How Broad Should Your Initial Literature Search Be?

Should Your First Search Be Broad or Highly Specific?

You have translated your research question into searchable concepts and are ready to run the first serious search. One decision immediately appears: how much should you include?

Suppose you are interested in how generative artificial intelligence affects academic writing among undergraduate students. You could begin with several highly specific requirements: generative AI, academic writing, undergraduate students, writing self-efficacy, perhaps a particular educational setting, and maybe even a date or language restriction.

That search may look impressively precise. It may also hide useful literature before you have had a chance to learn how the field describes your topic.

At the other extreme, searching simply for artificial intelligence could return an enormous and heterogeneous body of literature that tells you little about your specific problem.

The useful starting point lies somewhere between those extremes. Your initial search should give you enough room to discover relevant terminology and literature while remaining sufficiently connected to your research question to be informative.

Start Broad Enough to Learn, Then Narrow for a Reason

Search Breadth Is a Balance Between Finding More and Filtering More

Broad does not simply mean “a lot of results”

Researchers often judge a search by the number displayed above the result list. A search producing 20,000 records feels broad. One producing 40 feels narrow.

Result count is useful information, but it does not tell the whole story.

A search is broad in a more meaningful sense when its retrieval conditions allow a wide range of records to qualify. A large result count may reflect a broad strategy, a heavily researched topic, an ambiguous search term, or a very large database. A small result count might reflect a narrow strategy, a genuinely small evidence base, unusual terminology, or a highly specialized topic.

You therefore cannot determine appropriate breadth from the number alone.

Your first search is partly an information-gathering exercise

Early searching does more than retrieve papers for reading. It can teach you how the literature talks about your topic.

You may discover:

• terminology you had not considered;
• abbreviations and acronyms used in the field;
• older and newer names for the same concept;
• relevant subject headings or indexing terms;
• adjacent concepts that repeatedly appear in useful papers;
• unexpected ambiguity in one of your search terms.

This is one reason an initial search can benefit from some breadth. If you impose every conceivable restriction before seeing the literature, you may prevent the search from showing you terminology that would have improved the strategy.

Cochrane describes search development as iterative: terms can be modified based on what has already been retrieved. Its current Handbook also recommends using a wide range of appropriate free-text terms for selected concepts and combining free-text searching with controlled vocabulary where available.

The principle extends beyond systematic reviews. Early search results are evidence about the vocabulary and structure of the literature, not merely a pile of citations waiting to be screened.

Start with the concepts most necessary to identify the topic

If you have already translated your research question into searchable concepts, you may have several candidate concept blocks. That does not mean they all need to appear in your first search.

Consider this hypothetical question:

How does generative AI use for academic writing affect writing self-efficacy among first-year undergraduate students?

Candidate concepts might include:

• generative AI;
• academic writing;
• writing self-efficacy;
• undergraduate students;
• first-year students.

An initial search that requires all five may be unnecessarily restrictive. A more exploratory starting point might test the central topical concepts first, then examine whether population or outcome concepts need to be added.

The exact structure depends on the topic. The point is to avoid confusing the detail of the research question with the number of concepts that the database must initially require.

Every additional required concept changes what can survive the search

When concept blocks are combined with AND, adding another block generally reduces retrieval because a record must satisfy another condition.

This is why decisions about which concepts should actually become search terms matter so much. A detail can be central to the research question without necessarily belonging in the initial query.

Broad searching usually favors sensitivity; narrow searching usually favors precision

Two concepts are particularly useful for understanding search breadth: sensitivity, also called recall, and precision.

Cochrane defines sensitivity as the proportion of relevant reports retrieved from all relevant reports available in the resource, while precision is the proportion of retrieved reports that are relevant. Increasing search comprehensiveness will commonly lower precision because more irrelevant records are retrieved along with relevant ones.

This trade-off helps explain why there is no single ideal level of breadth.

The purpose of your search changes how much breadth you need

A student trying to find several strong sources for an assignment does not necessarily need the same retrieval strategy as a team conducting a systematic review.

For Cochrane intervention reviews, searches should seek to maximize sensitivity while striving for reasonable precision. This means retrieving irrelevant records is not necessarily evidence that the search is poor. Some irrelevant retrieval may be an acceptable cost of reducing the risk of missing eligible studies.

That principle should not be applied mechanically to every literature search. Search objectives differ, and so does the cost of missing relevant evidence.

