How do Amazon keywords work when AI reads your listing?

Rufus is now Alexa for Shopping. It answers questions in the search bar. What a keyword is now, who reads it, and what the Keyword Mastery framework checks first.

Amazon keywords now work on two levels. The index still matches the words in your title, bullets and backend to what a shopper types. Above it, Amazon's AI reads your listing for meaning: who the product is for, what problem it solves, when people use it. A keyword does both jobs only when your listing answers the question behind it.

You added every high-volume keyword to the title. Impressions went up. Orders stayed flat. The search term report shows why, if you sort it by orders.

Take a generic product: an insulated water bottle. Near the top of the report sits "water bottle". Big impressions, a long column of clicks, few orders. Further down sits "leak proof bottle for kids". Fewer clicks. More of them buy. Same product. Same listing. Two different people. You paid for both.

Now the second person opens the Amazon app and types a full question: which bottle won't leak in my son's school bag? This is a sentence. Since May 2026, the Amazon search bar answers sentences.

What is an Amazon keyword?

An Amazon keyword is the phrase a shopper types to describe a need, with the vocabulary they have at that moment. It tells you three things: how well the shopper knows the category, how specific the need is, and how close they are to a purchase.

"Water bottle" says almost nothing. With "leak proof", the shopper has a problem to solve. With "for kids with straw lid", the shopper has read reviews, compared options and knows the spec. Same root, three buyers.

This is the first idea in my Keyword Mastery framework. Read the person behind the keyword, then decide what your listing owes them.

What are the three parts of an Amazon keyword?

An Amazon keyword has three parts: a root, a modifier and a qualifier. The root names the category. The modifier narrows the product type or the problem. The qualifier narrows to one buyer, size, use or spec. Each layer you add shrinks the audience and moves the shopper closer to a purchase.

Keyword	Root	Modifier	Qualifier	Who is searching
water bottle	water bottle	none	none	Browsing, no clear need yet
leak proof water bottle	water bottle	leak proof (problem)	none	Has a problem, comparing products
leak proof water bottle for kids	water bottle	leak proof (problem)	for kids (audience)	Knows the problem and the user, close to buying

One word inside the root does a special job. "Bottle" is the format word. It separates your product from a tumbler, a flask or a jug, which hold water in a different form. Without it, "leak proof for kids" turns ambiguous. Keyword Mastery treats the format word as a category boundary. Searches without it carry shoppers who have not picked a format yet. Expect a lower CVR from them.

The layers also explain how one title phrase covers several searches. "Leak proof water bottle for kids" holds three shorter searches as unbroken word sequences: drop "for kids", drop "leak proof", or keep only the root. It does not hold "leak proof bottle", because "water" breaks that sequence. Keyword Mastery counts a phrase as covered only when its words sit unbroken, in order. Anything else is partial coverage.

Why is Amazon keyword research different from Google?

Amazon keyword research is different because every Amazon search comes from someone who plans to buy. Google serves every stage of a question, from "are steel bottles safe" to "where to buy one". Amazon serves the last stage. Google's four intent types collapse into one line of readiness. Your job is to read how ready each buyer is.

Amazon also learns from a signal a search engine for information does not have: the purchase at the end of the search. Amazon's science team built COSMO, its commonsense knowledge system, from two sources. Searches paired with the purchases after them. Products bought together in the same session. In that design, every order teaches the system what a product is for.

What is the Amazon intent spectrum?

The Amazon intent spectrum is the line every search sits on, from category browsing to a committed purchase. Five positions cover it. A search term's position on the line tells you what CVR to expect from it. It also tells you whether a low CVR is normal or a listing problem.

Position	Example	What the shopper is doing
Category exploration	water bottle	Open to any product in the category
Product-type aware	insulated water bottle	Knows the format, comparing brands
Feature-specific	insulated water bottle with straw lid	Knows the feature, narrowing the list
Specification level	18 oz insulated kids water bottle straw lid	Knows the spec, choosing between the last few
Brand-committed	[your brand] water bottle	Decided on the brand, looking for the listing

A low CVR on a category exploration term is normal. The same CVR on a specification-level term means the page lost a buyer who arrived ready. On your dashboard, both look like the same number.

