How to build an Amazon keyword list for one niche, step by step

Where Amazon keywords come from, from autocomplete and Brand Analytics to ad reports and reviews, and how to validate each one on page one.

A useful keyword list is not a dump of everything a tool suggests. It is a short, checked map of how shoppers search for one kind of product. This guide covers the sourcing side: where candidate keywords come from, what each source can and can't tell you, and how to validate every term before it goes into your list. How many keywords you need, and how to weight and prune them, is covered in how many keywords to track on Amazon.

Start with a seed and a niche definition

Write down two things before you collect anything:

The seed: the plainest generic name of the product, such as "lunch box".
The niche boundary: what counts as "your" product and what doesn't. For a hard-shell lunch box, a soft lunch bag might be a substitute, and an ice pack is a complement.

The boundary decides every later step. Amazon's own tooling thinks in the same terms: Product Opportunity Explorer defines a niche as "a collection of search terms and products that represent a specific customer need," grouped by the products customers view or buy. Your list is your own version of that collection.

Source 1: Amazon search autocomplete

Type the seed into the Amazon search box and read the drop-down. Amazon's seller guide on keyword research recommends exactly this as a starting point.

How it works. Amazon's researchers describe query auto-completion as "a ranked list of search queries as suggestions for a customer-entered prefix" (Ramachandran and Murthy, SIGIR 2019). In their production test, the default suggestions were driven by a behavioral model, and offered suggestions "are optimized for past search impact," meaning queries that previously led shoppers to products they bought. A 2024 Amazon paper notes that traditional systems "build upon query popularity" and struggle with prefixes they haven't seen before. These papers describe systems Amazon built and tested; the exact production setup today is not public.

What it reveals. Real phrasings that shoppers have used and that led to purchases: qualifiers (size, material, audience), word order, and the vocabulary people actually type.

Limits to keep in mind:

No volumes. A suggestion's position in the list is not a search volume.
Prefix-bound. You only see completions of what you typed. Run the seed with modifiers ("lunch box for", "insulated lunch", "lunch box with") and with each letter of the alphabet after the seed to surface more.
Context-dependent. Amazon research on session-aware auto-completion uses the shopper's previous query as input (Yadav et al., KDD 2021). Your own recent searches can shape what you see, so use a fresh, logged-out browser session.
Popular phrasings only. Rare but relevant long-tail queries often never appear.

Source 2: Brand Analytics and Product Opportunity Explorer

If you are a brand owner, Amazon's own search data is the most reliable source you have. Brand Analytics requires Brand Registry, a Professional selling account and the brand representative role.

Report	What it shows	What it can't show
Top Search Terms	Search frequency rank per term, department, the three most-clicked products with their click share and conversion share	Absolute search volume
Search Query Performance	Queries that led shoppers to your products, with query volume and total vs. your impressions, clicks, cart adds and purchases	Queries where your products never appeared
Product Opportunity Explorer	Niches as clusters of search terms and products, plus aggregated review themes	Terms outside the niches Amazon has built

Two practical uses. Top Search Terms can be filtered by keyword, so search it for your seed to find every popular term containing it, and look at which products win the clicks. Search Query Performance shows where your listing already gets impressions but few clicks, which often points to keywords worth tracking closely. Amazon says Brand Analytics data usually goes back three years, so seasonal terms are visible too.

Source 3: Sponsored Products auto campaigns

An automatic campaign is a paid discovery tool. Amazon Ads describes four targeting groups: close match (search terms closely related to your product), loose match (loosely related terms), substitutes (detail pages of similar products) and complements (detail pages of products that go with yours).

The output that matters is the search term report. Amazon says it lists the search terms shoppers entered that led to at least one click, and recommends adding strong terms as keywords and weak ones as negatives. For keyword research, read it as a list of real queries where Amazon judged your product relevant enough to show it. Some rows show ASINs instead of queries; Amazon says these can be moved into product targeting, and for research they point to the competitor products worth studying in the next step.

Two caveats. The report reflects your bids, budget and match settings, so it is a sample, not a census. And a click on an ad says the term is plausible, not that you can rank organically for it.

Source 4: Competitor titles and bullets

Pick the products that keep appearing on the first page of your seed and its main variants. Amazon's keyword guide recommends reviewing their titles, bullet points and descriptions. Look for:

Feature nouns you haven't used ("bento", "compartments", "leakproof lid").
Audience words ("for kids", "for adults", "for work").
Alternate names for the product itself.

Leave out brand names. They measure brand demand, and Amazon prohibits other brands' names in backend search terms.

Source 5: Reviews and customer questions

Sellers write titles; shoppers write reviews. Amazon's guide suggests reading reviews and customer questions to see which words customers use. That is where problem language lives: "keeps food cold until lunch", "fits in my work bag", "my toddler can open it". Turn each phrase into a candidate search ("small lunch box for work", "easy open lunch box for toddlers").