A broad initial search should still have conceptual boundaries

“Start broad” can be misunderstood as “search vaguely.” Those are not the same thing.

Searching education when your question concerns generative AI and academic writing is not strategically broad. It is simply disconnected from the information need.

A useful broad search retains the central concepts while initially avoiding unnecessary restrictions.

The distinction matters because broad searching should create useful discovery, not indiscriminate retrieval.

Do not narrow simply by adding every available filter

Databases often offer convenient filters for publication year, language, document type, subject area, population, access status, and other characteristics.

Those controls can be useful. They can also exclude literature.

A restriction should therefore have a reason connected to the research question, eligibility criteria, or search purpose. It should not be applied merely because the interface makes it easy.

Questions such as what date range the search should use, whether to restrict the search by language, and whether to search only peer-reviewed literature deserve separate methodological decisions.

For systematic reviews, restrictions can have methodological consequences. Cochrane advises that searches should capture as many eligible studies as possible and that restrictions such as publication date and format should be justified. Its guidance also cautions against language restrictions because they can exclude relevant records and potentially introduce bias.

Use known relevant papers as a diagnostic check

If you already know several papers that clearly belong in the literature you are seeking, ask whether your developing search retrieves them.

This is a practical diagnostic test.

If a supposedly broad initial strategy fails to retrieve obvious key publications that are indexed in the database, investigate why. Perhaps an essential synonym is missing. Perhaps phrase searching is too restrictive. Perhaps a concept block is excluding the paper. Perhaps the database uses different indexing terminology.

Cochrane specifically recommends checking whether a developing search finds key publications or studies included in similar reviews. It also cautions that retrieving only the papers you already know is not sufficient, since a strategy can become inadvertently biased toward known records.

Look at what the irrelevant results are telling you

Irrelevant records are not always useless. They can reveal why a search is broad.

Suppose your search retrieves hundreds of papers about generative AI in software engineering when your interest is academic writing. That pattern tells you that the technology concept is working, but the application context may need stronger representation.

Suppose instead that most irrelevant records concern writing by professional authors rather than students. The population concept may need attention.

This is more informative than simply reacting to a large result count.

Before narrowing, inspect a sample and ask:

• Which term is bringing these records in?
• Which important concept is absent?
• Is one term ambiguous?
• Would a phrase, field restriction, subject heading, or additional concept address the problem?
• Would the proposed change also remove relevant records?

That final question is essential. A refinement that removes irrelevant material can look successful until you notice that it also removed useful studies.

Narrow incrementally so you know what changed

When a search needs refinement, change one meaningful element at a time where practical.

If you simultaneously add a population block, an outcome, a five-year date limit, English-language restriction, and a document-type filter, the result count may fall dramatically. You will have little idea which change was responsible or which relevant records were lost.

Incremental refinement makes the search easier to diagnose.

This iterative approach gives you something more valuable than a smaller number: an explanation for why the search improved.

There is no magic number of search results

Researchers sometimes look for thresholds: Is 500 results too many? Should I aim for 100? Is 20 too few?

No universal threshold can answer those questions.

Five hundred records could be quite manageable and appropriately sensitive for one project. Fifty could be suspiciously restrictive for another. A systematic review may legitimately screen thousands of records, while a focused exploratory search may achieve its purpose with far fewer.

The meaningful question is not “How many results should I have?” It is “What does this result set tell me about the performance of my search relative to my purpose?”

The next diagnostic step is therefore to learn how to recognize when a search has become too broad, rather than judging breadth from an arbitrary result count.

Refining an Initial Search Without Jumping Straight to a Narrow Query

The initial broad search served a purpose: it revealed terminology and showed how the literature represented the population and outcome. Had the researcher required every detail from the beginning, those patterns might have remained invisible.

If the question itself contains many interacting elements, it can also help to break the complex question into searchable concepts before deciding which ones belong in the initial strategy.

Common Mistakes When Deciding How Broad to Search

Let the Results Tell You How to Refine the Search

A useful initial search is not necessarily your final search. Its job is partly diagnostic.

Start with a defensible representation of the central topic, examine what it retrieves, and narrow only when you can identify a retrieval problem that the proposed change is likely to solve.

If you are still deciding whether the current result set is excessive, focus on its composition rather than its size. A search becomes problematic when its breadth prevents it from serving the purpose for which it was designed.

Before You Narrow Your Initial Search

Questions About How Broad an Initial Search Should Be

Begin With Enough Breadth to See the Literature Before You Constrain It

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.