A reproducible search strategy is a written record that lets a colleague rerun your literature search and arrive at the same candidate set. The core task is not writing a narrative about what you searched; it is capturing the decisions that shaped the search — eligibility criteria, sources, term logic, filters, screening stages, and the date each step ran — so the process can be repeated and audited.
This guide is for clinicians and medical researchers assembling a focused evidence brief, research update, or team discussion. It walks through the starting state most teams are in, the artifacts a reproducible strategy needs, a step-by-step sequence with checkpoints, a worked example, and clear completion criteria. It stays at the level of search documentation and evidence appraisal; it does not offer individual treatment advice.
Why a search narrative is not a reproducible strategy
Most teams begin with a search they can describe but not reproduce. Someone ran a few queries, saved a handful of PDFs, and wrote "we searched PubMed and Google Scholar" in the methods. That sentence records that a search happened, but not what would let another person repeat it.
Cochrane's guidance frames the fix: plan searches from the eligibility criteria, use multiple sources, combine free-text terms with subject headings, document the process, deduplicate records, screen in stages, link multiple reports of one study, and rerun searches before publication. Each of those elements is a decision you can write down. A reproducible strategy is the sum of those written decisions.
A useful diagnostic: hand your current search notes to a colleague who was not in the room. If they cannot reconstruct your queries, sources, and screening rules from those notes alone, the strategy is not yet reproducible.
What artifacts does a reproducible search strategy require?
Before drafting, decide which artifacts you will produce. A complete record typically includes the following, and each maps to a decision you must make explicitly.
| Artifact | What it records | Decision it captures |
|---|---|---|
| Eligibility criteria | Population, exposure or intervention, outcome, study types | What counts as relevant |
| Source list | Databases and other sources searched | Where the search ran |
| Query log | Full strings, including field tags and headings | How terms were combined |
| Filter and limit log | Dates, languages, study types applied | What was excluded at search time |
| Deduplication note | How duplicate records were handled | How the record set was cleaned |
| Screening stages | Titles/abstracts, then full text | How records were narrowed |
| Study-linking note | How multiple reports of one study were grouped | How the unit of analysis was defined |
| Run dates and rerun note | When each search ran and whether it was refreshed | How current the evidence is |
Treat this table as your build checklist. If a row has no entry, you have an undocumented decision.
How do you turn eligibility criteria into search terms?
Start from the eligibility criteria, not from a favorite query. Break the question into its population, exposure or intervention, and outcome, then list the free-text synonyms and subject headings for each concept. Cochrane recommends combining free-text terms with subject headings, so plan both from the outset.
PubMed supports field tags, phrase searching, filters, and Advanced Search, and its Search Details view reveals how terms were translated and combined. That translation view is valuable documentation: it shows what the database actually did with your input, which is often more precise than what you typed.
At this stage, capture the query string exactly as run, including tags and any phrase syntax. A query you cannot paste back into the database is not yet documented.
Which sources belong in the record, and how do you log them?
Cochrane recommends using multiple sources rather than relying on a single database. Your source list should name each database and any other source you searched, and for each one record the date and the query as run in that interface.
Different sources behave differently, and your log should reflect that. PubMed's Search Details shows term translation and combination. Google Scholar documents searching by author and title, date sorting and limits, Cited by and Related articles, citation export, and email alerts. If you use those features to extend a search, note them, because a citation-chaining step is part of the strategy and should be reproducible too.
When you need to find academic papers across a focused topic, a scholarly search interface can serve as one of your logged sources. WisPaper's scholarly search is a place to run and gather candidate papers; record it in your source list exactly as you would any other database, with the query and date.
How should you document filters, limits, and deduplication?
Filters and limits are decisions, and each one narrows the candidate set. Log every date limit, language limit, and study-type filter you apply, and note the source where it was applied, since filters are interface-specific.
Deduplication is a separate documented step. Cochrane recommends deduplicating records, so record how you did it and roughly what it removed. If you exported citations, note the export format and the tool used to merge them.
Citation accuracy is part of this record. If you are screening a BibTeX file, TrueCite checks BibTeX against real academic databases to flag hallucinated references — a useful checkpoint before a reference list enters your documentation. Record that the check was run and what it flagged.
How do you document screening in stages?
Cochrane recommends screening in stages, typically titles and abstracts first, then full text. Document the rule applied at each stage and the outcome counts: how many records entered, how many were excluded, and why.
A simple screening log works:
- Stage 1 — title/abstract. State the inclusion rule in one sentence. Record records in and records excluded.
