A database search fails quietly. You type the words you already use for your topic, get a thin list of results, and assume the literature is thin. Usually the problem is vocabulary: your field calls the concept one thing, the database indexes it another way, and the papers you need are sitting one synonym away. This guide walks through building a search vocabulary from your eligibility criteria, pairing free-text synonyms with subject headings, testing the combination, and knowing when the search is ready to run and document.
Why does your first search return so few relevant papers?
The most common starting state is a search built from a single phrase lifted from your research question. That phrase reflects how you talk about the topic, not how authors or indexers describe it. Cochrane's guidance on searching for studies begins with eligibility criteria: the search should be planned from what you have decided counts as relevant, not from a convenient keyword. That reframing is the first decision. Before you collect synonyms, write down the concepts your question contains and what would make a study eligible or ineligible.
A second common cause is mixing free-text searching with subject-heading searching without noticing. Free-text terms match words in titles, abstracts, and author keywords. Subject headings are controlled vocabulary assigned by an index. PubMed, for example, supports field tags and shows search details that reveal how your terms were translated and combined, so you can see whether the database understood your phrase the way you intended. If you have never opened that translation view, you are guessing at what your query did.
What artifacts should you have before you touch a database?
Three artifacts make the rest of the work mechanical rather than improvised.
- A concept table. One row per concept in your question, with a column for free-text synonyms and a column for subject headings.
- An eligibility note. A short statement of inclusion and exclusion criteria, which Cochrane treats as the foundation for planning a search.
- A search log. A running record of each query, the database, the date, and the number of records returned, so the process can be documented and rerun.
If you are starting a fresh project rather than patching an old search, set up a workspace first so the log and the papers you keep stay together. WisPaper's Research Projects area is a place to create a project for this work, which keeps the vocabulary building and the resulting reading in one container instead of scattered across notes.
How do you turn a research question into a concept table?
Break the question into its independent ideas. A question about a treatment, a population, and an outcome has three concepts; each becomes a row. For each row, list every way you have seen the idea expressed: the formal term, the abbreviation, spelling variants, older terminology, and the phrase a practitioner would use. Cochrane's guidance is to combine free-text terms with subject headings, which means each row needs both kinds of entry.
Keep the table deliberately over-inclusive at this stage. You are not deciding what to search yet; you are collecting candidates. The pruning happens after you test.
Where do subject headings actually come from?
Subject headings are assigned by the database, so you find them inside the database rather than inventing them. PubMed's Advanced Search and its search-details view let you see how terms were translated, which is how you discover the controlled heading behind your free-text phrase. When you enter a term and the translation shows a mapped heading, that heading is a candidate for your table.
A practical route is to find two or three known-relevant papers first, then inspect how they are indexed. The headings attached to papers you already trust are strong candidates, because they describe literature you have decided is on target. Record each heading next to the free-text synonyms for the same concept, and note which database it belongs to, since headings do not transfer cleanly between databases.
How do you find synonyms without drowning in them?
Mine the papers you already have. Read the abstracts of your known-relevant set and copy the phrases authors use for your concept, including the ones that differ from yours. Cochrane's guidance on linking multiple reports of one study is a reminder that the same work can appear under different descriptions, so variant vocabulary is expected rather than a sign of a messy topic.
Use citation relationships to expand sideways. Google Scholar documents Cited by and Related articles features, which let you move from a paper you trust to papers that cite it or sit near it. Those neighbors often use the vocabulary of a different subcommunity, which is exactly the synonym you are missing. PubMed's Similar Articles feature serves the same purpose inside that database.
When you want to search a scholarly interface directly for candidate papers and their terminology, WisPaper's scholarly search is the entry point for finding academic papers; pull the abstracts of promising hits and harvest their phrasing into your concept table.
How do you combine free-text terms and subject headings in one query?
Build the query concept by concept. Within a row, connect synonyms and the subject heading with OR so that any variant can match. Between rows, connect with AND so that every concept must be present. This is the structure Cochrane describes when it recommends combining free-text terms with subject headings.