Every search term gets two labels. Relevancy: relevant, partially relevant or not relevant. Intent match: exact intent, close intent or broad intent. A relevant term with broad intent earns a test budget. A not relevant term with high volume earns a negative.

What are the 8 modifier types, and what does each one ask your listing?

A modifier is the word a shopper adds to a root to narrow the results. Keyword Mastery sorts modifiers into eight types. Each type asks your listing a different question. Each one needs its own line or image on the page to answer it.

Modifier type	Water bottle example	What the shopper needs to see
Feature	insulated, stainless steel	The feature named early, with proof it works
Audience	for kids, for men	The user in the image, and the right size
Problem	leak proof, no metal taste	The problem solved in the first bullet
Occasion	for gym, for school	The use scene, in images and in words
Outcome	keeps drinks cold	The result, said plainly, backed by reviews
Comparison	best, vs plastic	A clear reason to pick this one
Price tier	premium, budget	A price and a page that match the tier
Size or format	32 oz, 2 pack	Exact sizes, contents and fit

Read your top converting search terms with this table open. Say the shopper searched "leak proof". If your first bullet talks about the powder coat, the page answers a question nobody asked. If every image shows an adult at the gym, the parent packing a school bag leaves.

What changed when Rufus became Alexa for Shopping?

In May 2026, Amazon merged Rufus into a new assistant, Alexa for Shopping. Two things changed for your keywords. In the Amazon app, shoppers now ask the assistant questions in the main search bar. Bloomberg reports the second change. The search bar splits traffic. Long questions get an AI answer. Short searches keep the standard product grid.

EcomCrew and GeekWire reported Amazon retired the Rufus name. The assistant also has its own chat window. Amazon says the assistant combines Rufus's product expertise with your shopping history and your conversations with Alexa. It answers questions and compares items. Bloomberg gave two examples. A request to compare espresso machines triggers AI answers and personalized picks. A simple search like "pants" still returns the grid.

In its third-quarter results in late October 2026, Amazon reported:

Customers who used it in the last 12 months: more than 400 million
Customer interactions: up 570% year over year

Put that next to the intent spectrum. Your root keyword sits on the grid. The long question, full of modifiers and qualifiers, goes to the assistant. The assistant needs your listing to answer it.

So the qualifier kept its value and changed its job. On the grid it is an exact phrase to index. In the assistant it is a question your bullets answer in plain words.

How does COSMO read the meaning behind a keyword?

COSMO is a system Amazon's researchers built to connect searches and products to the human context around them. Amazon's science team described it in 2024. It runs large language models over shopping data to find commonsense relationships, such as who a product is for and what event it serves. It stores them in a knowledge graph.

The relationships have names, such as used for function, used for event and used for audience. Amazon's example is a search for "shoes for pregnant women". The goal is for the system to work out that the shopper wants slip-resistant shoes. The shopper never typed "slip-resistant".

Look at those names again. Used for audience is the audience modifier. Used for event is the occasion modifier. Used for function covers problem and feature. Keyword Mastery's modifier list and COSMO's relation list describe the same thing from two sides. The shopper adds a modifier to the search. In COSMO's design, the system looks for the matching fact on the product side.

Amazon's paper tests COSMO on search relevance. It does not publish COSMO's weight in live ranking. I read it as the direction Amazon is building in.

For your listing, the rule is plain. If the bottle fits a car cup holder, show the cup holder and write the words. A fact your listing never states is a fact the system has to infer. Inference is where you lose to the competitor who stated it.

What does this look like in a real account?

In a real account, the gap shows up as ad spend on searches the page never answers. A premium gifting brand at a $65+ price point came to us with ACoS in the mid-40s. The first read was a bid problem. The search term report said keyword problem.

Broad discovery and auto campaigns were pulling in commodity shoppers. A $65 product does not convert them at any bid. In the audit month:

Ad spend with zero sales: $6,846
Share of ad spend with zero sales: 76.7%

The listing had a second gap. Titles and bullets led with features and color. The competitors who ranked were winning on use case and outcome: who the gift is for and what moment it marks. In modifier terms, shoppers searched with audience, occasion and outcome modifiers. The page answered with feature modifiers.

