Start with the question, not the search
A review question fails most often because it was never pinned down. PICO forces the four decisions you would otherwise make silently, halfway through screening:
- Population — who, with what condition, in what setting. "Adults" is not a population; "adults aged 40–75 with type 2 diabetes in primary care" is.
- Intervention — the thing being done, with enough detail that a reader could deliver it.
- Comparator — usual care, placebo, an active alternative, or nothing. Write N/A if your question genuinely has no comparator, and say why.
- Outcomes — at least one primary outcome, decided before you search. Choosing outcomes after seeing results is selective outcome reporting.
Write the question down and date it. If you change it later, keep the earlier version and note what changed. Examiners treat an evolving question as normal; an undocumented one as a problem.
Register a protocol before you screen
PROSPERO registration takes an afternoon and closes off the most common criticism of student reviews — that inclusion criteria were adjusted to fit what was found. Record your eligibility criteria, search plan, outcomes, and how disagreements will be resolved.
Work from open-access sources
Building your review on sources anyone can reach means your supervisor, your examiner, and a reader three years from now all see the same evidence you did. Open and public sources cover a great deal: PubMed Central's open-access subset, Europe PMC, ClinicalTrials.gov and other trial registries, DOAJ journals, preprint servers, and regulator documents.
Where only an abstract is available, say so in your table. An abstract-only record can inform screening but rarely supports extraction — mark the difference rather than blurring it.
Search more than one database
A single-database search is the fastest way to lose marks. Search at least two bibliographic sources plus a trial registry, and record each search string, the database, the date run, and the number of hits. Deduplicate on DOI and registry number.
Screen with reasons
Every exclusion at full-text stage needs a stated reason — wrong population, wrong comparator, no extractable outcome, and so on. Those reasons become your PRISMA 2020 flow diagram. Two screeners are the standard; where a second screener is not available, say that plainly in your limitations.
Disclose AI use from day one
Most universities now expect a statement describing which tools were used and for what. The practical approach is to keep a running note as you work: tool, date, task, and what you did with the output. Retro-fitting that record at submission is painful and usually incomplete.
ICMJE and COPE are clear that AI cannot be an author and cannot be credited as one — responsibility for every claim stays with you. Disclosure is not an admission; it is the expected professional behaviour.
A realistic order of work
- Draft and date the PICO question.
- Scope the literature briefly to check the question is answerable.
- Write eligibility criteria and register the protocol.
- Run and record searches across multiple sources.
- Deduplicate, then title/abstract screen.
- Full-text screen, recording an exclusion reason for each.
- Extract data, assess risk of bias, synthesise.
- Assemble the PRISMA flow and AI-use declaration.