Keyword Research

Intent beats volume

The biggest keyword is almost never the best keyword. A high-volume head term is full of people who are still deciding what they want, and it is priced by every competitor who wants to be seen. A smaller, specific term is full of people who already know — and it is cheap because nobody bothered to look for it.

Keyword research is not list-building. It is the ongoing work of finding which searches contain buyers for your particular product, at a price you can afford, and routing your money toward them.

The three questions every keyword has to answer

Before a term deserves a bid, it has to pass three tests — and most keyword lists only ever check the first one:

• Volume — does anyone actually search this? Easy to check, and the only thing most tools measure.
• Relevance — is your product genuinely a good answer to this search? If not, you will buy clicks that cannot convert, and teach Amazon your listing disappoints.
• Economics — at the price this term clears at, can it convert often enough to pay for itself? A relevant, high-volume term is still a bad keyword if the CPC exceeds what a click is worth to you.

Where the first list comes from

Your seed list is a hypothesis, not an answer. Build it quickly from the places real demand already shows up, then let money test it:

• Your own listing — the words you already used in the title, bullets and backend terms.
• Your reviews and questions — the language customers use for the problem you solve, which is rarely the language you use.
• Competitors — the terms their listings clearly target, especially ones you have no coverage for.
• Amazon’s autocomplete and related searches — real queries, ranked by real demand.
• Amazon’s own keyword suggestions in the ad console, which are drawn from what already converts in your category.

Four kinds of keyword, doing four jobs

A list of fifty terms all doing the same job is not a keyword strategy. Amazon’s own framing splits them four ways, and the split is useful because each kind fails differently:

• Generic — short and broad (“water bottle”). Enormous volume, weak intent, expensive. Visibility, not efficiency.
• Long-tail — specific phrases (“32 oz insulated bottle with straw”). Small volume, strong intent, cheap. Your margin.
• Branded — your own name. Cheap and high-converting, but see the section below on whether you needed to buy them at all.
• Seasonal — terms that only exist for part of the year. Worth pre-loading before the window and pausing after it, rather than judging on annual averages.

Match types are a research instrument

Match types are usually taught as targeting settings. In research they are better understood as three different questions you can ask the market, at three different prices.

Broad match is wider than people expect

It is worth being precise about broad, because it surprises people who treat it as “the keyword, plus a few extra words”. Amazon matches broad keywords on meaning, not just wording: terms can appear in any order, and it will include singulars, plurals, variations, synonyms and related terms — and the keyword itself may not appear in the shopper’s query at all.

Amazon’s own example is instructive: the keyword “sneakers” can match queries like “basketball shoes”, “cleats”, “trainers” or “foam runners”. That is enormously useful for discovery and enormously expensive if you leave it unattended.

You can target products, not just searches

Keywords are not the only way to buy traffic, and research that only looks at search terms misses half the map. Product targeting places your ad on specific product detail pages and alongside those products in results — you pick the competitor, not the query.

Category targeting works the same way but wider, and can be refined by attributes such as brand, price range, star rating and Prime eligibility. That combination is the interesting part: you can target a category, then narrow to products priced above yours and rated below yours, which is a fairly precise description of a shopper who might switch.

Let discovery do the finding

A spreadsheet cannot tell you what converts. Auto and broad campaigns surface the exact phrases shoppers type — including long-tail winners with odd wording that no tool would have suggested and no human would have guessed.

Treat your early auto-campaign spend as paid research with a deliverable. The deliverable is the search-term report, and that report — not your seed list — is your real keyword list.

The harvest loop

This is the engine of the whole discipline, and it runs forever rather than being a task you complete:

• Discovery campaigns spend on loose targeting and produce search terms.
• Terms that converted repeatably get promoted into exact match, where you set the bid deliberately.
• The same term is negated in the discovery campaign, so you stop paying twice and your campaigns stop bidding against each other.
• Terms that spent real money and never converted get negated outright.

How much data before you judge a term?

The most common research mistake is not choosing badly, it is choosing early. One click and no sale tells you nothing. Neither does five.

