Search Filters vs. Better Query Terms: Which Should You Fix First?

Fix the query before you tighten the filters. When a search returns too few usable studies, the first suspect should be how concepts are expressed and combined, not how many records you have excluded. Metadata filters act on what a database or index has already recorded about a record; query repair changes what the search is looking for in the first place. If a relevant study is described with different vocabulary, sits under a different subject heading, or is indexed inconsistently, a filter applied early can remove it before you ever see it. Cochrane's guidance to build searches from eligibility criteria and to combine free-text terms with subject headings points in the same direction: get the concept layer right, then restrict.

This article compares the two moves on the same task — finding the eligible studies for a manuscript, review, or grant evidence summary — and shows where each fails. It then gives a decision table and a combined workflow that keeps an auditable method trail.

Take a concrete task: you are assembling the evidence base for a review section and need a defensible set of eligible studies. Two levers are available.

Metadata-based restriction narrows a result set using recorded attributes — date limits, field tags, filters, or similar record-level constraints. PubMed documents field tags, filters, Advanced Search, and saved searches, and its Search details show how terms were translated and combined. Google Scholar documents date limits and date sorting alongside author and title searching. These are restriction tools: they operate on the record as described.

Concept repair changes the search itself — the terms, synonyms, phrase forms, and subject headings that represent your eligibility criteria. Cochrane recommends planning searches from eligibility criteria and combining free-text terms with subject headings. That is concept work: making the query match the many ways a study might be described.

Both moves aim at the same outcome — a smaller, more relevant set — but they act at different layers. Restriction trusts the metadata; repair questions whether the concept was captured at all.

A filter can only act on what is recorded. If a relevant study is indexed under a different subject heading, described with a synonym you did not use, or captured in a source whose metadata is thin, a restrictive filter may drop it silently. The record is not wrong; the query never reached it in a form the filter could preserve.

Cochrane's recommendation to search multiple sources and to combine free-text terms with subject headings reflects this risk. A single vocabulary or a single index will not capture every eligible study, and a filter layered on top of an incomplete concept narrows an already incomplete set. The failure is quiet: you see fewer results and may read that as precision, when it is actually loss.

PubMed's Search details are useful here because they reveal how your terms were translated and combined. If the translation shows your concept was mapped narrowly, that is a signal to repair the query before adding restrictions. Google Scholar's Related articles and Cited by features offer a different check: they let you move from a known relevant paper to adjacent work, which can surface studies your terms missed.

Restriction is the right move when the concept is already well captured and you need to bound the set. If your query returns a broad but relevant pool, date limits and field tags can make it manageable. PubMed's filters and Advanced Search support this, and saved searches let you keep a stable, documented query. Google Scholar's date sorting and date limits serve the same bounding purpose.

Restriction also helps when you need reproducibility. A saved search with explicit filters is easier to document and rerun than an ad hoc string. Cochrane recommends documenting the search process and rerunning searches before publication; explicit filters make that trail legible.

Repair is the right first move when results are sparse, when known relevant papers are missing, or when your eligibility criteria involve concepts with unstable vocabulary. Cochrane's guidance to build from eligibility criteria and to combine free-text terms with subject headings is the core method. In practice, that means listing the concepts in your criteria, generating synonyms and phrase variants, and pairing free-text terms with subject headings rather than relying on either alone.

PubMed's Similar Articles and citation lookup can help you test whether a known paper's vocabulary is represented in your query. Google Scholar's author and title searching, plus Related articles, let you probe whether a specific work is reachable at all. If a paper you know is eligible cannot be found, the concept layer needs repair before any filter is applied.

Run both moves against the same eligibility criteria and judge them on the same dimensions.

• Recall of known eligible studies. Concept repair is designed to reach studies described in varied vocabulary; restriction can only preserve what the query already found. Test with a small set of papers you know should appear.
• Precision of the result set. Restriction reduces volume quickly; repair may broaden the set before you narrow it. Expect repair to increase results and restriction to decrease them.
• Auditability. Both can be documented. PubMed's Search details and saved searches make the translation and combination visible; Cochrane's documentation guidance applies to whichever move you make.
• Reproducibility. Explicit filters and saved searches are easy to rerun. Concept repair is reproducible when you record the terms, synonyms, and headings you used.
• Risk of silent loss. Restriction carries the higher risk when metadata is inconsistent; repair carries the risk of an unwieldy set that still needs bounding.

A defensible sequence moves from concept to restriction, not the reverse.

1. State eligibility criteria. Write the concepts your review or grant summary requires. Cochrane's guidance starts here.
2. Build the concept layer. List free-text terms, synonyms, and subject headings for each concept, and combine them. Record the versions you try.
3. Test against known papers. Confirm that a handful of studies you know are eligible can be found. If not, repair before restricting.
4. Add restrictions deliberately. Apply date limits, field tags, or filters only after the concept layer reaches your known papers. Note each restriction and its rationale.
5. Search multiple sources. Cochrane recommends multiple sources; vocabulary and metadata differ, so test the concept layer in each.
6. Document and deduplicate. Record the query, the source, the date, and the restrictions. Cochrane recommends deduplicating records and screening in stages.
7. Link reports and rerun. Cochrane recommends linking multiple reports of one study and rerunning searches before publication.

For the search step itself, WisPaper search provides a scholarly search interface where you can run and refine queries. When you need a public entry point to finding academic papers, WisPaper Scholar Search explains the search capability. As you settle on a concept layer and begin collecting candidates, My Library lets you save or upload papers and organize them, which keeps the candidate set in one place while you document decisions. If you want to ask questions grounded in the papers you have collected, Library QA supports questions based on papers uploaded to My Library or added from search results.

When a reviewer asks why a study was excluded, the answer should point to a documented decision, not a filter setting. If your method section says only that you applied date limits, a reader cannot tell whether a relevant study was filtered out or never found. Record the concept layer — terms, synonyms, headings — alongside the restrictions, so the trail shows both what you looked for and what you bounded.

For a revision, rerun the search before resubmission, as Cochrane recommends, and note any changes to terms or filters. If a reviewer names a missing study, first test whether your concept layer reaches it; only then decide whether a restriction should be relaxed. That order keeps the fix at the right layer.

• Treating a small result set as precision. Fewer results after a filter may mean loss, not relevance. Check known papers first.
• Repairing the query but never bounding it. Concept repair can produce an unwieldy set; plan the restriction step.
• Applying filters per source without rechecking vocabulary. Metadata differs across sources; a filter that works in one may hide records in another.
• Leaving no trail. Record terms, synonyms, headings, filters, sources, and dates so the method is reproducible.