- A search requires you to think about strategy first. Map your concepts, collect synonyms for each, then join them with Boolean logic.
- Pair controlled vocabulary such as MeSH with free-text keywords. Using either alone will miss relevant studies.
- Google Scholar is excellent for scoping and citation chaining, but it cannot produce the reproducible strategy a systematic review requires. Choose databases like Web of Science and Scopus for your main search.
- Record every search line on the day you run it, including database, platform, date, and result count.
What Is a Literature Search?
A literature search is a structured, documented process for finding published and unpublished evidence on a defined question. It is a method, not a reading task, and someone else should be able to repeat it.
The kind of search you need depends on why you are searching:
- Background scan: you need orientation in an unfamiliar field, fast.
- Evidence synthesis: you need every eligible study, because omissions bias the result, i.e., you are conducting a literature review for publication.
- Find research gaps: you need to prove what has not yet been studied, in order to justify a study and get funding or approval.
Literature Search vs Literature Review
| Feature | Literature search | Literature review |
| Purpose | Locate all relevant evidence on a question | Interpret and synthesize that evidence |
| Output | A set of records plus a documented strategy | A written synthesis or argument |
| Scope | Set by the protocol and eligibility criteria | Set by the analytic or narrative aim |
| Core skills | Databases, operators, filters, chaining | Appraisal, extraction, synthesis |
| What you record | Search lines, dates, platforms, result counts | Study characteristics, findings, quality ratings |
| When it ends | When new searches return only known records | When the argument is complete and defensible |
Structured vs Unstructured Literature Search
| Feature | Structured search | Unstructured search |
| Starting point | A defined question broken into 2-4 concepts | A topic and a hunch |
| Terms | Synonyms plus subject headings, listed in advance | Whatever comes to mind at the keyboard |
| Logic | Boolean blocks combined deliberately | Words typed into 1 box |
| Sources | Databases chosen by coverage, plus grey literature | Whichever database or search engine is nearest |
| Record kept | Every line, date, platform, and result count | None |
| Repeatable | Yes, by someone else | No, often not even by you |
| Stops when | New searches return only known records | You have enough to start writing |
| Best for | Research papers, systematic reviews, theses and dissertations, grant applications | Orientation, background reading, teaching prep, classroom presentations, undergraduate coursework |
Unstructured searching is the right tool for finding your footing in an unfamiliar field. The problem is only when it gets reported as though it were structured.
When Is a Structured Search Required?
- You are writing a protocol, dissertation, grant application, or any review that will be peer reviewed.
- The conclusion will inform practice, policy, or spending decisions.
- You must demonstrate that a gap exists, rather than simply assert it.
- Several people are screening, so the method has to be explicit enough to share.
- You will update the work later and need to rerun the same searches.
What Is Controlled Vocabulary?
Controlled vocabulary is a fixed list of subject terms that indexers assign to every record, so that 1 concept carries 1 label regardless of the words the authors happened to choose.
It solves the synonym problem at the source. A record indexed under “Myocardial Infarction” is retrievable whether the paper says heart attack, MI, or coronary thrombosis.
MeSH, Emtree, CINAHL Headings, and the APA Thesaurus
| Database | Thesaurus | Notes |
| PubMed and MEDLINE | MeSH | Enter as “term”[Mesh]; narrower terms are included automatically |
| Embase | Emtree | Larger drug and device vocabulary; add /exp to explode a term |
| CINAHL | CINAHL Headings | Nursing and allied health focus; searched in the MH field |
| PsycINFO | APA Thesaurus | Behavioral science terms; searched in the DE field |
| Cochrane Library | MeSH | Many CENTRAL records are unindexed, so free text is essential |
| Scopus, Web of Science, Google Scholar | None | No subject headings at all; free-text searching only |
Explosion, Subheadings, and Major Topic
- Explosion retrieves a heading together with every narrower heading beneath it in the tree; it is on by default in PubMed.
- Turn explosion off only when the narrower terms are clearly irrelevant to your question.
- Subheadings, such as drug therapy or adverse effects, restrict a heading to 1 aspect of the topic.
