How to Write a Systematic Literature Review

A rigorous, reproducible review method distinct from a traditional narrative review, protocol, search, screening and synthesis you can defend.

A systematic literature review differs fundamentally from a traditional narrative review: it follows an explicit, pre-planned and reproducible method to minimise bias. Increasingly required in dissertations and standard in health and social sciences, it is judged on the transparency and rigour of its process as much as its conclusions.

Narrative vs Systematic

A narrative review synthesises literature through the author's expert selection, valuable but vulnerable to bias, because the reader cannot see how sources were chosen. A systematic review documents every step so another researcher could repeat it and reach the same set of included studies. That reproducibility is the whole point.

Start with a Protocol and a Question

Define a focused question, often using a framework like PICO (Population, Intervention, Comparison, Outcome) or, for qualitative reviews, SPIDER. Write a protocol specifying your question, inclusion/exclusion criteria, databases and methods before searching, and register it where appropriate (e.g. PROSPERO) to prevent shifting the goalposts.

Comprehensive, Documented Search

Search multiple databases with a documented combination of keywords and controlled vocabulary. Record the exact search strings and dates. The search must be exhaustive within your criteria and reproducible from your documentation alone, supplement with reference-list and forward-citation searching.

Transparent Screening

Apply your pre-set criteria in stages, title/abstract, then full text, ideally with two independent reviewers to reduce selection bias. Record numbers excluded and why, and present the flow with a PRISMA-style diagram so the reader sees exactly how you moved from thousands of records to your final set.

Appraise and Synthesise

Assess the quality of included studies with an appropriate appraisal tool, then synthesise. Where studies are comparable and quantitative, meta-analysis may be possible. Where they are heterogeneous or qualitative, synthesise narratively or thematically around the patterns across studies. Either way, the synthesis should answer your question, not merely list what each study found.

Report Rigorously

Follow a recognised reporting standard (PRISMA for many fields). The methods section should be detailed enough to reproduce. The results should present the synthesis clearly. And limitations should acknowledge where the evidence base or your process fell short.

Checklist

Did I write (and ideally register) a protocol before searching?
Is the search comprehensive and fully documented?
Is screening against pre-set criteria, shown in a flow diagram?
Does the synthesis answer the question, with quality appraised?
Are exclusion reasons recorded at full-text stage?
Have I searched beyond databases, including reference lists and grey literature?

Inclusion and Exclusion Criteria, Set Before You Search

Criteria written after screening are not criteria, they are justifications. Fix them in advance, write them down, and apply them mechanically. Where a judgement call arises, record it and note the rule you applied so the next borderline case is handled the same way.

Dimension	Include	Exclude	Why the boundary
Population	Defined by your question	Adjacent groups	Keeps the synthesis about one thing
Study design	Designs that can answer it	Designs that cannot	An opinion piece cannot evidence effectiveness
Date range	A justified window	Older material	State the reason, such as a change in practice or policy
Language	Usually English for student work	Others	A real limitation. Name it rather than hiding it
Publication type	Peer-reviewed, plus grey literature if justified	Editorials, blogs	Excluding grey literature risks publication bias
Setting	Contexts that transfer to your question	Others	Prevents pooling incomparable material

Two of those deserve a note. A date cut-off needs a reason beyond recency, because "the last ten years" is arbitrary unless something changed ten years ago. And an English-only restriction introduces a known bias, so state it in limitations rather than passing over it.

Screening in Two Stages

Screening runs in two passes, and conflating them is a common error that produces an unusable audit trail.

Title and abstract. Fast, inclusive, and biased toward keeping. If you cannot tell from the abstract, keep it. You need no reasons at this stage, only counts.
Full text. Slower and decisive. Here you must record a reason for every exclusion, because PRISMA requires them and because the reasons themselves tell the reader about the evidence base.

Keep a simple spreadsheet with one row per record, columns for each screening decision, and a reason column. Building it as you go takes minutes. Reconstructing it afterwards from memory takes days and produces something you cannot fully trust.

