Amazon keyword tiers: how to decide which keywords count most

A practical method to tier Amazon keywords by demand, relevance, conversion and margin, with a decision matrix, a lunch box example and re-tier rules.

Once you track a real keyword set, the next question is which keywords should count most. A drop on "lunch box" and one on "lunch box with ice pack for toddler" are not the same event, yet a flat list treats them as equals. Tiers fix that. This guide explains how to assign them: which signals to use, where Amazon's own data helps, why coarse buckets beat precise volume numbers, and how often to revisit the decision.

For how many keywords to track and why weighting matters statistically, see how many keywords to track on Amazon. For how tiers feed into a visibility score, see measuring your share of page one. This article focuses on the step in between: deciding which tier each keyword gets.

What a tier is (and isn't)

A tier is a coarse statement of importance: "this keyword matters about as much as these others, more than those, less than that one." A common four-level scheme is Core, Important, Standard and Minor.

A tier is not a search volume: it folds demand together with relevance, conversion and business value. And it isn't permanent, because seasons and product changes move keywords between tiers. The goal is a ranking you can defend in one sentence per keyword, not a precise number.

The six criteria

Six questions decide most tiers. The first three carry the most weight; the last three break ties and handle exceptions.

1. Demand: how many shoppers search it?

Amazon gives brand owners two demand signals, and they measure different things.

Search Frequency Rank (SFR). In the Brand Analytics Search Terms report, Amazon's report schema defines it as the relative ranking of a search term within a department based on how often it was searched in the report period. Lower is more popular. It is a rank, not a count, and it can be pulled by day, week, month or quarter.
Search query volume. The Search Query Performance (SQP) report defines it as the number of times customers searched for a query in a period. That's an actual count, but SQP only lists queries that led customers to your products, and it requires Brand Registry.

So SFR covers the whole store but only gives an order; SQP gives counts but only for queries you already appear on. Neither tells you the volume of a keyword you have never touched.

2. Relevance: is this keyword about your product?

Amazon's Shopping Queries benchmark (Reddy et al., 2022) labels each product for a query as Exact, Substitute, Complement or Irrelevant. Use the same lens on your own keywords:

Exact: your product fully satisfies the query. "Insulated lunch box" for an insulated lunch box.
Substitute: your product could serve the need but misses a stated detail. "Bento box" for a soft-sided lunch bag.
Complement: shoppers want something used with your product. "Ice packs for lunch box."

High demand with substitute-level relevance is a trap. You may rank, but shoppers who wanted the exact thing will mostly buy something else. Relevance caps a keyword's tier regardless of its volume.

3. Conversion evidence: do searches turn into purchases?

Search Query Performance has a field most sellers overlook. Its schema defines total purchase rate as purchases divided by search query volume, for all products on that query, not just yours. Alongside it sit your own purchase count and purchase share.

That gives you two readings per query:

Market conversion. A query with high volume and a low total purchase rate is often browsing behavior: shoppers comparing, researching or not yet ready. It deserves less weight than its volume suggests.
Your conversion. A query where your purchase share is well above your click share shows that shoppers who find you there tend to buy you. Losing rank there costs real sales.

The Sponsored Products search term report adds evidence for queries outside SQP. Amazon Ads describes it as listing shopper search terms that led to at least one click, with the clicks and sales they generated. It reflects your bids and match types, not organic demand.

4. Margin

Two keywords with the same demand and conversion are not equal if one sells your higher-margin size or bundle. If "lunch box for adults" mostly sells your large insulated version and "kids lunch box" mostly sells a low-margin small one, the first deserves a higher tier. SQP reports by product, so you can usually tell which item sells on which query.

5. Strategic value

Some keywords matter beyond direct sales:

Defensive: the terms that define your category, where losing page one would hand competitors the shoppers who define your market.
Expansion: an attribute or use case you're about to launch into, such as a leakproof bento compartment.
Competitive: keywords where a key competitor is gaining fast (see tracking competitors on Amazon).

Strategic value can lift a keyword one tier. It shouldn't lift a weak keyword to Core.

6. Intent

Search research has long distinguished navigational, informational and transactional intent (Broder, 2002). On Amazon nearly every query is commercial, but intent still varies. "How to keep lunch cold" is closer to informational; "lunch box" is broad and transactional; "stainless steel bento box 3 compartment" is a shopper who knows what they want. Sellers commonly treat keywords that name a specific product type as closest to purchase; your SQP purchase rates show whether that holds in your niche. Competitor brand names are a separate case: they measure someone else's brand demand and belong outside your tiered set.

A decision matrix

Combine demand and relevance first, then let conversion, margin and strategy move a keyword up or down by at most one tier.

Demand band (from SFR or SQP volume)	Exact	Substitute	Complement
High	Core	Standard	Minor (or leave out)
Medium	Important	Standard	Minor
Low	Standard	Minor	Leave out

Then adjust:

Up one tier if your purchase share clearly beats your click share on the query, if it sells your highest-margin item, or if it's strategically important.
Down one tier if the query's total purchase rate is low compared with similar queries, or if page one is dominated by products unlike yours.
Never above Important for a substitute keyword, whatever the adjustments. If it really converts like an exact match, your relevance label was probably wrong.

