apply: Treat the filter categories shown for your product's search results as a checklist and ensure your listing explicitly states each relevant attribute (e.g., waterproof, washable, lightweight) so Cosmo can qualify you for those filters.
Avoid mismatched-intent targeting — A caution against optimizing a listing to be suggested for features or use cases the product doesn't actually satisfy. Apply: Only write copy claiming features/benefits your product genuinely delivers, since mismatched suggestions cause returns that train Amazon/Rufus to stop recommending you for those terms.
Sponsored Products Prompts / Sponsored Brands Prompts — An AI-powered enhancement to existing Amazon ad campaigns that automatically surfaces relevant product information as clickable prompts in shopping results and product detail pages, opening a Rufus dialogue or responding on-page. Apply: Monitor this PPC prompt feature (US-only as of the referenced article) as an early signal of which intents and features Amazon's AI considers relevant for your product, and incorporate those into organic listing copy.
The mismatch-suggestion feedback loop means over-optimizing for intents you don't fully satisfy can actively hurt future visibility, since returns from disappointed customers train the system to stop suggesting you for those terms.
PPC prompt data (Sponsored Products/Brands Prompts) is described as a leading indicator of where Amazon's AI ranking is heading, giving sellers early visibility into which intents/features Amazon considers salient for their product category.
The search-results-page filter checkboxes, which most shoppers ignore, are reframed as a direct checklist of the structured attributes Cosmo evaluates across competing listings.
Because Cosmo layers on top of A9 rather than replacing it, the practical strategy is additive, not a rewrite: keep existing keyword indexing intact and layer intent/benefit language on top, rather than discarding keyword-era listing practices.
«A9 asks, 'Does the listing contain the words the customer typed?' Cosmo asks, 'Does this product solve the problem the customer described?'»
— 03:33
«Keywords get you considered, intent signals get you recommended.»
— 05:51
«Rufus is basically... think of it as an AI chat agent. So, imagine you were talking to ChatGPT or to Gemini, what question would you ask that?»
— 09:19
«For beach picnics, you'll want a blanket that's both waterproof and sand-proof.»
— 10:05
«If Rufus does suggest your product and it's for a feature your product doesn't actually have, when the customer gets there, they might be unhappy and then they return that product and Amazon and Rufus will learn from that.»
— 12:24
«Sponsored products prompts and sponsored brands prompts address this by functioning as a 24/7 virtual product expert.»
— 16:28
«Anytime you talk about a feature... what is the benefit and use case to the customer?»
— 08:26
«Now, if you're not sure how to do your keyword research in the first place, I've recently recorded a video that shows you how I'm doing it now in 2026 using Data Dive.»
A practical, well-sourced explainer that grounds its claims in an Amazon-adjacent article and live demonstrations of Rufus prompts, though the presenter openly frames current guidance as provisional and plans a follow-up once Rufus/Cosmo optimization best practices mature.