The Flow Diagram Is Evidence, Not Decoration

The numbers must reconcile. Records identified, minus duplicates, gives records screened. Records screened, minus those excluded, gives full texts assessed. Full texts assessed, minus exclusions with reasons, gives studies included. Markers do check the arithmetic, and a diagram that does not add up undermines everything downstream of it.

Extracting Data Consistently

Build one extraction table and use it for every study, even where a field comes out empty. Consistent fields make the synthesis possible, whereas free-form notes leave you comparing things that were never recorded the same way.

Field	What to record
Citation	Author, year, country
Design	Study type and duration
Participants	Number, and defining characteristics
Intervention or focus	What was done or examined
Comparator	What it was measured against, if anything
Outcomes	What was measured, and how
Key findings	Direction and size, with the numbers where reported
Quality appraisal	Tool used and result
Notes	Anything that affects comparability

This table usually goes in an appendix, and a condensed version of it belongs in the body as your characteristics-of-included-studies table.

Synthesis Is Argument, Not Sequence

The commonest failure in a student systematic review is a results section that walks through the included studies one at a time. That is a list, and it leaves the reader to do the synthesis.

Organise instead by finding, by outcome or by theme, and bring studies together under each. Where studies agree, say so and say how strongly. Where they disagree, explain the disagreement by reference to design, population or measurement rather than reporting both and moving on. Where the evidence is thin, say that too, because an honest account of a weak evidence base is itself a finding.

A useful discipline: every paragraph of your synthesis should begin with a claim about the evidence, not with an author's name. If your paragraphs open with "Smith (2019) found...", you are still listing.

Common Mistakes and Fixes

Mistake	Fix
Study-by-study results section	Reorganise by theme or outcome, with studies as evidence
Criteria written after screening	Set them first, and record any later change with its reason
Quality appraised but never used	Weight the synthesis by quality, and say where weak studies drive a finding
Search string absent	Reproduce it per database in an appendix, with dates
Duplicates not removed or not counted	Deduplicate in a reference manager and report the number
Grey literature excluded silently	Either search it or state the decision and its publication-bias implication

Where Ethical Support Fits

Subject librarians build search strategies for a living, and asking one to check your string before you run it is among the best uses of an hour in the whole project. Departments also expect you to seek help with appraisal tools you have not used before.

What stays yours is screening the records, appraising the studies and writing the synthesis. Those judgements are the review.

Frequently Asked Questions

How is this different from a narrative literature review?

A systematic review follows a pre-specified, reproducible protocol and reports the process in full. A narrative review selects sources by judgement. Both are legitimate. Only one may be described as systematic.

How many databases should I search?

At least three relevant to your field, plus reference-list checking. Which three depends on the discipline, and a librarian will tell you in five minutes.

Do I need to appraise quality if I am not doing meta-analysis?

Yes. Appraisal is what lets you weight the evidence. Without it, a small flawed study carries the same rhetorical weight as a large careful one.

What if I find only a handful of studies?

Report it as a finding. A sparse evidence base tells the reader something real. Check your search first, since an unexpectedly small yield sometimes indicates a string that is too restrictive.

Can I include qualitative studies?

Yes. Qualitative evidence synthesis is an established approach with its own appraisal tools, such as the CASP qualitative checklist. You may also combine both in a mixed-methods review, provided you explain how.

Should I register a protocol?

For student work it is usually optional, though good practice. PROSPERO registration is standard for health reviews intended for publication.

Your Next Step Today

Build your screening spreadsheet before you run a single search. Columns for record, title-abstract decision, full-text decision and exclusion reason. Ten minutes now saves the reconstruction later, and it is what makes the flow diagram truthful rather than approximate.

Trusted Sources

PRISMA statement, checklist and flow diagram templates. Accessed 11 August 2026.
Louisiana State University Libraries, Steps in a Systematic Review. Accessed 11 August 2026.
Paré G and Kitsiou S, Methods for Literature Reviews, in Handbook of eHealth Evaluation, NCBI Bookshelf, on review typologies. Accessed 11 August 2026.
EQUATOR Network, reporting guidelines across study types. Accessed 11 August 2026.
University of Southern California, The Literature Review. Accessed 11 August 2026.

Review expectations differ by discipline and supervisor. Where this guide and your department's guidance differ, follow your department.