- Major topic, entered as [majr], keeps only records where the concept is a main focus rather than a passing mention.
- Subheadings and major topic both raise precision and lower sensitivity, so avoid them in a systematic review unless results are unmanageable.
Why Do You Still Need Keywords Alongside Subject Headings?
- Indexing lags publication, so the newest records often carry no headings yet.
- Indexing is done by people and is inconsistent, particularly for secondary outcomes.
- Emerging concepts have no heading until the vocabulary catches up, sometimes years later.
- Ahead-of-print and publisher-supplied records in PubMed are not yet indexed.
- Several major databases have no thesaurus at all.
What Are Boolean Operators?
Boolean operators are the connectors AND, OR, and NOT, which tell a database how to combine your terms. They decide whether your results narrow, widen, or shrink by exclusion.
AND, OR, NOT: What Each Operator Does
| Operator | Effect on results | Example |
| AND | Narrows: every record must contain both concepts. Use it between different concepts. | delirium AND prevention |
| OR | Widens: a record needs only 1 of the listed terms. Use it between synonyms. | teenager OR adolescent OR youth |
| NOT | Excludes: removes every record containing the second term. Use it sparingly. | dementia NOT alzheimer |
Most databases require operators in capitals. If typed in lowercase, a Boolean operator is often treated as an ordinary word and your results will change.
Nesting, Parentheses, and Order of Operations
Databases process OR inside parentheses first, then AND, then NOT. Without parentheses, a mixed query is read in an order you did not intend.
- Correct: (teenager OR adolescent) AND vaping. Both synonyms are tied to the second concept.
- Incorrect: teenager OR adolescent AND vaping. This returns every record about teenagers, plus a much smaller set about adolescents who vape.
- Keep 1 concept per parenthesized block, then join the blocks with AND.
- Build each block as a separate line in the search history, so you can test and swap blocks independently.
What Is Truncation?
Truncation is a symbol added to a word root so that the database retrieves every ending of that root. Typing nurs* returns nurse, nurses, and nursing in a single line.
| Symbol | Function | Where it works |
| * | Truncation: stands for any number of characters after the root | PubMed, Embase, CINAHL, Scopus, Web of Science, Cochrane |
| $ | Truncation, in the Ovid syntax | Ovid MEDLINE, Ovid Embase, Ovid PsycINFO |
| ? | Wildcard: stands for 1 character or none | Embase, CINAHL, Scopus, Ovid |
| # | Wildcard: stands for exactly 1 character, which must be present | CINAHL, Ovid |
| Not supported | Enter each variant as a separate term joined by OR | Google Scholar |
- Keep at least 4 characters before the symbol. Truncating cat* pulls in cataract, catastrophe, and catheter.
- In PubMed, truncation switches off automatic term mapping, so no MeSH terms are added for you.
- PubMed expands a truncated root to a maximum of 600 variants, then warns you; very short roots exceed this silently in older interfaces.
- Truncation is best used for for guessing at words humans chose freely. Don’t use truncation for subject headings as there’s nothing left to guess. Here’s an analogy: truncation is like searching your phone contacts for “Jen” to catch Jen, Jenny, and Jennifer, because you don’t remember how you saved her number. Subject headings are like picking a country from a dropdown menu on a form. Typing “Den*” into a dropdown does nothing useful. You just scroll to Denmark and click it.
What Are Wildcards?
A wildcard is a symbol standing in for 1 character, or for no character at all. It catches spelling variants inside a single term: wom?n retrieves both woman and women.
- Use wildcards for US and UK spellings: behavio?r covers behavior and behaviour.
- Use them for irregular plurals, where truncation would not help.
- Wildcards are placed inside the word whereas truncation is placed at the end of it.
- Support varies by platform, so check the symbol table above before you paste a string into a new database.
- Google Scholar supports neither, so spell out each variant and join the variants with OR.
5 Common Boolean Mistakes
- Using AND between synonyms, which demands that both words appear and collapses your results to near zero.
- Using NOT to remove a topic, which also removes relevant records that merely mention it in passing.
