Amazon Ads keyword strategy: choose reader intent before keyword volume
A useful keyword strategy gives each target a reason to exist, observes the traffic it matches and keeps or rejects it using reader fit and economics. A long keyword list is not a strategy. A phrase can be relevant to books generally but wrong for your edition, audience or price.
This guide focuses on Sponsored Products book campaigns. For the initial setup, use the beginner walkthrough. Pair targeting decisions with bid limits and a campaign allowance; useful language alone cannot make unaffordable clicks sustainable.
Write a reader-fit brief
Fictional example: "An illustrated beginner guide to growing herbs in balcony containers. It covers small outdoor spaces, not commercial farming, houseplant diagnosis or advanced greenhouse systems."
That sentence produces both target ideas and exclusions. For fiction, replace task and skill level with genre, subgenre, tone and central reading experience. For puzzle books, distinguish difficulty, age suitability, print format and actual puzzle type. Keep the brief grounded in the book, not in a category you wish it occupied.
1. Build a small, explainable seed list
Use your table of contents, reader problem, comparable books and existing campaign reports. Inspect Amazon shopping suggestions as possible language clues, not verified search-volume data. A keyword tool's estimate is also an estimate; do not publish invented demand figures for a phrase.
| Intent group | Possible seed | Check before testing |
|---|---|---|
| Task | grow herbs in containers | The contents actually explain this process |
| Setting | balcony herb gardening | The advice applies to small outdoor spaces |
| Experience | container gardening for beginners | The book supplies beginner instructions |
| Comparable reading | A genuinely similar gardening title or author | Shared audience, not just a famous name |
| Wrong audience | commercial greenhouse management | Exclude from the seed list: the book does not solve this task |
Keep comparables defensible
Read the description and sample of a comparable title before using it as a targeting idea. A subject overlap is not enough if the reader's skill level or desired format is different. Selecting a competitor-related target does not authorize copying its cover, claiming endorsement or placing misleading branding in your creative.
Separate advertising from KDP metadata
Advertising targets are not the same thing as KDP's book metadata fields. Adding a phrase to a campaign does not put it into the book's metadata, guarantee indexing or buy an organic ranking position. Maintain the two workflows separately; see KDP keyword rules before editing publishing metadata.
2. Use match types to answer different questions
Amazon's match-type help distinguishes manual broad, phrase and exact matching from automatic targeting groups. Exact matching still allows specified close variants; it is not a promise that every report row will reproduce your typed phrase literally.
| Approach | Useful question | Main risk |
|---|---|---|
| Broad | Which related language deserves investigation? | Paying for meanings outside the book's promise |
| Phrase | Do more constrained phrase-based contexts fit? | Assuming every longer match has the same intent |
| Exact | Can a selected query work under tighter control? | Assuming tighter matching guarantees sales |
| Automatic discovery | What relevant contexts does Amazon find? | Treating automatic groups as keyword match types |
| Product targeting | Which comparable book contexts work? | Mixing ASINs with typed queries in analysis |
The Sponsored Products targeting guide covers automatic, manual and negative targeting. Use its controls selectively. You do not need every approach active on day one, and a small budget should not be divided across dozens of nearly empty experiments.
3. Structure the test so a result is interpretable
Group by a real difference
Separate materially different editions, countries or reader intents when they need different economics or decisions. Match-type separation can help control a test, but it is not a universal requirement to create a campaign for every word. Add structure only when you can supervise it and explain what decision it enables.
A simple worksheet can contain the seed, rationale, country, edition, targeting method, bid, allowance, start date and review date. Keep a separate status for "idea", "testing", "controlled follow-up" and "excluded". Preserve the reason for rejection so the same bad idea is not reintroduced by a future tool export.
Discovery and controlled follow-up
When a relevant search term shows promising matured results, consider testing it as a tighter manual target with its own affordable bid. Confirm that the new target is eligible and can deliver before excluding its source traffic. Copying a term does not transfer its performance history or prove the same result will repeat.