- Stage 2 — full text. State the full-text inclusion rule. Record records assessed, included, and excluded with reasons.
- Study linking. Cochrane recommends linking multiple reports of one study. Note how you grouped reports so that one study is not counted twice.
As papers move from candidate to included, keep them organized. My Library lets you save or upload papers and organize them, which gives your included set a stable home that matches the screening log.
How do you record evidence level, applicability, indirectness, and limitations?
A reproducible search strategy documents not only what you found but how you characterized it. For each included study or study group, record four separate judgments, and keep them distinct in your notes:
- Evidence level. The study design and the strength of the evidence it can support, recorded per study.
- Applicability. How closely the studied population, setting, and intervention match your brief's question.
- Indirectness. Where the evidence departs from your question — for example, a different population or a surrogate outcome.
- Limitations. The study's own constraints, recorded separately from applicability and indirectness so the three are not conflated.
Keeping these as four columns rather than one narrative paragraph prevents a common error: treating a limitation of the study as if it were a mismatch with your question, or vice versa. When you need to ask questions based on papers you have saved, Library QA supports asking questions based on papers uploaded to My Library or added from search results, which helps you interrogate the included set consistently.
Worked example: a focused population and outcome brief
Consider a team preparing a brief on a defined adult population and a single clinical outcome. The steps below show how the artifacts come together. The example illustrates documentation mechanics only; it makes no claim about what the evidence shows.
Step 1 — State eligibility. Population: adults with the defined condition. Outcome: the single outcome of interest. Study types: state which designs are eligible. Checkpoint: a colleague can read the criteria and classify a sample citation without asking you.
Step 2 — Build the concept map. List free-text synonyms and subject headings for population and outcome. Checkpoint: every concept has both free-text and heading entries.
Step 3 — Run and log queries. Run the string in each named source, including a scholarly search such as WisPaper search, and paste the exact query and date into the query log. Checkpoint: each query can be rerun verbatim.
Step 4 — Apply and log filters. Record each date, language, and study-type limit and where it was applied. Checkpoint: the filter log accounts for every narrowing step.
Step 5 — Deduplicate and verify citations. Record the deduplication method and run a citation check with TrueCite on any BibTeX you rely on. Checkpoint: the reference set has been screened for hallucinated entries.
Step 6 — Screen in stages. Apply the title/abstract rule, then the full-text rule, recording counts and exclusion reasons. Checkpoint: the counts reconcile from records found to studies included.
Step 7 — Link reports and appraise. Group multiple reports of one study, then fill the four appraisal columns — evidence level, applicability, indirectness, limitations — for each included study. Checkpoint: no study appears twice, and each appraisal column is populated.
Step 8 — Rerun before finalizing. Cochrane recommends rerunning searches before publication. Record the rerun date and any new records added. Checkpoint: the rerun is logged and the screening log is updated.
For a longer brief, a Research Project gives the work a shared home where the query log, screening counts, and appraisal table can live together. If the brief grows into a fuller literature review, Scholar Agent supports longer research tasks including literature review work.
Reusable checklist for your next brief
Use this checklist at the end of drafting. Each item should be answerable from your written record alone.
- Eligibility criteria written for population, exposure/intervention, outcome, and study types
- Every source named, with the date searched
- Full query strings saved verbatim, including field tags and headings
- Free-text terms combined with subject headings for each concept
- Every filter and limit logged with its source
- Deduplication method recorded
- Citation check completed on any BibTeX used
- Screening stages documented with counts and exclusion reasons
- Multiple reports of one study linked
- Evidence level, applicability, indirectness, and limitations recorded separately
- Rerun date logged before finalizing
How do you know the search strategy is ready?
The strategy is ready when a colleague who was not involved can rerun it and reach the same candidate set. Test that directly: hand over the record and ask them to reconstruct the queries, sources, filters, and screening rules without your help. If they can, and their record counts reconcile with yours, the documentation is complete.
Use these completion criteria as your final gate:
- Reproducible queries. Every query can be pasted back into its source and returns a comparable set.
- Complete source and filter log. Every source, date, and limit is recorded.
- Reconciling counts. Records found, deduplicated, screened, excluded, and included add up.
- Linked studies. No study is counted twice through multiple reports.
- Separated appraisal. Evidence level, applicability, indirectness, and limitations are each recorded per study.
- Logged rerun. The pre-publication rerun is dated and reflected in the screening log.
If any criterion fails, the gap points to the artifact that needs another entry, not to a rewrite of the narrative.