- Omitting parentheses around an OR block, so the operators execute in the wrong order.
- Typing operators in lowercase, so the database reads them as search terms.
- Combining every concept in a single line, which leaves you no way to see which block is failing.
What Is a Search String?
A search string is the complete query you paste into 1 database, combining terms, Boolean operators, field tags, and truncation in the syntax that database understands.
A string is never universal. The same concepts must be retranslated for each platform, because subject headings, field tags, and proximity syntax all differ.
Building Blocks: Concepts, Synonyms, and Field Tags
- Break the question into 2 to 4 concepts. More than 4 usually over-restricts the search.
- For each concept, list synonyms, abbreviations, brand and generic names, spelling variants, and older terminology.
- Add the relevant subject headings for that database alongside the free-text terms.
- Join everything within a concept using OR, and wrap the block in parentheses.
- Attach field tags to the free-text terms, usually title and abstract, to keep precision reasonable.
- Join the concept blocks with AND, then run and inspect the result.
Phrase Searching and Proximity Operators
Quotation marks force words to appear together in order. Proximity operators are looser: they require the terms to be close, in any order, which often recovers records that a rigid phrase misses.
| Syntax | Meaning | Platform |
| “quality of life” | Exact phrase, in this order | Most databases, including Google Scholar |
| pain ADJ3 management | Within 3 words, either order | Ovid MEDLINE, Ovid Embase |
| pain N3 management | Within 3 words, either order | CINAHL, EBSCO platforms |
| pain W3 management | Within 3 words, in this order | CINAHL, EBSCO platforms |
| “pain management”~3 | Within 3 words, either order | Scopus, Web of Science variants |
| Not supported | Use phrases instead | PubMed basic search, Google Scholar |
How Do You Test and Refine a Search String?
- Assemble 5 to 10 papers you already know are eligible, then check that your string retrieves all of them.
- If a known paper is missing, open it from that same database and take note of the subject headings and title words you overlooked.
- Scan the first 50 titles. If they’re mostly highly irrelevant, it usually means a term is too broad or a field tag is too wide.
- Run each concept block on its own and compare result counts. A block returning very few records is the likely weak point.
- Ask a second reviewer or a librarian to check the string before you run the final version.
Best Databases for a Literature Search
Choose databases by subject coverage first, then by the features you need, such as citation tracking or a thesaurus. Searching more databases is not automatically better; searching the wrong ones wastes screening time.
Core Databases by Discipline
| Database | Best for | Access |
| PubMed and MEDLINE | Biomedicine, clinical practice, public health | Free |
| Embase | Drugs, medical devices, European and conference literature | Subscription |
| Cochrane Library, including CENTRAL (Cochrane Central Register of Controlled Trials) | Trials and existing systematic reviews | Free in many countries |
| CINAHL (Cumulative Index to Nursing and Allied Health Literature) | Nursing and allied health | Subscription |
| PsycINFO | Psychology and behavioral science | Subscription |
| Scopus | Multidisciplinary coverage with strong citation tracking | Subscription (Scopus Preview is free and so is author search and journal rankings search) |
| Web of Science | Multidisciplinary citation indexes, cited reference searching | Subscription |
| ERIC (Education Resources Information Center) | Education research and policy | Free |
| Google Scholar | Scoping, citation chaining, grey literature | Free |
How Many Databases Do You Need?
For a systematic review, plan on 3 to 4 subject databases plus trial registries and grey literature. For a background scan, 1 database plus citation chaining is usually enough.
- In health research, the common core is MEDLINE, Embase, and CENTRAL, with CINAHL or PsycINFO added by topic.
- Add 1 multidisciplinary database, such as Scopus or Web of Science, when the topic crosses fields.
- Add a regional or language-specific database when the evidence is likely to be concentrated outside English-language journals.
- Stop adding databases when a new source returns almost nothing you have not already seen.
Overlap, Deduplication, and Reference Managers
- Export from each database separately and keep the original file, named with the database and date.
- Import everything into a reference manager, such as EndNote, Zotero, or Mendeley, before you deduplicate.