A workable rule: judge a term on clicks, not impressions or days, and give it enough clicks that a conversion would have been likely if the term were any good. If your product converts at roughly 10%, a term with eight clicks and no sale is unremarkable — you would expect to wait about ten clicks for the first one. At thirty clicks and no sale, you have an answer.

Amazon suggests letting match-type tests run one to two weeks before pausing anything, for the same reason: you are waiting for the sample to mean something.

Negatives are half the job

Research is not only about the terms you want. Every wasteful term you remove sharpens the account and protects the budget that should be flowing to winners — and unlike bid changes, a good negative keeps paying back forever.

Work down from the obvious: searches for a different product entirely, competitor brands you cannot serve, wrong sizes or variants, informational searches from people who are reading rather than buying, and free-shopper terms that never convert in any category.

Mine the waste by pattern, not one term at a time

Negating individual search terms is endless, because shoppers phrase the same intent a hundred ways. The faster method is to look for the repeated word — the n-gram — inside your losing terms.

Split your non-converting search terms into their individual words and total the spend behind each. When one word appears across many wasteful searches and almost no converting ones, that word is the problem. Negating the root kills the whole family at once, including the variants you have not seen yet.

Do this in SellerMateBulk Actions →Harvest converting search terms and push a whole pattern of waste into negatives in a single pass — which is what makes pattern-level negation practical above a few dozen campaigns.

Should you bid on your own brand?

Branded terms look wonderful in a report. They convert at rates nothing else matches and the ACoS is often single digits, which makes them the easiest number in the account to feel good about.

They are also the place where attribution flatters you most, because a shopper searching your brand name was already looking for you. Some of those clicks are genuinely won — a competitor was advertising against your name and would have intercepted them. Some are traffic you already had, bought back at a price.

The honest answer is that both are true and the ratio depends on how contested your brand is. What settles it is not a report but a test: pause branded advertising for a defined period and watch total sales rather than ad sales.

Build the map by intent, not alphabetically

Organise what you find by the job each tier does, because each deserves a different bid and a different expectation:

• Head terms — high volume, broad intent, expensive. Visibility plays. Rarely your best ACoS, sometimes worth it anyway.
• Mid-tail — more specific, usually better converting and materially cheaper. This is where most accounts should live.
• Long-tail — precise intent, small volume, excellent efficiency. Individually tiny, collectively your margin engine.

Look outside your own account

Your search-term report only shows searches you already paid to appear in. It cannot show you the demand you are missing entirely, and that blind spot is where category share is quietly lost.

If you are brand-registered, Brand Analytics fills it in. The Search Query Performance dashboard shows the queries leading shoppers to your products with the full funnel — query volume, impressions, clicks, cart adds and purchases — alongside your brand’s share of each against the whole query on Amazon.

That share view answers a question your ad reports cannot: not “how did I do”, but “how much of this search did I get”. A term where you hold a small share of a large volume is an opportunity. A term where you already hold most of the purchases is a term to defend, not to bid harder on.

Keywords the listing cannot cash

Research and listing work are the same job seen from two ends. A term you bid on but never mention in the listing is a term you pay full price for — Amazon has less reason to consider you relevant, and the shopper who clicks finds a page that does not obviously answer their search.

When a term proves itself in the search-term report, feed it back into the listing: the title if it is a head term, the bullets if it qualifies the use case, backend if it is a variant spelling. That is how paid discovery turns into organic rank, which is the flywheel from the White Belt.

Research pitfalls to avoid

Make the loop survive a busy week

Keyword research fails on cadence rather than on insight. The analysis is not hard; doing it every week for a year is. That is precisely the part worth automating.

Set it up so harvest candidates and waste candidates arrive together on a schedule — terms that converted enough to promote, and terms that spent enough to kill — and the weekly pass becomes a short review rather than an afternoon in a spreadsheet.

Master Sifu — Put the loop in the calendar, not in your intentions. Keyword research does not fail on insight, it fails on the fourth consecutive busy week — so make the weekly pass small enough to survive one.