How to Do a Systematic Literature Search: A Practical Guide

Learning how to do a systematic literature search means building a planned, documented, and reproducible search that aims to find every study relevant to your question rather than a convenient sample. You translate the question into concepts, develop a search strategy from each concept, run it across several databases, supplement it with grey literature and citation methods, and record every step so another reviewer could rerun it and get the same records.

Why a structured workflow beats ad-hoc searching

Typing a few keywords into one database returns a tidy-looking list, but it quietly misses studies that used different wording, indexing, or synonyms. A systematic search trades that convenience for sensitivity: you accept a larger, noisier set of records in return for confidence that you have not silently excluded eligible studies. The whole point is that the result reflects the literature, not the gaps in your vocabulary. Every choice below is made before searching and written into the review protocol, so the search is auditable rather than improvised.

Step one: turn the question into searchable concepts

Start by breaking the question into its core concepts using a framework such as PICO. A clear review question usually yields two or three concepts worth searching, often the population and the intervention. You rarely search the outcome, because outcomes are often poorly indexed and searching them throws away relevant records. Choosing which concepts to include is the single most important decision in the whole search, and the PICO framework exists to make that choice explicit.

Step two: gather terms for each concept

For every concept you collect two kinds of terms. The first is subject headings, the controlled vocabulary a database assigns to records, such as MeSH terms in MEDLINE. The second is free-text or keyword terms, the synonyms, spelling variants, and phrases that authors actually use. Mining a handful of known relevant papers for their titles, abstracts, and assigned headings is the fastest way to build a comprehensive term list. Truncation symbols and proximity operators let one term capture many word forms at once.

Step three: combine terms with Boolean logic

Within a concept you join synonyms with OR so any one of them retrieves the record. Between concepts you join the synonym blocks with AND so only records touching both concepts survive. This OR-within, AND-between structure is the backbone of every database search, and we cover the mechanics in detail in our guide to Boolean operators for literature searching. Build and test each concept block on its own before combining them, so you can see how each line changes the result count.

Step four: run, translate, and supplement

Run the finished strategy in your lead database, then translate it for each additional source, because syntax and subject headings differ between platforms. Picking the right mix is its own decision, covered in choosing databases for systematic reviews. Database searching alone is rarely enough, so you add grey literature searching for reports and trials registries, plus citation chasing and snowballing to catch studies the term-based search missed.

Step five: quality-check and document

Before you commit, have the strategy independently checked against the PRESS peer review checklist to catch missing terms, logic errors, and syntax slips. Finally, record every database, every search line, the date run, and the number of records retrieved, as set out in documenting and reporting your search. That record is what makes the search reproducible and what lets you report it against PRISMA 2020.

A worked two-concept search, line by line

Seeing the workflow as concrete search lines makes the abstract rules stick. Suppose the question asks whether mindfulness reduces anxiety in adults. The searchable concepts are the population and the intervention; the outcome stays out of the strategy for the reason given above. In an Ovid MEDLINE interface, a sensitive first draft might read:

1. exp Mindfulness/ (the exploded subject heading)
2. (mindful* or “mindfulness based” or MBSR or MBCT).ti,ab.
3. 1 or 2 (the finished intervention block)
4. exp Adult/
5. (adult* or men or women).ti,ab.
6. 4 or 5 (the finished population block)
7. 3 and 6 (the two blocks combined)

Notice the structure: each concept is built and OR-ed on its own lines, the two finished blocks are joined with AND at the end, and the .ti,ab. field tag restricts free-text to title and abstract. The asterisk in mindful* is truncation, which captures mindful, mindfulness, and mindfully in one term. The same logic is set out in detail in our guide to combining terms with Boolean operators, and the heading side is covered in choosing and exploding subject headings.

Validating the search against known studies

A search you cannot test is a search you cannot trust. Before running the full strategy, assemble a small set of known relevant studies you are certain belong in the review, often the studies that prompted the question or those cited in a recent overview. Run the draft and confirm it retrieves every one of them. If it misses a study, open that record, inspect the subject headings the indexer assigned and the exact wording of its title and abstract, and add whatever term or heading would have caught it. This relative recall check, repeated until the strategy finds the whole test set, is the most practical evidence that the search is sensitive enough. Keep the test set and the changes it prompted, because they belong in the development history you report later.

Common mistakes that quietly cripple a search

Most failed searches share a handful of avoidable errors, and almost all of them are invisible until the search is peer reviewed or rerun:

• Searching the outcome. Outcomes are inconsistently indexed, so an outcome block silently drops eligible studies; let title and abstract screening filter for the outcome instead.
• Free-text only, no headings. Skipping controlled vocabulary loses every study whose authors phrased the concept unexpectedly.
• Ungrouped operators. Mixing AND and OR on one line without grouping makes the database combine terms in the wrong order.
• One database only. Each index covers a different slice, so a single source builds a hidden gap into the review.
• Pasting a strategy across platforms. Syntax, field tags, and headings differ, so an untranslated strategy fails on the new platform without any error message.

A search built to avoid these is the foundation the rest of the review rests on, which is why it is the first concrete deliverable in our wider map of the review process and why a flawed search is so expensive: every later stage inherits its gaps. If the build, translation, and peer review are more than your team can take on, that is precisely what our literature search service delivers to a reproducible standard.