- Deduplicate in 2 passes: automatic matching first, then a manual check of near-duplicates with differing page numbers or DOIs.
- Record the number of records before and after deduplication; the PRISMA flow diagram requires both.
- Screening tools such as Rayyan or Covidence handle deduplication and blinded screening in 1 place.
How to Use Google Scholar for a Literature Search
Google Scholar indexes the full text of an enormous, undocumented range of material, including books, theses, and preprints. That breadth makes it excellent for discovery and poor for reproducibility.
Advantages of Google Scholar for a literature search:
- Free, and it spans nearly every discipline.
- Full-text indexing surfaces phrases buried deep in a paper, which subject databases never index.
- “Cited by” makes forward citation chaining fast and free.
- Library Links connects results to your institution’s subscriptions.
Limitations for a literature search:
- No controlled vocabulary, no truncation, and no wildcards.
- Queries are capped at roughly 256 characters, and anything beyond that is silently ignored.
- Only about the first 1,000 results of any search can be viewed or exported.
- Coverage is undisclosed and changes without notice, so the same search can return different results next month.
- Bulk export is impractical, since records must be saved a few at a time.
Advanced Search, Operators, and Date Ranges
- Open the Advanced search panel from the menu to set phrase, author, journal, and year fields without memorizing syntax.
- Quotation marks work for phrases; OR must be capitalized; a minus sign immediately before a word excludes it.
- allintitle: restricts every following word to the title, which is the fastest way to cut a bloated result set.
- author: and source: limit by author and publication.
- Set a custom year range in the left sidebar, and untick patents and citations to remove clutter.
Cited By, Related Articles, and Library Links
- “Cited by” lists later papers that cite the record: this is forward citation chaining.
- Tick “Search within citing articles” to run a keyword search inside that citing set, which is far more precise than browsing it.
- “Related articles” surfaces papers with similar text and citation patterns, useful when your terminology is uncertain.
- Configure Library Links under Settings so results show your institution’s full-text access.
- “All versions” often reveals a free repository copy of a paywalled paper.
Alerts, My Library, and Export Workflow
- Create an alert on a saved search to receive new matches by email.
- Save promising records to My Library, then export them in batches to your reference manager.
- Set the bibliography manager format under Settings so each result shows an import link.
- Screen a predefined number of results, commonly the first 100 to 200 sorted by relevance, and record where you stopped.
Is Searching Google Scholar Enough for a Research Paper or Thesis?
No. Google Scholar is a valuable supplementary source, but its undisclosed coverage, 1,000-result ceiling, and lack of reproducible syntax rule it out as a primary database.
- Use it to find grey literature, theses, and reports that the subject databases miss.
- Use it for forward chaining from key papers and for validating your main search.
- Always report how many results you screened and how you sorted them.
- Never present it as your only source in a review, especially if you’re describing it as systematic.
How to Run a Literature Search in PubMed
PubMed is free, covers biomedicine comprehensively, and combines MeSH indexing with flexible free-text searching. Work in the Advanced Search Builder rather than the single search box.
Step-by-Step Walkthrough
- Write the question and split it into 2 to 4 concepts.
- Open the MeSH Database and identify the best heading for each concept.
- Run each MeSH heading as its own search line and note the result count.
- Build a free-text line for the same concept using [tiab], synonyms, and truncation.
- Combine the MeSH line and the free-text line for that concept with OR.
- Repeat for every concept, so each has a single combined line.
- Combine the concept lines with AND in the Advanced Search history.
- Apply filters last, save the search to My NCBI, and export the results.
Using the MeSH Database and Search Builder
- Search the MeSH Database for your concept, then read the scope note to confirm the term means what you assume.
- Check the tree display for broader and narrower headings you may want to include or exclude.
- Use “Add to search builder” to generate correct syntax automatically.
- Note the entry terms listed on the MeSH record: they are a ready-made synonym list for your free-text line.
- Remember that MeSH indexing does not cover ahead-of-print records, so the free-text line does that work.
