03 · What You Need to Know
How to Use Generative AI to Develop Better Literature Search Terminology
Why Are Search Terms So Important?
A database search retrieves records according to the relationship between your query and the information indexed by the database. If your terminology is too narrow, relevant studies may be missed. If it is too broad, the search may retrieve large numbers of irrelevant records.
The difficulty is that scholarly concepts rarely have one universally accepted label. Researchers working in different disciplines may describe similar phenomena using different theoretical frameworks or terminology.
For example, studies investigating educators' use of technology may discuss technology acceptance, behavioral intention, adoption, implementation, or actual usage. These concepts are related, but they do not necessarily measure the same thing.
A useful search strategy therefore requires more than listing words that appear similar. Researchers must decide which expressions adequately represent each concept in the research question.
Cochrane's guidance on searching emphasizes using a broad range of appropriate free-text terms, including synonyms, spelling variants, and acronyms, alongside controlled vocabulary where available. This principle is particularly relevant to searches intended to identify evidence comprehensively. Cochrane Handbook, Chapter 4.
What Types of Search Terms Can AI Generate?
Generative AI can propose several categories of terminology. These should be distinguished because they serve different retrieval purposes.
| Term category |
Illustrative examples |
How to evaluate it |
| Direct synonyms |
"higher education institution" and "tertiary education institution" |
Check whether the expressions are used equivalently in the relevant context |
| Alternative terminology |
"faculty members," "university lecturers," and "academic staff" |
Determine whether each term represents the intended population |
| Related constructs |
"technology acceptance," "adoption intention," and "technology resistance" |
Assess conceptual overlap without assuming equivalence |
| Acronyms and abbreviations |
"generative artificial intelligence," "generative AI," and "GenAI" |
Check ambiguity and actual usage in scholarly records |
| Spelling variants |
"behavior" and "behaviour" |
Include relevant variants when database mapping does not already account for them |
| Broader or narrower expressions |
"artificial intelligence" and "generative artificial intelligence" |
Determine whether broadening or narrowing changes the intended scope |
| Historical terminology |
Earlier terms used for a technology or construct |
Verify that the terminology refers to the phenomenon under investigation |
The examples illustrate potential search expressions rather than universally interchangeable terms. Their suitability depends on the research question, discipline, and database.
Why Should You Separate Synonyms From Related Concepts?
This is one of the most consequential decisions when using AI for search term generation.
A synonym is an alternative expression that conveys substantially the same meaning within the relevant context. A related concept may share theoretical or empirical connections without referring to the same construct.
Consider a researcher investigating trust in generative AI. AI might suggest confidence, perceived reliability, acceptance, perceived usefulness, and willingness to use.
Some of these terms may be relevant to the literature. However, trust, perceived usefulness, and behavioral intention are distinguishable constructs in many theoretical models.
Combining them indiscriminately could change the search from studies specifically examining trust to a much broader collection of studies about technology attitudes.
Synonym expansion
Adds alternative expressions intended to represent the same search concept.
Conceptual expansion
Adds related constructs that may broaden the theoretical or empirical scope of the search.
Both can be useful, but they should not be treated as equivalent decisions. Synonym expansion primarily addresses vocabulary variation, while conceptual expansion may alter what counts as relevant evidence.
How Should You Prompt AI to Generate Useful Search Terms?
A vague prompt such as "Give me keywords about AI in education" may produce a long list containing general topics, theoretical constructs, and unrelated expressions.
A more useful prompt specifies the research question, population, setting, and conceptual boundaries.
Example Prompt
Generate Search Terms Without Confusing Related Constructs
"I am conducting a literature review on university faculty members' trust in generative AI for teaching. Identify potential search terms for the population, technology, and trust constructs. Organize the suggestions into direct synonyms, alternative terminology, acronyms, spelling variants, and related but conceptually distinct constructs. Do not assume that related constructs are interchangeable. Flag ambiguous terms and explain which suggestions require verification against scholarly literature."
This prompt asks AI to make its conceptual distinctions visible. It also discourages the common tendency to treat every associated word as an equally useful synonym.
Researchers can refine the output further by providing definitions of the constructs, known relevant papers, or explicit inclusion and exclusion criteria.
Even a carefully constructed prompt cannot guarantee that the suggested terminology appears in the literature. That requires checking actual scholarly records.
How Can You Verify AI-Generated Search Terms?
Start with a small set of known relevant publications. Examine their titles, abstracts, author keywords, and indexed subject terms where available.
Compare the language used in those records with the AI's suggestions. Terms appearing repeatedly in relevant publications may be useful candidates, although frequency alone does not establish conceptual suitability.
Next, test the terms in the intended database. Determine whether they retrieve relevant records, duplicate results already found through other expressions, or introduce substantial irrelevant material.
Pay attention to ambiguous words. For example, "acceptance" may appear in studies about technology adoption, institutional acceptance, manuscript acceptance, or social acceptance. Its usefulness depends on how it is combined with other concepts.
For searches requiring greater methodological rigor, document why important terms were included or excluded. A librarian or information specialist can help evaluate whether the terminology adequately represents the research question.
Should You Include Every Synonym AI Suggests?
No. More terms do not automatically produce a better search.
Adding genuine alternatives within a concept can improve retrieval sensitivity. However, adding weakly related or ambiguous expressions may substantially increase irrelevant results.
The goal is not to maximize the number of terms. It is to represent the intended concept adequately while maintaining a manageable balance between sensitivity and precision.
For systematic reviews, sensitivity is often prioritized because missing eligible studies can affect the conclusions. Nevertheless, Cochrane guidance emphasizes that search strategies should be carefully structured rather than indiscriminately expanded.
A useful practice is to maintain a candidate term list with brief explanations of why terms were retained, modified, or rejected.
Can AI Identify Controlled Vocabulary Terms?
AI can suggest possible controlled vocabulary terms, but it cannot establish that a term belongs to a particular database thesaurus without verification.
For example, PubMed uses Medical Subject Headings (MeSH) for MEDLINE indexing. Other databases may use different vocabularies or subject indexing arrangements.
Researchers should verify suggested headings in the authoritative vocabulary interface rather than assuming that a familiar phrase is an official subject heading.
Controlled vocabulary can complement free-text searching by retrieving records indexed under standardized concepts even when the title or abstract uses different wording.
The process of identifying and validating controlled vocabulary deserves particular attention because database-specific indexing rules can affect retrieval.
How Do Search Terms Become a Database Search Strategy?
Once suitable terminology has been selected, researchers can organize terms into conceptual groups.
Alternative expressions for the same concept are commonly combined with OR. Different concepts are then combined with AND when the research question requires their intersection.
For example, a search on faculty trust in generative AI might include one group of terms for university faculty, another for generative AI, and another for trust.
The final expression must be adapted to the database's search fields, phrase handling, and operator syntax.
AI can help organize terms into Boolean search strings, but that step should follow conceptual validation rather than replace it.
Can AI-Generated Terms Help Discover Literature That Initial Searches Missed?
Yes, particularly when the initial query relies on a narrow or unfamiliar expression.
Suppose a researcher searches for "faculty AI resistance" but discovers that relevant studies describe reluctance, distrust, perceived risks, or adoption barriers. Carefully incorporating appropriate alternatives may retrieve additional publications.
However, those terms may also represent different constructs. The researcher must decide whether the additional studies genuinely address the research question.
In this sense, term generation and discovering literature missed by keyword searches are closely related but distinct activities. One improves the vocabulary used in searching; the other concerns the records ultimately retrieved.
