Single-word keywords are common in publisher metadata. In a separate Kadaxis review of 2.4 million publisher keywords, 44.57% of keywords were comprised of only one word.
The appeal is easy to understand. A publisher can fit many terms such as running, memoir, and Kenya into a limited keyword field.
The theory goes that Amazon might then add ordinary book wording, use the book's category, or combine separate keywords into unspecified searches or “expanded keywords” such as:
running booksrunning memoirKenya running memoirrunning sports books
Explicitly spelling out search phrases such as these takes up more space, but it also tells Amazon which reader search the book is meant to answer.
Rather than rely on anecdotal evidence or speculative advice, we set out to empirically uncover which approach—single-word keywords or search phrases—gives books more meaningful visibility in Amazon search. To do that, we compared books using keyword lists made mostly from single words with closely matched books using natural reader-search phrases. We compared only the first 500 keyword characters for each book, so the two approaches were judged similarly.
We also addressed the obvious commercial question: broad, shorter keywords may attract more searches than longer, more specific phrases. Does that extra demand make single-word keywords the better use of a limited field?
The findings were clear:
Search phrases ranked their books more often than single-word keywords, including when those single words were matched in expanded form. Some short-tail keywords may carry higher search demand, but their low probability of ranking a book almost always makes them a poor trade-off for limited keyword space.
Therefore, our practical recommendation is:
Use search phrases as the default. Add a single-word keyword only when there is a specific, evidence-based reason to believe it is worth the space.
In this article, a visible search means that the correct book, or another edition of the same title, appeared within the first 24 organic results—roughly the first two Amazon search pages. Unless stated otherwise, results are normalized to the first 500 keyword characters.
The short version
- Search phrases worked far more often when searched exactly as supplied in ONIX. Books appeared for 41.8% of phrase searches, compared with 2.3% for mostly single-word keywords.
- Fitting more keywords into the field did not create more visibility. About 50 mostly single-word entries fit into the first 500 characters, compared with about 20 phrases. Those 50 entries produced about 1.2 visible exact searches; the 20 phrases produced about 8.2.
- Phrases also produced substantially more visibility after expansion. Across a balanced 250-pair analysis, expanded parts and combinations from phrase-led records produced about 23.8 visible searches per 500 characters, compared with 8.4 for mostly single-word records. The advantage held in both fiction and nonfiction: about +13.3 searches per 500 characters in fiction and +18.4 in nonfiction.
- Across the complete keyword record, phrases remained far ahead. Exact wording plus all measured expansion searches produced about 30.5 visible searches per 500 characters for phrase-led records, compared with 8.5 for mostly single-word records.
- Broad demand matters, but it must be reachable. A short keyword may be searched more often, but that demand is useful only if the book can actually appear for the search and the query is relevant to the reader.
What happened when we searched the keywords exactly as they were defined in ONIX?
We began with the most direct test: searching each keyword exactly as it was found in ONIX.
Within the first 500 characters, books using mostly single-word keywords contained about 50 entries. Books using search phrases contained about 20.
The phrases returned the books far more often:
| Exact keywords as supplied | Mostly single-word keywords | Search phrases |
|---|---|---|
| Keyword searches tested per 500 characters | 49.9 | 19.6 |
| Searches that returned the book | 1.2 | 8.2 |
| How often a tested search returned the book | 2.3% | 41.8% |
| Books with at least one top-10 result | 24.4% | 89.2% |
| Books with at least one first-two-page result | 36.0% | 95.2% |
The book using phrases produced more exact-search visibility in 224 of the 250 matched pairs. The mostly single-word book was higher in 16 pairs, and ten pairs tied.
More keyword entries fitted into the mostly single-word record. The phrase record made much more effective use of the same 500-character comparison window.
A phrase can state a complete relationship: subject plus audience, setting, genre, method, tone, problem, activity, format, or desired outcome. A single word names only one part of that request and leaves the relationship to be inferred elsewhere.
Does expansion change the answer?
No. Single-word keywords can participate in additional searches, but the phrase-led records produced much more expanded visibility overall.