Field Tags and Search Details
| Tag | What it searches | Example |
| [tiab] | Title and abstract | delirium[tiab] |
| [tw] | Text words across several fields | delirium[tw] |
| [mh] or [Mesh] | MeSH heading, exploded by default | “Delirium”[Mesh] |
| [majr] | MeSH heading flagged as a major topic | “Delirium”[majr] |
| [au] | Author name | smith j[au] |
| [ta] | Journal title | lancet[ta] |
| [dp] | Date of publication, including ranges | 2015:2026[dp] |
| [pt] | Publication type | randomized controlled trial[pt] |
| [la] | Language | english[la] |
Open “Search details” on the results page to see how PubMed translated your query. Automatic term mapping adds MeSH terms for you, but it is switched off by quotation marks, truncation, and field tags.
Search History, Combining Lines, and Saving to My NCBI
- Every line in the Advanced Search history gets a number, such as #3, which you can reuse in later lines.
- Combine lines by entering #1 AND #2, or by using the Actions menu beside each line.
- Copy the full history into your protocol before you leave the session, since it clears after 8 hours.
- Sign in to My NCBI to save searches permanently and to schedule email alerts.
- Use the same account to store collections of records and your filter preferences.
Exporting Results and Setting Alerts
- Use Save, then choose PubMed format or CSV, to export up to 10,000 records at a time.
- Choose the RIS-compatible option when importing into EndNote, Zotero, or Mendeley.
- Set the alert frequency to match your project; monthly is adequate for most reviews.
- Rerun and report the search close to submission, since reviewers expect currency.
Common Filters in Databases and When to Apply Them
Filters are prebuilt limits applied after your search runs. They save time, but most depend on indexing that is incomplete, so they can discard eligible studies without you noticing anything.
Filter Types and Typical Use
| Filter | Use it when | Caution |
| Date | A guideline or technology changed the field at a known point | Cuts out foundational work. Always justify the cutoff in your literature review. |
| Language | Translation is genuinely unavailable to you | Introduces language bias; report it as a limitation |
| Publication type | You need 1 design, such as randomized trials | Depends on indexing that is often missing or wrong |
| Species | Only human evidence is eligible | Unindexed recent records are dropped |
| Age group | The question is specific to 1 age band | Poorly indexed; free-text age terms work better |
| Free full text | You need something to read this afternoon | Skews to open access journals; never use it in a review |
| Sex | The question is genuinely sex-specific | Frequently absent from records |
When Should You Avoid Filters?
Avoid a filter whenever it depends on indexing that may be missing. Free full text, age group, and publication type filters all silently discard recent or poorly indexed records.
- Avoid all filters in a scoping review, where breadth is the point.
- Avoid the free full text filter in any review, because it selects for your library rather than for relevance.
- Avoid stacking 3 or more filters, since the losses multiply and become impossible to trace.
- If you must limit, do it with search lines you can see and report, not with checkboxes.
Better Alternatives: Validated Search Filters and Hedges
- The Cochrane Highly Sensitive Search Strategies identify randomized trials in MEDLINE with tested sensitivity.
- The ISSG Search Filter Resource collects filters for study designs, populations, and topics across platforms.
- The McMaster Hedges offer tested strategies for etiology, prognosis, diagnosis, and qualitative research.
- Paste a validated filter as its own search line, then combine it with AND, so it appears in your reported strategy.
- Report the filter and its version, exactly as you would cite any other method.
How Do You Find Grey Literature?
Search repositories, agency and government sites, trial registries, preprint servers, and dissertation databases, then run targeted site: and filetype: searches on the open web.
Grey literature matters because negative and inconclusive findings are less likely to reach journals. Omitting it biases your review toward positive or statistically significant results.
Where to Look
| Source type | Examples | What you get |
| Theses and dissertations | ProQuest Dissertations and Theses, CORE, OpenAIRE | Full studies, often never published elsewhere |
| Trial registries | ClinicalTrials.gov, ISRCTN, WHO ICTRP | Ongoing, completed, and unreported trials |
| Preprint servers | medRxiv, bioRxiv, SSRN, PsyArXiv | Findings ahead of peer review |
| Government and agency reports | WHO IRIS, CDC, NICE, CADTH Grey Matters | Guidelines, evaluations, national statistics |
| Conference material | Embase conference records, society websites | Abstracts and posters, often years before publication |
| Institutional repositories | University archives, OpenDOAR, Zenodo | Working papers, datasets, technical reports |
Search Tactics for Websites and Repositories
- Keep queries short. Most site search boxes cannot handle nested Boolean logic, so use 2 or 3 words.