We rebuilt around the intent the account had already converted on. Five exact match structures around the converting intent clusters. Targeting moved to occasion, recipient and outcome. We rewrote titles and bullets to lead with use-case language.

One month later:

Ad sales: $19,143 to $31,107
ACoS: 46.57% to 37.23%
Ad spend: up 29.9%

In the same month we also added negatives, paused ad formats and moved spend to a hero variation. The change started with one decision: stop buying broad intent the page did not answer. Then make the page answer the intent that converted.

Where does the Keyword Mastery framework fit?

Keyword Mastery is my framework for turning a keyword file into listing and campaign decisions. A Cerebro export or search term report goes in. Out comes a map. Every keyword in it holds one job: title, bullet, description, backend, exact match, phrase match, monitor or negative.

This is the work I do. I run Amazon accounts. I help brands and their freelancers read keyword files as buyer signals.

It runs in layers: root pools, the format word, the intent labels from this article, semantic coverage of every fact the modifiers ask about, then campaigns built to mirror the intent tiers.

AI does part of this work in my own setup. It cut keyword segmentation from 50 minutes to 5. It groups the terms. I decide which demand earns the spend. I check every line.

This article covers the why and the what. The full Keyword Mastery manual stays a separate guide. If the case above made you look hard at your ACoS, the next read is my article on what a good ACOS is: break-even ACOS and CPA.

What to do this week

In Brand Analytics, open Search Query Performance and pull your top 20 search terms by sessions.
Mark each one root only, root plus modifier, or root plus modifier plus qualifier.
For every modifier, write its type: feature, audience, problem, occasion, outcome, comparison, price tier, or size and format.
Open your listing on a phone. For each modifier type, find the line or image answering it. Mark the gaps.
In your search term report, list the broad intent terms with spend and zero orders in the last 30 days. Decide which ones the page will never answer. Make those negatives.

FAQ

Do keywords still matter now that Amazon has an AI shopping assistant?

Yes. Bloomberg reports simple searches still return the product grid. The assistant takes the long, question-shaped searches.

Should I write my listing for Amazon's AI assistant or for the search index?

Both. Keep exact phrases unbroken in the title for the index. Write bullets as direct answers to the questions your modifiers imply. The assistant then has facts to repeat.

What is the Keyword Mastery framework?

Keyword Mastery is Umair Ali's framework for reading Amazon keywords as buyer signals. It labels every term by root, format, intent and modifier type, then gives each one a job in the listing or the campaigns.

The keyword was always the shopper's sentence. Now the search bar reads it like one.

Sources

Amazon Science, "Building commonsense knowledge graphs to aid product recommendation" (COSMO, SIGMOD 2024), 10 May 2024. https://www.amazon.science/blog/building-commonsense-knowledge-graphs-to-aid-product-recommendation
Bloomberg via The Star, "Amazon puts Alexa inside the shopping search bar in AI push", 14 May 2026. https://www.thestar.com.my/tech/tech-news/2026/05/14/amazon-puts-alexa-inside-the-shopping-search-bar-in-ai-push
Amazon, "Meet Alexa for Shopping, your personalized, agentic AI assistant on Amazon", 13 May 2026. https://www.aboutamazon.com/news/retail/alexa-for-shopping-ai-assistant
EcomCrew, "Amazon retires Rufus and launches Alexa for Shopping as its unified AI assistant", 13 May 2026 (GeekWire reported the same, via AI News 18 May 2026). https://www.ecomcrew.com/amazon-retires-rufus-and-launches-alexa-for-shopping-as-its-unified-ai-assistant/
Amazon third-quarter 2026 results (29 Oct 2026): Alexa for Shopping "used by 400 million+ customers in the last 12 months, with customer interactions up 570% year-over-year", as reported by Cord Cutters News (29 Oct 2026) https://cordcuttersnews.com/amazon-posts-201-9-billion-in-q3-sales-aws-revenue-climbs-39/ and James Sharp & Co. (30 Oct 2026) https://www.jamessharp.co.uk/market-news/amazon-q3-2026-results/