How do I do keyword research for an Amazon listing?

Start with the queries you already appear for

The best keyword list for a listing is mostly already inside the account, and it is specific to that product rather than to its category. Three reports hold it. Search Query Performance, in Brand Analytics where you have Brand Registry, gives the queries your ASIN was shown for and splits each one into impressions, clicks and purchases, with your share of each stage. The Sponsored Products search-term report gives the queries that actually triggered your ads and what those queries did once the shopper clicked. The Business Report gives the listing's own sessions and unit session percentage, which is the baseline every later comparison is read against.

Put those before any third-party tool, and be clear about why. Amazon does not publish keyword search volume, so every tool's volume number is a model — built from panels, scraped rank movements and inference — and the estimates disagree with each other, most of all on the long tail where the useful phrases live. That does not make tools useless; it makes them discovery instruments rather than decision instruments. Use them to find phrases you had not thought of, and your own reports to decide which phrases get space on the page.

The first pass is mechanical. Export the queries, sort them by purchases, and split them into three groups: queries that convert, queries that take clicks and produce nothing, and queries you are shown for but rarely clicked. The first group is where rank is cheapest to gain, because Amazon already has evidence you satisfy that query. The second group is a page-and-product mismatch, which keyword work cannot repair. The third is usually a main-image or price problem showing up in the keyword data, and reading it as a keyword problem is how a month gets spent on the wrong field.

Where the rest of the list comes from

Amazon's own autocomplete is the cheapest source of real query strings. Type the root word in the search bar of the marketplace you are working on and read the completions, then repeat with a letter appended to pull different branches, and collect the related-search suggestions on the results page. These are strings shoppers type, not phrases a tool composed. Two cautions make them usable: suggestions are marketplace-specific and shift with season and demand, so date what you collect and gather it in a clean browser session rather than one carrying your own search history.

The category itself is the second source, and the most neglected. The filters shoppers narrow by in the left-hand rail — size, material, compatibility, count, age range, certification — are simultaneously query vocabulary and structured attribute fields on your listing. A phrase that appears as a filter is worth having in your copy and worth filling in as an attribute, because the attribute is what makes the listing eligible for the filtered view in the first place.

Competitor listings extend the vocabulary further. Read the titles and bullets of the products that hold the placements you want, and, if you use a reverse-ASIN tool, treat its output as a candidate list rather than a target list. You are taking their vocabulary, never their claims: copying a compliance-risky phrase or a differentiator you cannot substantiate transfers their problem to your listing.

The last source is customer language, and it is the one that produces phrases no tool suggests. Reviews, stated return reasons and the customer questions thread show the words buyers use for the problem your product solves, which are frequently not the industry term the listing was written in. Where a recurring question uses a word your copy never contains, you have found both a keyword and a conversion fix in the same sentence.

Decide which phrases are worth owning

A list of three hundred phrases is not a plan, and the filtering is where the research earns its keep. The first filter is relevance, judged strictly: would a shopper who typed this query be satisfied by this exact product, in this size, for this use? If the answer is no, the phrase costs you twice over. It wins impressions you do not convert, which pushes click-through and conversion down, and both of those are inputs to where Amazon ranks you for the queries you do deserve.

The second filter is whether the phrase is winnable in the period you are planning. Search it and read the top of the result page as a shopper would: the price band, the review counts, whether one brand holds several slots, and whether the products shown are even the same form factor as yours. A query whose first screen is filled by products at a different price point, or with an order of magnitude more reviews, is not this quarter's target however good the volume estimate looks. Note it, and come back when the listing has the sales history to compete there.

Specificity is what makes the shortlist work. Longer, more specific queries carry clearer intent, convert at higher rates and face fewer established competitors, and the arithmetic of whether one is worth pursuing is reproducible from your own numbers: the most you can afford per click on a phrase is its selling price multiplied by the conversion rate you get on it multiplied by your break-even ACOS. A phrase where that figure sits comfortably above what the clicks cost is worth both ad budget and listing space; a head term where it does not is a phrase you can rank for and still lose money on.

Do the same filtering for the organic side, because rank follows evidence. A specific phrase you convert well on accumulates the click-through and sales-velocity signals that move you up, and those gains then carry into broader queries. Starting at the broad end instead means competing on signals you have not earned yet, which is the slowest available route.

Map each phrase to one place, once

Once the list is short, each phrase gets exactly one home. The handful you most want goes in the title, where readability constrains how many you can carry. Phrases that need context go in the bullets, as sentences a shopper benefits from rather than a comma-separated pile. Words you need indexed but would not want a shopper to read — synonyms, alternative product names, informal terms — go in the backend search terms field. Anything that corresponds to a category attribute goes in the attribute, which looks clerical and is the part most sellers leave half-filled.

Repetition does not compound. Once a word is indexed on the listing, adding it again does not increase relevance; it consumes space and makes the copy worse. Nor do you need a phrase to appear as an unbroken block, because Amazon can match a query from words present across different fields. Both facts shorten the work considerably: the exercise is coverage of distinct words, not density of favourite phrases.

Two categories stay off the list whatever the volume estimate says. Competitor brand names and trademarks are against Amazon's policy, expose the listing to complaints, and do not produce the ranking the seller is hoping for. Claims you could not defend to a reviewer — superlatives, medical or safety language, wording restricted in your category or marketplace — belong nowhere on the listing, and a keyword tool has no idea which words those are in your category.

Test the list instead of trusting it

Advertising is the fastest honest test of a keyword hypothesis, and it is usually cheaper than rebuilding a listing around a guess. Put the candidate phrases into exact match with a bid you are willing to pay, leave them alone, and read the funnel: impressions, then clicks, then orders. A phrase that takes clicks and produces no orders has told you that the intent behind it does not match the product or the page, which is worth knowing before the title is rewritten around it. A phrase that gets no impressions at a sensible bid is telling you the demand or the eligibility is not there.

Give the test enough clicks that a zero means something. As a rule of thumb rather than a rule, a listing converting somewhere around one click in ten needs a few dozen clicks on a phrase before an absence of orders is evidence rather than noise, and a handful of clicks proves nothing in either direction. Decide the number before you start, write it down with the date, and read the result then instead of checking it daily and killing phrases on the first bad afternoon.

Then confirm what you added actually indexed, per marketplace, by searching the exact phrase together with your ASIN. An edit that Seller Central accepted is not the same as an edit that is live and indexed, and the backend field in particular has no shopper-visible proof it saved. Allow for recrawl time before concluding a phrase failed.

Finally, treat the list as a living document on a cadence rather than a launch task. New query strings appear in Search Query Performance as the product's history grows, vocabulary shifts with season and with new entrants to the category, and a phrase that was unwinnable last quarter can be winnable after the listing has sales behind it. A dated list, with what was added where and what the test showed, is what turns six months of keyword work into something cumulative instead of six separate guesses.