- Use site:who.int or site:nice.org.uk in a web search engine to search 1 organization at a time.
- Add filetype:pdf to obtain reports rather than news pages.
- Browse the publications section of key organizations directly, since their internal search is often weak.
- Contact authors and topic experts for unpublished work, and record who you contacted and when.
Documenting Grey Literature Searches
- Record the site name, URL, date searched, terms used, and the number of results screened.
- Decide in advance how deep you will go, such as the first 100 results, and apply that rule everywhere.
- Save a copy or a web archive link of anything you cite, because agency documents are moved and withdrawn.
- Report grey literature separately in the PRISMA flow diagram, under “records identified from other sources”.
What Is Citation Chaining?
Citation chaining means following citations away from a known paper: backward into its reference list, and forward into the papers that have since cited it.
Backward vs Forward Chaining
- Backward chaining reads the reference list of a key paper to find earlier work, including material your terms never matched.
- Forward chaining finds newer papers citing that key paper, which is the fastest route to recent developments.
- Do both from every included study, not only from the papers you started with.
- Chaining is a supplement to database searching, never a replacement, because it inherits the biases of the paper you start from.
Tools That Automate It
- Scopus and Web of Science offer cited reference searching with the cleanest citation data.
- Google Scholar provides free forward chaining through “Cited by”.
- citationchaser, a free and open-source R package, runs backward and forward chaining in bulk from a list of DOIs.
- Semantic Scholar and Lens.org provide open citation data and free bulk export.
When is Citation Chaining Most Important?
Citation chaining is always meant to complement database searching, but is especially important in the following situations:
- The concept has no agreed name, so keyword searching cannot work reliably.
- The field is interdisciplinary and scattered across databases you cannot all access.
- You are building or testing a theory, as in a realist review.
- You need to verify recall: if chaining surfaces eligible papers your search missed, the search needs revision.
How to Set Inclusion and Exclusion Criteria for a Literature Search
Criteria define what counts as eligible before you see the results. Setting them afterward invites bias, because you are then choosing studies rather than applying a rule.
Frameworks: PICO, PCC, SPIDER, SPICE
| Framework | Best for | Elements |
| PICO | Clinical effectiveness questions | Population, Intervention, Comparator, Outcome |
| PCC | Scoping and mapping reviews | Population, Concept, Context |
| SPIDER | Qualitative and mixed-methods questions | Sample, Phenomenon of Interest, Design, Evaluation, Research type |
| SPICE | Service, program, and policy evaluation | Setting, Perspective, Intervention, Comparison, Evaluation |
| CMO | Realist reviews | Context, Mechanism, Outcome |
Criteria Domains to Define Before Screening
- Population: age, condition, severity, setting, and how each will be verified.
- Intervention and comparator: dose, duration, delivery, and acceptable variations.
- Outcomes: which outcomes make a study eligible, and which are merely of interest.
- Study design: which designs are included, and whether protocols and abstracts count.
- Date and language, each with a stated reason.
- Publication status: whether preprints, theses, and conference abstracts are eligible.
Title and Abstract Screening vs Full-Text Screening
- At title and abstract stage, exclude only on the basis of criteria visible in that text, and keep anything uncertain.
- Move to full text whenever the abstract is missing, ambiguous, or silent on a key criterion.
- Record a single reason for exclusion at full-text stage; PRISMA requires these counts by reason.
- Use 2 independent screeners where possible, and agree in advance how disagreements are resolved.
Piloting Criteria and Recording Reasons for Exclusion
- Pilot the criteria on 30 to 50 records before full screening begins.
- Measure agreement between screeners and rewrite any criterion that produces frequent disputes.
