Amazon keyword gap analysis: build a worksheet you can act on

By Scoutvo · Published October 16, 2026 · Updated October 16, 2026 · 6 min read

Define a gap as a testable difference in text

An Amazon keyword-gap analysis compares relevant terms or phrases in a defined competitor corpus with the fields collected from your own listing. The result is a candidate set for review. It is not a list of proven lost searches, and it does not establish that inserting every term will improve sales.

This guide focuses on the comparison worksheet: which documents qualify, how a phrase is matched, what counts as missing, and how a candidate becomes an approved change. For query sources and the TF-IDF formula, use the broader Amazon keyword research guide.

Choose one question before collecting text. For example: which capacity and usage phrases appear in comparable storage-container listings but are absent from the collected title and bullets of my product? Keep backend fields and public copy separate; you cannot observe a competitor's backend search terms from its public detail page.

1. Fix the corpus and field boundaries

Use the competitor-selection method to choose close substitutes on one marketplace. Create one row per included document, with ASIN, parent where known, language, collection time, title status, bullet status, and source. Record the exact own-listing fields used for comparison.

Choose whether a document means a child or a deduplicated family. Repeated copy across children can otherwise inflate a phrase's prevalence. If one listing's bullets were unavailable, do not automatically count that listing as lacking every bullet phrase. Exclude it from comparisons that require those fields and disclose the changed denominator.

A useful manifest distinguishes collected text, valid empty text, failed retrieval, and excluded documents. Retain the original wording beside cleaned text. This lets a reviewer inspect a supposed gap without rerunning the entire project.

2. Write the matching rules before calculating

Start with consistent case, Unicode normalization, whitespace, punctuation, and units. Decide whether plural forms, spelling variants, hyphens, and language-specific compounds count as the same candidate. Keep an explicit mapping rather than silently merging terms until the result looks plausible.

Run two comparisons where useful: an exact normalized phrase match and a human-reviewed meaning match. “Dishwasher safe” and “suitable for dishwasher cleaning” can express related meanings; “hand wash only” does not. Do not remove negation and turn a limitation into a positive feature.

Store the match type with each row. A missing exact phrase may already be covered semantically. A word occurring somewhere in a listing may still fail to explain the capability clearly. Both cases matter for copy review, but neither proves Amazon's indexing behavior.

3. Build the candidate set and calculate coverage

Let C be the approved vocabulary extracted from comparable competitor documents, and O the matched vocabulary in the collected own-listing fields. The initial candidate set is G = C − O. This is a text operation, not a demand model.

For each candidate, calculate coverage = matching eligible documents / eligible documents × 100. Count a document once regardless of repetitions. Keep exact-match coverage separate from reviewed semantic coverage and record the field boundary for each.

The table describes a hypothetical container category. It is not current Amazon data or a Scoutvo result. Six of eight documents equals 75% coverage; it does not mean 75% of shoppers search for the phrase. If only six documents had usable bullets, coverage for a bullet-dependent comparison would need that denominator instead.

4. Apply a relevance and evidence gate

Review each row against the product facts before considering frequency. A phrase describing a competitor's material, size, or tested performance can be common and still be wrong for your product. Competitor wording never substantiates your own claim.

• Relevant and supported: retain the candidate with its specification or test source.
• Relevant but unverified: assign a product check; do not publish the claim yet.
• Not applicable: exclude with the product reason.
• Already explained differently: classify it as a phrasing review rather than an absent capability.
• Brand or restricted wording: review applicable rights and content rules before any use.

Keep high coverage as a descriptive signal. There is no universal coverage band that guarantees a useful keyword opportunity. A rare phrase can explain an important verified use; a frequent phrase can be redundant or misleading.

5. Add query evidence without changing its meaning

Where available, use Amazon Brand Analytics or authorized advertising reports to check customer language. Keep the report period, marketplace, source, and metric definition with the phrase. Advertising data covers the advertising activity represented; competitor text does not fill gaps in organic query reporting.

Label third-party search-volume values as estimates when they are estimates. An unavailable query metric means unverified demand, not zero demand. A candidate supported only by listing text may still be useful for accurate explanation, but should not be presented as a proven traffic driver.

Prioritize with transparent judgments: product relevance, strength of claim support, available query evidence, customer clarity, and implementation effort. Avoid a weighted score that gives the impression of calibrated sales potential when the inputs are subjective.

6. Turn approved rows into a focused content brief

For each approved candidate, write the intended customer question, proposed wording, destination field, supporting product fact, and reviewer. Make the placement fit the information. A core identity term may belong in a title; a care restriction needs a clear explanation in the appropriate content or attribute.

Amazon's keyword guidance supports readable relevant wording. Check the current search-term rules for backend fields in your marketplace. Public content and backend fields have different roles; avoid promising identical indexing across all content surfaces.

Preserve the original copy and change only approved fields. Use the listing optimization workflow for submission, child checks, imagery, and claim verification. Do not optimize toward an arbitrary keyword count.

7. Close each gap with a verification status

Confirm the saved value and rendered child after processing. Mark the row as published and checked, rejected, pending product evidence, or unresolved. A phrase becoming visible closes a text gap; it does not confirm indexing or a performance gain.

Record any search observation with query, time, location, device context, and ad status. Evaluate suitable account metrics or an eligible experiment separately. Note price, inventory, advertising, and offer changes. If the result remains inconclusive, retain that conclusion rather than assigning invented revenue to the candidate.

Refresh the corpus when the product or competitive set changes. Compare manifests between versions so a change in coverage is not mistaken for market movement when it came from different products or missing fields.