We tested a broad but fixed set of ways each keyword record could lead to additional searches. These included:
- individual words;
- ordinary book wording such as
booksorbooks about; - useful two-word parts already contained within a keyword;
- combining words from different keywords; and
- adding the book's category to the keyword.
The same rules were applied across the two approaches. Duplicate searches were removed. For the conservative comparison, we also removed results that could already be explained directly by the title, subtitle, author, series, or category.
Across the balanced 250-pair analysis, the result was:
| Expansion searches, excluding the exact supplied keyword | Mostly single-word records | Phrase-led records |
|---|---|---|
| Visible searches per 500 keyword characters | 8.4 | 23.8 |
| Phrase advantage | +15.3 |
The phrase approach therefore produced about 2.8 times as much observed expanded visibility.
This is the strategy-level answer to the expansion question. Amazon sometimes created additional visibility from single words. But when we counted the full range of expansion available to both strategies, phrases still produced much more.
Where did the phrase expansion come from?
Two sources were especially important.
Useful two-word parts inside phrases
A phrase such as:
FBI procedural thrillercontains shorter searches such as:
FBI proceduralprocedural thriller
These are not random fragments. They preserve part of the relationship expressed by the complete phrase.
Across the balanced 250-pair analysis, these adjacent two-word searches produced:
- about 8.5 visible searches per 500 characters for phrase-led records;
- about 0.3 for mostly single-word records; and
- a phrase advantage of about 8.2.
The phrase-led book was higher in 198 pairs. The mostly single-word book was higher in ten, and 42 pairs tied.
Combinations across separate keyword entries
We also created controlled combinations from different keyword entries. Phrase-led records again produced more visibility:
- about 15.6 visible searches per 500 characters for phrase-led records;
- about 7.0 for mostly single-word records; and
- a phrase advantage of about 8.5.
The phrase-led book was higher in 176 matched pairs, the mostly single-word book in 66, with eight ties.
These two views should not be added together because some searches overlap. They show why the net expansion result favored phrases: phrases preserve useful relationships within each entry and also create a more productive set of relationships across the wider keyword record.
What about ordinary book wording and categories?
Amazon sometimes returned books when ordinary product wording was added, for example:
running booksbooks about runningrunning memoir books
Those forms were included in the wider expansion analysis. They helped in some cases, but they did not reverse the overall result.
Categories also mattered. In a separate controlled comparison, the correct category helped far more often than an unrelated category. That tells us that relevant category information can support retrieval.
But a category is already part of the book's metadata. We therefore did not treat a category-assisted result as independent proof that the keyword alone created the visibility. Our practical conclusion is:
Assign accurate categories for the context they provide. Do not assume a category term will combine with your keywords to match a reader's query. Write the full phrase into the keyword field.
Are phrases only useful for longer, highly specific searches?
No.
A common concern is that phrases may rank only for narrow searches with very little demand, while single-word keywords provide broader short-search coverage.
The data did not support that as a general strategy-level conclusion.
In our comparison of nonfiction titles, we looked at all visible one- and two-word searches produced by each metadata record. This included individual words, two-word parts of phrases, and combinations across keywords.
| All one- and two-word expansion searches in nonfiction | Mostly single-word records | Phrase-led records |
|---|---|---|
| Visible searches per 500 characters | 13.8 | 32.2 |
Phrase-led records therefore produced more than twice as much one- and two-word visibility, even before their exact three- and four-word phrases were added.
This matters because it changes the presumed tradeoff. A publisher is not choosing between:
- single words that provide short-search coverage; and
- phrases that work only for one narrow exact search.
A good phrase can do both. It can state a specific reader request and contain shorter relationships that remain useful on their own.
What happened across the complete keyword record?
The most practical comparison is the whole record: the exact keywords as supplied, plus every measured expansion.
Across the balanced 250-pair analysis:
| Complete keyword record | Mostly single-word records | Phrase-led records |
|---|---|---|
| Visible searches per 500 characters | 8.5 | 30.5 |
| Phrase advantage | +22.0 |
The phrase-led records produced approximately 3.6 times the observed search footprint. Even where broad single-word searches may carry more demand, phrases produced far more reachable visibility across both direct and expanded searches, more than offsetting the theoretical advantage of targeting broader terms.