Define demand bands relative to your own set, for example the top fifth by SFR or SQP volume as "high." Absolute cut-offs don't transfer between categories.

Why coarse tiers beat exact volume estimates

Weighting each keyword by an estimated monthly search volume looks more precise. It isn't, for three reasons.

There is no public ground truth. Amazon publishes SFR as a rank, and SQP counts only for queries where your products appeared. Third-party volume figures are modeled estimates. We found no independent benchmark of their accuracy; the comparisons that exist are vendors testing each other, with conflicting results.
Ranks don't convert linearly to volume. Search demand is heavily skewed toward a few frequent queries, so the volume gap between two neighboring ranks near the top is very different from the gap between two neighboring ranks deep in the list. Any rank-to-volume conversion is itself a model.
Errors carry through. If a volume estimate is off by a factor of two, a weighting built on it is off by a factor of two as well. Tiers only need the order to be roughly right, which is a far more robust claim.

This is reasoning rather than a published result, but it follows directly from what Amazon does and doesn't disclose.

What the long tail is worth

Practitioner wisdom says long-tail keywords convert better because they're more specific. The neutral evidence is mixed:

Head concentration. In an analysis of three paid search campaigns covering 4,908 keywords over 36 weeks, Skiera, Eckert and Hinz (2010) found that the top 20% of keywords drew on average 98.16% of searches and 97.21% of clicks. They also noted that the set of top keywords changes over time. That's web paid search, not Amazon, but it cautions against assuming the tail carries the business.
Small samples. Meisel, Benndorf and Ittner (2013) point out that the performance of a single low-frequency query can't be judged reliably, because its sample is too small. Grouping tail queries by shared patterns can still reveal profitable ones.
Amazon's view. Amazon researchers describe tail queries as lacking the behavioral data head queries have, and work on mapping them to equivalent head queries with the same purchase intent (Hiranandani et al., 2019).

The practical conclusion: judge tail keywords as groups, not one by one. A cluster of twenty precise, exact-match queries can deserve Standard as a group even if each one alone is Minor. And don't promote a single tail keyword to Core because it showed three purchases last month.

How often to re-tier

Amazon's own research shows why tiers can't be set once. In "Seasonal Relevance in E-Commerce Search" (Yang et al., CIKM 2021), Amazon researchers use the example of "jacket," which has different relevant products in winter than in summer. They report that 39% of queries are highly seasonally relevant to the time of search. On average, each query was highly seasonally relevant for about 31% of the year, and those periods accounted for close to 39% of annual query volume and up to 42% of purchases. Their model also picked up a bookbag's sales spike in the back-to-school period near August.

A workable rhythm:

Quarterly: re-pull SFR and SQP, re-run the matrix, move keywords that changed band.
Before your peak season: pre-promote seasonal keywords so your averages reflect the period that decides your year.
After a product change: a new size, material or bundle changes relevance and margin for whole keyword groups.
When a competitor reshapes a page: if page one of a keyword fills with substitutes, its relevance to you dropped even if your product didn't change.

When you re-tier, keep the history and note the date. A visibility trend that jumps on the day you changed tiers reflects your weighting, not the market.

Example: a generic lunch box set

The table is illustrative, for an insulated adult lunch box. Demand bands are hypothetical, not data.

Keyword	Demand band	Relevance	Adjustment	Tier
lunch box	High	Exact	—	Core
lunch box for adults	High	Exact	Sells the high-margin size	Core
insulated lunch box	Medium	Exact	Strong purchase share	Core
lunch bag	High	Substitute	—	Standard
leakproof lunch box	Medium	Exact	—	Important
lunch box for men	Medium	Exact	—	Important
bento box	High	Substitute	Page one mostly hard-shell boxes	Minor
kids lunch box	High	Substitute	Seasonal: promote before back-to-school	Standard (Important in season)
lunch box with ice pack	Low	Exact	Grouped with other tail terms	Standard
ice packs for lunch box	Medium	Complement	—	Minor

Note the biggest words: "bento box" and "lunch bag" may out-search half the list, but neither is what this product is, so neither drives the score.

Common mistakes

Tiering by gut. The keyword you typed first feels most important. Check it against SFR and SQP before giving it Core.
Everything is Core. If most of your set is Core, tiers carry no information. Keep Core to a small group of defining keywords; most keywords should sit in Standard.
Never revisiting. Tiers set in January will mislead you by August, especially for seasonal products.
Promoting single tail queries. A handful of purchases on a rare query is noise until it repeats.

For diagnosing a sudden drop once your tiers are set, see why your Amazon ranking dropped.

How JustRanks uses tiers

In JustRanks, each keyword in a Market gets a tier (Core, Important, Standard or Minor) that sets its weight, with up to 100 keywords per product. Positions are scored on the whole first page of each keyword, so grid and list pages stay comparable, and ads never count. JustRanks records the whole organic first page of Amazon.com search results for more than 200,000 keywords over time, from a fixed US delivery address, so history already exists when you re-tier. Keywords saved in any user's market are captured every day, all others every 48h. Related keywords can be found from products that rank on several of your keywords, and no Seller Central access is needed. For how tracking works overall, see how an Amazon rank tracker works.