Overlapping campaigns complicate interpretation. Do not state that every overlap automatically raises your own auction price, or that applying an exact negative creates a perfect routing system across all contexts. Inspect actual delivery and preserve a reversible change log.
4. Read the search-term report as evidence, not as a complete demand map
Amazon's search-term reporting help says the report includes terms with at least one ad click. Rows may also represent ASIN contexts or inferred terms in non-search contexts. Consequently, its impressions need not equal campaign-manager impressions, and an absent phrase is not proof that nobody searches for it.
Record the report period and download date. Keep edition, country and currency consistent. Allow recent interactions to mature under the applicable attribution rules; the mechanics guide explains why early sales totals can change. Evaluate purchases and expected author contribution separately from attributed retail sales.
5. Turn fictional report rows into cautious decisions
Assume each one-unit purchase in this example contributes $3 before advertising and other overhead, with no KU income included. These rows are invented teaching examples, not campaign results or performance benchmarks. Assume the observation period is sufficiently matured for the exercise.
| Matched context | Clicks / spend / purchases | Contribution after ads | Possible action |
|---|---|---|---|
| balcony herb gardening | 20 / $4.80 / 3 | $9 - $4.80 = $4.20 | Relevant; consider a controlled follow-up, not instant scaling |
| indoor houseplant pests | 8 / $2.40 / 0 | $0 - $2.40 = -$2.40 | Wrong task; inspect exclusion scope |
| container gardening guide | 18 / $5.40 / 0 | $0 - $5.40 = -$5.40 | Relevant but uneconomic so far; pause or reduce exposure within the allowance |
| herb pots for beginners | 3 / $0.60 / 1 | $3 - $0.60 = $2.40 | Too little evidence for an aggressive increase |
The first row has 15% observed purchase conversion and $0.24 average CPC, but only twenty clicks. It supports a bounded follow-up, not a claim of stable 15% conversion. The third row can be paused because your allowance is limited even though it does not prove the phrase can never work.
Do not turn a click count into a universal rule
In a simplified independent-click model with a true 5% conversion probability, the chance of zero purchases after 20 clicks is 0.95 to the power of 20, approximately 35.8%. The assumptions are restrictive, but the example shows why twenty clicks do not guarantee an order. Relevance, cost and uncertainty matter together.
6. Apply negatives at the narrowest useful scope
A clearly unrelated subject can justify exclusion without waiting for an arbitrary number of paid clicks. A relevant but underperforming term calls for a different decision: inspect maturity, economics and the remaining allowance before blocking an entire family of searches.
Exact exclusion versus phrase exclusion
A negative exact target is generally the narrower choice; a negative phrase can block a broader set of queries containing that phrase or permitted variants. Check the current help and whether the setting is campaign-wide or ad-group-specific. Review the impact before applying a shared exclusion to multiple books.
For the balcony guide, excluding an unrelated houseplant-pest query is different from excluding the word "indoor" everywhere. The latter could remove useful apartment-related research. Keep the excluded phrase, match type, scope, reason and date. Do not copy a generic negative list without checking every entry against your book.
7. Use a repeatable maintenance routine
- Export current reports and retain the prior version.
- Mark obviously wrong reader intents separately from relevant weak performance.
- Calculate contribution, not just sales or CTR.
- Identify a few controlled follow-up candidates.
- Check delivery before changing source exclusions.
- Record bid and negative changes, then observe a comparable period.
Keep fewer, better-understood targets rather than expanding a list simply because a tool can generate it. Revisit the reader brief after a substantive book revision or price change. A term that worked for one edition or country is not automatically validated for another.
Continue with the bid-adjustment guide and budget allocation. Use the Ads hub for the wider cluster.
Sources and scope
Official Amazon Ads documentation checked: 14 October 2026. Source-linked statements describe targeting and report behavior. Examples, worksheet structure and decision routines are original editorial aids, not reported results. Product availability and matching rules can change; confirm the current account help before applying exclusions.