Product Opportunity Explorer's review insights group review themes for a whole niche, which saves reading hundreds of reviews. Note that sellers report Amazon's Rufus assistant now answers many product questions, so the classic question list is less visible on some pages; reviews remain the richer source.

Check the category behind each query

Amazon researchers note that e-commerce platforms "store and structure product information and search data in a hierarchy," and that short, ambiguous queries are hard to place in it (He et al., WWW 2024). For your list, that means a keyword can belong to a different part of the catalog than your product.

Two quick checks:

Department filters. Run the query and look at the department and category refinements on the left of the results. If they lean toward kitchen storage containers while your product is a kids' lunch bag, the query belongs to another niche.
Top Search Terms department field. The report assigns each term to a department, which helps sort large lists quickly.

Synonyms, plurals and misspellings

Amazon's search does not rely on exact word matching alone. Amazon's semantic product search paper explains that pure lexical matching fails on synonyms, morphological variants such as "woman" and "women", and spelling errors, and describes a model trained on shopper behavior to bridge those gaps (Nigam et al., KDD 2019). Amazon Ads' phrase and exact match also include plurals.

What that means for sourcing:

Synonyms deserve a check, not an assumption. "Lunch bag" and "lunch box" may show very different products. Keep both only if both pages fit your niche.
Plurals are usually duplicates. "Lunch boxes" rarely needs its own row.
Misspellings are mostly handled for you. An Amazon staff tip on backend search terms says to include spelling variations but that there is "no need for misspellings," and Amazon's auto-completion research reports fewer misspelled searches when completions are offered inline. Track a misspelling only if its page one looks different from the correct spelling's.

Validate every keyword on page one

This is the step most lists skip. For each candidate, open the search and look at the whole first page of organic results, ignoring ads.

Amazon's Shopping Queries Dataset labels each result for a query as exact, substitute, complement or irrelevant (Reddy et al., 2022). Use the same lens:

Mostly exact matches to your product: keep it.
Mostly substitutes: keep it at a low weight, or move it to a neighboring niche.
Mostly complements or irrelevant results: drop it.

Then check overlap. If the same competitors appear on a candidate and on your seed, it belongs to your market. If the page is full of products you never see elsewhere, it is probably a different market, even if the words look close.

Dedupe, then split tracking from backend

Merge candidates that return near-identical pages: plurals, word-order swaps, and terms that differ only by a filler word. Keep one representative.

Then separate two lists that are often confused:

The tracking list is every validated keyword you want to watch. It can be long.
The backend search terms field is for indexing words that aren't already in your title and bullets. Amazon staff guidance says to stay within 250 bytes, include synonyms, abbreviations and alternate names, skip punctuation and stop words, and not repeat words already in your title or brand. Amazon prohibits brand names (yours or others'), ASINs, temporary statements such as "new" or "on sale now", subjective claims such as "best" or "cheapest", and abusive terms. Amazon has warned that an over-limit entry may not be indexed at all, and non-English characters can take more than one byte each.

A worked example: "lunch box"

The table is illustrative. It shows how the method sorts candidates, not real suggestions or volumes; your sources will return different terms.

Candidate	Source	Page-one check	Decision
lunch box	Seed	Mostly exact	Keep (core)
lunch box for kids	Autocomplete	Mostly exact	Keep
insulated lunch box	Autocomplete, search term report	Mostly exact	Keep
bento box for adults	Competitor titles	Mix of exact and substitutes	Keep, lower weight
leakproof lunch container	Reviews	Mostly meal-prep containers	Drop or separate niche
lunch bag	Synonym	Mostly soft bags	Keep only if bags are in scope
lunch boxes	Plural	Near-identical to seed	Merge into "lunch box"
lunch box ice pack	Search term report	Mostly complements	Drop
lunchbox	Spelling variant	Near-identical to seed	Merge

A repeatable checklist

Define the seed and the niche boundary.
Collect autocomplete suggestions from a fresh session, with modifiers and alphabet sweeps.
Pull Top Search Terms and Search Query Performance if you have Brand Registry; check Product Opportunity Explorer niches.
Run an auto campaign and harvest its search term report.
Mine competitor titles and bullets, minus brand names.
Mine reviews for problem and use-case language.
Check the department behind each query.
Validate each candidate on page one: exact, substitute, complement or irrelevant.
Merge plurals, variants and near-identical pages.
Split the tracking list from the backend field, and apply Amazon's backend rules.
Repeat monthly with fresh reports, then weight and prune as described in how many keywords to track.

Where JustRanks fits

Steps 8 and 9 are the slow part, because they mean reading many pages and comparing who ranks where. JustRanks records the whole organic first page of Amazon.com search results for more than 200,000 keywords over time, so history already exists for keywords in its database, and missing keywords can be added at no extra cost. Explorer lets you look up any keyword, and related keywords can be found from products that rank on several of your keywords. Keywords saved in any user's market are captured every day, all others every 48h. To understand why the same product ranks differently across these keywords, see how Amazon organic ranking works.