Amazon keyword tracking: how many keywords you really need

Why a handful of keywords misleads, how to build, weight and prune a keyword set per product, and what statistics say about how many is enough.

Most sellers track the keyword they think of first, plus a few close variants. The result is a rank chart that jumps around, a "position" that means less than it seems, and blind spots exactly where competitors are gaining. This guide explains why that happens, how to build a keyword set that reflects your real market, and how many keywords it takes before an average becomes trustworthy.

The short answer

There is no universal number, but there is a logic:

Too few keywords (under ~10) give you a noisy, lopsided picture of a market that mostly lives in the long tail.
A deliberate set of a few dozen relevant, weighted keywords per product is where averages become stable for most niches.
Beyond that, extra keywords help only if they add new demand. Near-duplicates add cost and clutter, not information.

The rest of this article shows where those ranges come from.

Why one keyword (or five) misleads

Search demand is not concentrated in a few big terms. Research on web search logs found that "a large fraction of queries submitted to Web search engines occur very infrequently" (Downey, Dumais and Horvitz, SIGIR 2007). Amazon's own researchers describe the same split in product search: engines perform well on frequent "head" queries with rich behavioral data, while rare "tail" queries lack that data (Hiranandani et al., Amazon Science, 2019).

How much of a category sits in the tail varies a lot. One agency analysis of estimated Amazon US search volumes (PMG, September 2024 data) found that the top 100 search terms covered about 75% of volume in video projectors, about 54% in automotive lights and accessories, and only about 15% in women's clothing. These are third-party estimates, but the direction is clear: in many categories, your biggest keyword is a small slice of the demand that decides your sales.

Tracking one head term causes three problems:

Coverage. You see where you rank for one query, not for the thousands of shoppers who phrase the need differently.
Noise. A single keyword's rank moves for reasons unrelated to your listing's strength, so one data point can't separate a trend from a blip.
Blind spots. If a competitor is gaining on mid-tail and use-case queries, a head-term-only view won't show it until it reaches the head term, if it ever does.

What Amazon's own reports can and can't tell you

Brand Analytics is useful for building a keyword list, but it does not replace tracking:

Search Query Performance lists the top queries that led customers to your products, with impressions, clicks, cart adds and purchases. Queries where you never appeared are not in it, and access requires Brand Registry.
Top Search Terms shows a search frequency rank, not a search volume, plus the three most-clicked products for each term.
The Sponsored Products search term report shows which shopper queries your ads matched and what they generated. It reflects your bids and match types, not organic demand.

All three describe your own funnel or a slice of the market. None shows how the whole first page of a search changes from day to day, or who moved into the spots you used to hold.

Build the set in layers

Think of a keyword set as a map of how shoppers describe the need your product solves. Build it in layers, all generic, with no brand names:

Head terms. The two or three phrases that define the product. High volume, broad intent.
Mid-tail variants. Head term plus one qualifier: size, material, form factor.
Attribute terms. Specific features shoppers filter on, such as "double door" or "foldable".
Problem and use-case terms. What the shopper wants to achieve: travel, puppy training, small apartments.
Audience terms. Who the product is for, by size, age or situation.
Adjacent generic terms. Other names for the same solution that competitors rank for, such as "kennel" instead of "crate".

Leave out competitor brand names. They measure brand demand, not your category, and Amazon tells sellers not to put other brand names in backend search terms. The backend field is also capped at 250 bytes, so the set you track will be larger than the list you can index.

A worked example: "dog crate"

The table below is an illustrative example, not a data set. The tiers are qualitative; no search volumes are implied.

Layer	Example keywords	Suggested tier
Head	dog crate, dog kennel	Core
Mid-tail	large dog crate, metal dog crate, collapsible dog crate	Important
Attribute	double door dog crate, dog crate with divider, heavy duty dog crate	Important / Standard
Use case	dog crate for car travel, puppy training crate, dog crate for apartment	Standard
Audience	dog crate for small dogs, crate for 50 lb dog	Standard
Adjacent	pet playpen, dog cage indoor, dog crate furniture	Minor (check relevance first)

A real set for this product would add more variants per layer until new keywords stop surfacing new products on page one.

Weight keywords, don't just average them

An unweighted average treats "dog crate" and "dog crate for 50 lb dog" as equally important. If you drop five places on the long-tail term and gain two on the head term, the plain average says you got worse, while your traffic almost certainly went up.