Then test the combination against your known-relevant papers. If a paper you are certain belongs in the review does not appear, a synonym or heading is missing from the table. If the result set is enormous and full of off-topic records, a term is too broad or a concept is under-specified. PubMed's field tags let you aim a term at a specific field when a word is matching in the wrong place, and filters let you narrow by attributes rather than by vocabulary.
Record each version of the query in your log with its date and result count. That record is what makes the search reproducible and what lets you rerun it later.
Worked example: a winter-break cleanup of a thesis chapter search
Suppose a student is revising a thesis chapter over winter break and the supervisor's feedback says the literature base looks thin. The chapter's question concerns a specific intervention in a specific setting.
Step 1 — concepts. Two rows: the intervention and the setting. The student writes the formal intervention name, its abbreviation, and the phrase used in practice notes. For the setting, the student writes the formal term and two colloquial variants.
Step 2 — known-relevant set. The student selects three papers already cited in the chapter and treats them as the test set.
Step 3 — headings. Searching each known paper in PubMed and opening the search details reveals the controlled headings attached to them. Those headings go into the table beside the free-text synonyms.
Step 4 — first query. Synonyms OR-ed within each row, rows AND-ed together. The query is run and the three known papers are checked against the results. One is missing, so the student adds the phrase that paper uses in its abstract and reruns.
Step 5 — sideways expansion. Using Related articles and Cited by on the strongest known paper, the student finds a cluster using a different term for the intervention and adds it to the table.
Step 6 — save and organize. Promising records go into a library so the reading list and the search log stay linked. WisPaper's My Library is where papers can be saved and organized, which keeps the harvested set separate from the rest of the project.
Step 7 — document. The final query, database, date, and count go into the log, and the concept table is saved with the chapter files.
How do you know the search vocabulary is ready?
Use explicit completion criteria rather than a feeling that the list is long enough.
- Every concept in the question has its own row, with at least one subject heading and several free-text synonyms.
- Every paper in your known-relevant test set is retrieved by the query.
- The result count is large enough to contain the field but small enough to screen, and you can explain which term is doing the narrowing.
- The query has been run in each database you plan to use, because headings differ between them.
- The search log records the final query, the database, the date, and the count.
- The concept table is saved alongside the project files so the next revision starts from it.
Cochrane also recommends documenting the search process and rerunning searches before publication, so treat the log as a living artifact rather than a one-time note.
How do you screen and deduplicate what the search returns?
Cochrane describes deduplicating records and screening in stages. In practice, export the results, remove duplicates, then screen titles and abstracts before reading full texts. Keep a record of why records were excluded at each stage so the process is auditable.
When you need to ask questions across the papers you have collected, WisPaper's Library QA supports asking questions based on papers uploaded to My Library or added from search results. That is useful at the screening stage, when you want to check whether a batch of saved abstracts addresses your concept before committing to full-text reading.
How do you keep the vocabulary current across a long project?
Vocabulary drifts. New terms appear, and your own understanding of the concept sharpens as you read. Set a checkpoint: before each major revision or submission, rerun the saved queries, check the new records against your criteria, and update the concept table with any new synonyms or headings you encountered while reading.
PubMed supports saved searches and email alerts, and Google Scholar offers email alerts, both of which turn the rerun into a recurring task rather than a manual one. For longer stretches of work where the literature review itself is the task, WisPaper's Scholar Agent is the area for extended research tasks including literature review, which can run alongside the manual search rather than replacing the log you keep.
Reusable checklist
| Stage | Artifact | Done when |
|---|---|---|
| Frame | Eligibility note | Inclusion and exclusion criteria are written down |
| Decompose | Concept table | Each concept has its own row |
| Collect | Synonym and heading columns | Each row has free-text variants and at least one heading per database |
| Test | Known-relevant set | Every test paper is retrieved |
| Tune | Query versions in the log | You can name the term doing the narrowing |
| Run | Final queries per database | Query, database, date, and count are recorded |
| Screen | Deduplicated record set | Screening stages and exclusion reasons are logged |
| Maintain | Saved searches or alerts | A rerun date is set before submission |