- Log every change to the criteria, with the date and the reason, and report changes made after the protocol was registered.
- Keep the excluded full texts and their reasons; reviewers frequently ask to see them.
How to Evaluate Sources
Evaluation runs alongside screening. A study can be perfectly on topic and still too weak to inform your conclusion, so relevance and quality are separate judgments.
Authority, Currency, Relevance, and Methods
- Authority: author expertise, institutional affiliation, funding, and declared conflicts of interest.
- Currency: publication date relative to changes in practice, technology, or policy in that field.
- Relevance: how closely the population, intervention, and outcomes match your question.
- Methods: sampling technique, sample size, comparator, blinding, attrition, and whether the analysis matches the design.
- Transparency: registration, an available protocol, and accessible data or code.
- Consistency: whether the findings align with the wider body of evidence, and what explains the difference if not.
Critical Appraisal Tools by Study Design
| Study design | Tool | Focus |
| Randomized trial | Cochrane RoB 2 | Randomization, deviations, missing data, measurement, reporting |
| Non-randomized intervention study | ROBINS-I | Confounding, participant selection, classification of interventions |
| Cohort or case-control study | Newcastle-Ottawa Scale | Selection, comparability, outcome or exposure assessment |
| Systematic review | AMSTAR 2 | Protocol, search adequacy, risk of bias handling, synthesis |
| Diagnostic accuracy study | QUADAS-2 | Patient selection, index test, reference standard, flow and timing |
| Qualitative study | CASP Qualitative Checklist | Design, recruitment, reflexivity, rigor of analysis |
| Mixed-methods study | MMAT | Quality of both strands and the quality of their integration |
Checking for Retractions and Corrections
- Search the Retraction Watch Database before relying on any older study.
- In PubMed, look for the Retracted Publication notice on the record, and check for published errata.
- Enable retraction alerts in Zotero, or use the Crossmark and Crossref data shown by most publishers.
- Recheck your included studies immediately before submission, since retractions appear continuously. Paperpal’s Reference Checker is a good resource for this.
Search Strategies by Review Type
The review type determines how sensitive your search must be, how many sources you need, and what you have to report. Match the strategy to the method before you run a single line.
| Review type | Search approach | Typical scope | Documentation |
| Scoping or mapping review | Broad and highly sensitive; few limits; iterative as concepts emerge during screening | 3 or more databases, plus grey literature and registries | PRISMA-ScR; publish the full strategy for every database |
| Rapid review | Deliberately narrowed; date, language, and design limits agreed with the commissioner in advance | 1 to 2 major databases, plus 1 guideline or grey literature source | Cochrane RREC guidance; state every shortcut and its likely effect |
| Integrative review | Sensitive database searching plus purposive chaining across quantitative, qualitative, and theoretical work | 2 to 4 databases spanning the relevant designs | Whittemore and Knafl stages; explain how differing designs were combined |
| Realist review | Theory-driven and iterative; cluster searching around key papers; stops when the theory is saturated | Purposive rather than exhaustive; databases, grey literature, and expert contact | RAMESES standards; justify each iteration and the point of saturation |
- Sensitivity matters most in scoping reviews; feasibility matters most in rapid reviews.
- Iterative searching is a feature of scoping and realist reviews, not a sign of poor planning, but each iteration must be recorded.
- Every review type requires the same core reporting: databases, dates, full strategies, and result counts.
How Do You Document and Report a Search?
Record the database, platform, date run, the full strategy line by line, and the number of results, at the moment you run it. A search you cannot rerun cannot be reported.
What Should You Record for Each Database?
- Database name and the platform or interface used, since syntax differs between them.
- Date the search was run, and the date range covered.
- The complete strategy, copied and pasted, including every line number and filter.
- Number of records retrieved, before and after deduplication.
- Any limits applied, with the reason for each.
- Who designed the search, and who peer reviewed it.
PRISMA-S and Flow Diagram Basics
- PRISMA-S is a 16-item extension covering how searches are reported, and it pairs with the PRISMA 2020 statement.
- Place full strategies in an appendix or supplementary file, not paraphrased in the methods section.