Weight by two factors:

Demand. Bucket keywords into tiers by volume, using whatever source you trust. Coarse tiers are more robust than exact volume figures, which are estimates anyway.
Relevance. A high-volume keyword that only partly fits your product deserves a lower tier than a smaller, exact-fit keyword.

Weighting has a statistical cost worth knowing. Survey statisticians use Kish's effective sample size, (Σw)² / Σw², to measure how much unequal weights shrink a sample. Twenty keywords where one carries as much weight as the other nineteen combined behave like roughly four keywords. A set with four heavy, six medium and ten light keywords (weights 4, 2, 1) still behaves like about fifteen. Use tiers to reflect importance, but avoid a single keyword that dominates everything.

Prune what doesn't belong

Irrelevant keywords drag down averages and hide real movement. Two checks work well:

Does page one show products like yours? In a public benchmark, Amazon researchers label each result for a query as exact, substitute, complement or irrelevant (Reddy et al., 2022). If the first page of a keyword is mostly complements (crate covers, crate beds) or substitutes (playpens, gates), you are not competing head-on there.
Do the same products rank across keywords? Amazon researchers treat two queries as equivalent if they lead to purchases of the same or similar products, and search research has clustered queries by shared clicked results since at least 2000. You can't see purchases, but you can see rankings. A keyword whose first page shares many products with your core terms belongs in the set. One with little overlap is probably a different market.

Also merge near-duplicates. Amazon matches queries semantically, handling synonyms, word variants and misspellings (Nigam et al., KDD 2019), so "dog crates" and "dog crate" often return very similar pages. Tracking both adds little.

How many is enough? The statistics

Think of each keyword's rank change as signal plus keyword-specific jitter. Averaging cancels jitter: the standard error of a mean falls with the square root of the number of observations (σ/√n). Relative to a single keyword, the jitter that remains is about:

Keywords in the set	Remaining keyword-specific noise
5	~45%
20	~22%
50	~14%
100	10%

Two caveats matter. First, returns diminish: halving the noise takes four times as many keywords. Second, the math assumes independent jitter. When keywords move together, the variance of the mean levels off near the shared part instead of falling toward zero. Near-duplicates are strongly correlated, so ten variants of one phrase count as far fewer than ten keywords. Breadth across layers beats depth within one.

Practical ranges

Combining coverage, noise and correlation gives working ranges. They are reasoning, not industry benchmarks:

Narrow niche, one product type: roughly 20 to 40 keywords. Noise is already well below that of a single keyword, and most narrow niches run out of genuinely distinct, relevant queries in this range.
Broad category or many use cases: 50 to 100 keywords. Long-tail-heavy categories need the breadth to cover demand at all.
Fewer than 10: fine for a quick check, not for judging trends or competitors.

The right stopping point is empirical: stop adding keywords when new ones keep showing the same products on page one.

Refresh the set

Demand shifts with seasons, trends and new product types. Amazon's own search reports can be pulled by week, month or quarter, which makes those shifts visible. A light routine works:

Monthly: review Search Query Performance and your ad search term report for new queries that convert, and add the relevant ones.
Quarterly: re-run the relevance checks, demote or drop keywords whose page one drifted away from your product, and re-tier by demand.
When you change the product: a new size or feature opens new attribute and use-case keywords.

Keep history when you prune. Removing a keyword from the average is different from deleting what happened on it.

The per-keyword pricing trap

Many trackers charge per keyword or per keyword-slot. That pricing pushes sellers toward exactly the wrong behavior: tracking the fewest, biggest terms. The statistics above show that this produces the noisiest, least representative picture, and the long tail, where shifts stay invisible longest, becomes the first thing cut. When comparing tools, check what a properly sized set across all your products would cost, not the entry price for ten keywords.

How JustRanks handles keyword sets

JustRanks records the whole organic first page of Amazon.com search for more than 200,000 keywords over time. A Market combines up to 100 keywords per product, and each keyword gets a tier (Core, Important, Standard or Minor) that sets its weight. Related keywords can be found from products that rank on several of your keywords, and missing keywords can be added at no extra cost. Keywords saved in any user's market are captured every day, others every 48h. Pricing is one flat price rather than per keyword, ads never count toward rankings, and no Seller Central access is needed.