At the book-pair level:
- the phrase-led book was higher in 205 of 250 pairs;
- the mostly single-word book was higher in 40; and
- five pairs tied.
This complete-record result already gives the mostly single-word approach credit for every expansion search it produced. The phrase advantage remained large after those gains were included.
Publishers allocate a complete keyword field, not one isolated expansion mechanism. Across that complete field, the phrase-led approach produced substantially more visibility.
What about search volume?
A broad word may be searched more often than a specific phrase.
romance may attract more searches than witty regency historical romance. That matters.
But demand is only one part of the equation:
Commercial search opportunity = the searches where the book appears × how often readers make those searches × where the book ranks × what readers do next.
This research measured:
- whether the correct book appeared;
- how many distinct controlled searches returned it; and
- whether it appeared in the top 3, top 10, or first 24 organic results.
It did not measure:
- exact query volume;
- reader impressions;
- click-through;
- conversion;
- sales; or
- revenue.
That means we should not claim that every successful phrase is more commercially valuable than every successful broad term.
But the opposite assumption is also unsafe. A broad keyword term has little practical value to a book if the book does not rank for a search for the keyword. The phrase-led strategy produced far more visible searches per 500 characters, including far more one- and two-word expanded keywords. A single-word strategy therefore needs a substantial demand advantage among the relatively few searches it wins in order to make up for its smaller observed search footprint.
To reach the top two pages, a book has to finish ahead of nearly every other title a query returns. Across ten query pairs, broad terms returned a median of more than 70,000 books competing for 24 places, a requirement of the top 0.03% of the field. Specific phrases returned a median of 177 for the same 24 places.
This is why single-word keywords and their expansions perform so poorly. A broad term is a lottery ticket: the prize is large and almost no one collects. Demand converts to readers only once a book ranks. A high-volume term the book never reaches returns nothing. A lower-volume phrase it does reach returns readers on every search.
The correct objective is not maximum breadth or maximum specificity.
Publishers should target reachable demand: searches that are relevant, used by readers, and competitive enough for the book to enter a visible position.
Do more specific searches express stronger reader intent?
Usually, they express more clearly defined intent.
A search for romance identifies a broad department. A search for witty regency historical romance specifies genre, period, and tone.
That does not prove the second query converts better. A reader using a broad search may still be ready to buy, and a highly specific query may have almost no demand.
But the second search gives Amazon and the publisher a clearer description of the desired book. When the title fits closely, the resulting traffic is more qualified.
Is making a keyword longer enough?
No.
The phrase-led keywords were not random strings created by adding words to a topic. They were selected to describe natural reader searches and reviewed as a portfolio.
The research therefore compares two complete strategies:
- records dominated by short topic terms; and
- records built around selected reader-search phrases.
It does not prove that any multiword keyword will work, that making a phrase longer is automatically beneficial, or that punctuation and spacing alone change ranking.
What should publishers do?
The evidence supports a phrase-first strategy, with single words added by exception.
1. Write the important reader searches as phrases
Use the limited field to state relationships that matter:
- subject plus audience;
- genre plus tone;
- setting plus activity;
- problem plus method;
- topic plus format; or
- experience plus desired outcome.
2. Favor phrases that contain useful two-word relationships
A strong three- or four-word phrase can serve two purposes:
- it states the complete reader request; and
- it contains shorter adjacent searches that preserve part of the meaning.
Do not add filler merely to increase word count.
3. Do not reserve a default percentage for single words
The data does not support a universal 10%, 20%, or 30% allocation.
A single-word keyword should be retained only when there is a clear reason, which is typically a combination of:
- the book already appearing for it;
- the search having meaningful demand;
- realistic competition;
- a distinctive keyword term; or
- creation of a valuable search not already covered by the phrase portfolio.
4. Test the exact wording and the wider search footprint
After the keywords have been indexed:
- search the exact phrase;
- search useful two-word parts;
- test a small number of sensible combinations across entries;
- record top-3, top-10, and first-two-page visibility separately; and
- evaluate demand and commercial results as a separate step.