- The flow diagram tracks records from identification through screening to inclusion, with exclusion reasons at full-text stage.
- Report database records and other-source records separately, including those found by chaining.
- Consider a PRESS peer review of the strategy before you run the final search.
Reproducibility Checklist
- Could another researcher rerun your search from the appendix alone?
- Is every result count in the flow diagram traceable to a specific search line?
- Are all limits and filters stated, with justifications?
- Are grey literature sources listed with URLs and dates?
- Is the search date close enough to submission to satisfy a reviewer?
Frequently Asked Questions
How Long Does a Literature Search Take?
A background scan takes 2 to 4 hours. A systematic review search commonly takes 2 to 6 weeks from first concept map to final export, including peer review of the strategy and grey literature searching.
- Concept mapping and synonym gathering: 1 to 3 days.
- Building and testing strings across databases: 3 to 10 days.
- Grey literature and citation chaining: 2 to 5 days.
- Deduplication and export: 1 day.
What Do I Do if I Get Too Many Results in My Search?
Add a concept with AND before you start cutting terms. Most oversized result sets come from searching 2 concepts where the question actually has 3.
- Revisit your research question or topic. “Exercise and mobility” is too broad; you should focus your research question on “does low-impact exercise improve mobility in community-dwelling older adults”.
- Move free-text terms from [tw] into [tiab], which removes matches that appear randomly in the full text of a paper but aren’t the real focus of that paper.
- Replace loose word pairs with phrases in quotation marks, or with a proximity operator.
- Restrict subject headings to major topic, but only if you can accept the loss in sensitivity.
- Check each concept block separately: 1 runaway OR block is usually responsible.
- Tighten over-broad truncation, such as cardi*, which pulls in cardiology, cardiac, and cardinal.
- Apply a justified date limit last, and report it.
How Many Databases Should I Search for a Systematic Review?
You should search a minimum of 3, and health reviews conventionally use MEDLINE, Embase, and CENTRAL. Add subject databases such as CINAHL or PsycINFO where the topic requires them, plus trial registries and grey literature.
Is Google Scholar Good Enough for Academic Research?
It is good enough for discovery, orientation, and citation chaining. It is not good enough as the sole source for a review, a thesis chapter, a journal article, or any work that must be reproducible, because its coverage is undisclosed and its results are unstable.
How Do I Find Free Full-Text Articles Legally?
- Check PubMed Central and the free full text link on the PubMed record.
- Use Unpaywall or the Open Access Button, which locate legal author copies in repositories.
- Search the Directory of Open Access Journals for fully open titles.
- Look at “All versions” in Google Scholar for a repository copy.
- Request a copy from the corresponding author, who is usually permitted to share it.
- Use your library’s interlibrary loan service for anything still paywalled.
How Do I Know When to Stop Searching for Literature?
Stop when new sources stop producing new eligible records. This is saturation, and it should be observed rather than assumed.
- A new database returns almost nothing you have not already screened.
- Citation chaining from included studies surfaces no new eligible papers.
- Your known-item test papers are all retrieved by the final strategy.
- You have searched every source your protocol committed to.
Can I Use AI Tools for a Literature Search?
Yes, for generating synonyms, drafting strings, and prioritizing screening, but not as the search itself. AI literature search does not provide reproducible coverage, and some fabricate references.
- Useful for suggesting terminology and alternative phrasings you had not considered.
- Useful for screening prioritization in tools such as Rayyan or Covidence.
- Verify every reference an AI tool produces against the database record.
- Disclose any AI assistance if the journal or institution requires it.
- Use a purpose-built literature search tool like R Discovery to supplement your database search, not replace it.
How Do I Search PubMed for Systematic Reviews Only?
Apply the Systematic Review article type filter, or add “systematic review”[pt] to your search line. Both approaches rely on indexing, so combine them with a search of the Cochrane Library.
- The publication type “systematic review” was introduced in 2019, so older reviews may not carry it.
- The broader subset tag systematic[sb] retrieves more, with lower precision.
- Search Epistemonikos or PROSPERO to find reviews in progress and avoid duplicating effort.
Comment