5. Treat expansion as additional coverage, not the plan
Amazon sometimes returned books for ordinary book wording, categories, shorter parts, and combinations. But the full analysis still favored phrases by a wide margin.
When a search matters, write it explicitly.
How did we make the comparison fair?
We did not compare an arbitrary group of highly visible books with an arbitrary group of books that were difficult to find.
We created 250 book pairs. In each pair:
- one book used keywords made mostly from single words;
- the other used natural reader-search phrases;
- the books were in a similar broad category;
- they were published around a similar time;
- they had a similar number of reader ratings;
- they had similar keyword character counts; and
- where possible, they used the same format, such as paperback with paperback.
The books were paired before we looked at their Amazon search results. We then analyzed the first 500 keyword characters for every book, giving both approaches the same comparison window.
For the headline result, every exact keyword search was checked twice. A search counted only when the correct book, or another edition of the same title, appeared within the first 24 organic results in both checks.
The books were similar, not identical. Matching reduces obvious differences, but it cannot remove every difference in audience, competition, sales history, author recognition, or other book-level factors. The study compares two complete keyword strategies; it does not prove that word count or formatting alone caused the result.
How did we run the research?
The article draws on several connected tests. Each answers a different part of the question.
The repeated exact-keyword comparison
- 250 matched book pairs, or 500 books in total;
- one book in each pair used mostly single-word keywords and the other used search phrases;
- the first 500 keyword characters were compared for every book;
- books were paired before Amazon search results were examined;
- every exact search was run twice;
- a result counted only when the correct title appeared within the first 24 organic results in both checks; and
- all keyword records were checked after collection and had not changed.
This is the basis for the 41.8% versus 2.3% exact-search result.
The wider expansion analysis
To understand the complete search footprint, we examined a balanced set of 250 matched pairs: 150 fiction pairs and 100 nonfiction pairs. We tested:
- individual words;
- useful adjacent two-word parts already present inside keywords;
- ordinary product and relationship wording;
- limited combinations across different keyword entries; and
- category-supported forms.
Each distinct book-search combination was counted once. We removed duplicates and excluded results that could already be explained directly by the title, subtitle, author, series, or category.
The broader analysis used one complete observation per unique search. It shows observed visibility at that time; it does not prove that every result stayed in place over a longer period.
What does the study show—and what does it not show?
The evidence supports this statement:
Among closely matched books with comparable keyword character counts, selected reader-search phrases worked far better when searched exactly as supplied and created substantially more Amazon search visibility after the full set of measured expansion searches was included. The phrase-led approach also produced more one- and two-word visibility, so its advantage was not confined to highly specific long searches.
The study does not show that:
- phrase formatting alone caused the difference;
- every search phrase has meaningful demand;
- every single-word keyword is ineffective;
- every word inside a phrase will rank independently;
- separate keywords will always combine;
- a one-time expansion result will remain stable indefinitely;
- visibility caused impressions, clicks, sales, or revenue; or
- one exact mixture of phrases and single words is best for every book.
Conclusion
Single-word keywords can create additional visibility through expansion. When we tested the expanded searches generated from both single-word and phrase-led records, the phrase-led approach still produced more visibility overall, including more one- and two-word searches. Phrases also performed far better when searched exactly as supplied.
The competition figures explain why. Broad queries returned a median of more than 70,000 books competing for 24 places; specific phrases returned 177 for the same 24. Demand converts to readers only once a book ranks, and a high-volume term the book never reaches is of no value.
Search phrases made substantially better use of the available keyword space than records built mostly from single words. Publishers should not fill keyword records with isolated words because more of them fit, or reserve a fixed share of the field for single words in the hope that Amazon will expand them. Use the field to state the reader searches the book should answer, and add a single word only when its demand and ranking opportunity justify the phrase it displaces.
Research note: This article reports observed Amazon Books search visibility for matched books. Results were normalized to 500 keyword characters and conservatively removed cases directly explained by title, subtitle, author, series, or category wording. Generated component searches are controlled tests, not claims that every query has meaningful reader demand.