Choosing keywords for a journal submission is not a one-time label-picking task. It is a research decision that affects how editors match reviewers, how indexing services classify the article, and how future readers discover the work. The practical question behind the search task is usually narrower: Which terms should I put in the keyword field right now, given my manuscript, my target journal, and the evidence I actually cite?
This article walks through that decision as a submission-readiness task. It shows how to move from a working manuscript to a defensible keyword set, how to test the set against real literature, and how to document the choices so a co-author or project lead can approve them without redoing the analysis.
The workflow treats keywords as an artifact with provenance: each candidate term comes from somewhere, each kept term has a reason, and the final set has a confidence level attached. A worked example at the end applies the method to a decision memo for a clinical research team, and the completion criteria tell you when the keyword field is ready to submit.
What Problem Are We Solving When We Choose Keywords?
The reader facing this task is usually in one of three states. The first is a first-time submitter with a finished manuscript and a blank keyword field. The second is a reviser whose journal reviewer asked for better keywords after the initial submission. The third is a research team lead consolidating multiple drafts into one traceable submission package. Each state has the same underlying need: a keyword set that is specific, discoverable, consistent with the cited literature, and defensible.
Keywords do two jobs at once. For the journal's workflow, they help the editorial office route the manuscript to the right handling editor and reviewers. For downstream discoverability, they become retrieval terms in indexing systems. In biomedical publishing, the National Library of Medicine's Medical Subject Headings (MeSH) is the controlled vocabulary used to index and search biomedical information. MeSH provides headings, entry terms, scope notes, qualifiers, and tree positions that help expand or focus a search. Authors in many biomedical journals are therefore choosing between natural-language keywords and controlled vocabulary terms, even when the journal does not require MeSH explicitly.
The key insight is that keyword selection starts with the manuscript's evidence base, not with a brainstorming session in the abstract. The cited sources and the terms already inside them reveal the vocabulary that the field actually uses.
Where Do Strong Keyword Candidates Come From?
Before testing anything, collect the terms that are already present in the manuscript and its reference list. This step turns keyword selection from an open-ended guessing game into a constrained extraction task.
Terms from the Abstract and Title
The title and abstract contain the manuscript's claim in its most compressed form. Extract the nouns and noun phrases that identify the population, intervention, outcome, method, and study type. For example, a manuscript on medication adherence in older adults with diabetes will surface candidates such as medication adherence, older adults, diabetes mellitus, and the study design term. These are the natural starting points, but they are rarely the final set on their own.
Terms from the Cited References
The reference list is the strongest signal of which vocabulary the manuscript's evidence base shares. Look at the titles and abstracts of the twenty most relevant cited works. The terms that recur across those sources are the terms reviewers and readers in that subfield are likely to search. If the manuscript cites ten papers that consistently say antihypertensive medication adherence rather than blood pressure pill compliance, the former is the better-aligned keyword even if the drafting team used the latter throughout the manuscript.
A practical way to do this at scale is to gather the manuscript's core references in a searchable workspace. For teams that want to save and organize the relevant papers while checking their terminology against one another, My Library provides a place to keep the source set together. When the comparison requires asking a specific question across those saved papers, Library QA lets the team query the papers directly rather than re-reading each one.
Terms from the Journal's Own Vocabulary
The International Committee of Medical Journal Editors (ICMJE) describes the standard parts of a biomedical manuscript and directs authors to follow the selected journal's current instructions for article type, format, required statements, and submission materials. That instruction extends to keywords: the journal's author guidelines often specify whether keywords must be MeSH terms, drawn from a particular list, or free text. Read that section before drafting the keyword list, because a technically strong set can still be noncompliant if the journal mandates a controlled vocabulary.
How Do Controlled Vocabularies Change the Decision?
If the target journal requires MeSH or expects biomedical indexing compatibility, keyword selection becomes a mapping exercise rather than a pure writing exercise. The National Library of Medicine's MeSH vocabulary includes headings, entry terms, scope notes, qualifiers, and tree positions. Each heading has a defined scope note that states what the heading includes and excludes, and entry terms list the synonyms that map to the heading. Tree positions show where the heading sits in the broader hierarchy.
The practical workflow is three steps. First, look up each natural-language candidate in MeSH to find its approved heading. Second, read the scope note to confirm that the heading actually covers the manuscript's meaning; two headings can sound similar while scoping different populations or outcomes. Third, check the tree position to see whether the selected heading is too broad or too narrow for the claim. A clinician may search broadly, but a keyword belongs to the specific article, not to the entire disease class.
This controlled-vocabulary step also produces better search testing later. PubMed's user guide explains that the platform supports field tags, phrase searching, filters, Advanced Search, saved searches, citation lookup, Similar Articles, and Search Details showing how terms were translated. Knowing the MeSH heading for a term lets a researcher test that term as a precise field query rather than a loose keyword match.
How Do I Test Whether My Candidates Will Actually Be Found?
The strongest keyword list is not the one that sounds best in the manuscript; it is the one that retrieves the manuscript's own reference neighborhood. The test is simple: search the literature using the candidate terms and check whether the results overlap with the manuscript's cited evidence base.
The retrieval test works like this. Take each candidate term or phrase and run it in a scholarly search interface with the same field filters the target journal's readers would use. Record what comes back. If a candidate retrieves the manuscript's core references and other highly relevant work, it is doing its discovery job. If a candidate returns mostly irrelevant studies, the term is either too generic or is being used differently in the field than in the manuscript. If a candidate returns almost nothing, it may be too narrow or idiosyncratic, even if the term is technically correct.
For a direct search of academic papers during this testing step, WisPaper search offers the scholarly search interface. The test should compare the retrieval set against the manuscript's own reference list, which is why keeping the reference set organized in My Library makes the overlap check faster: the team can see whether the search is returning the sources the manuscript already trusts.
How Do I Turn Candidate Terms Into a Defensible Keyword Set?
Once each candidate has evidence behind it, the decision becomes a structured comparison rather than a vote. Build a working table with four columns: candidate term, source of the term, retrieval result, and confidence. The source column records where the term came from—title, abstract, cited reference, MeSH heading, or journal instruction. The retrieval result records the overlap with the manuscript's reference neighborhood. The confidence column records how certain the team is that the term belongs in the final set.
A term earns a high confidence level when three conditions hold at once: it appears in the manuscript's own abstract or title, it appears in multiple cited references, and it retrieves relevant literature in a search test. A term earns a medium confidence level when two of the three hold, or when it is the correct controlled-vocabulary map but appears infrequently in the cited set. A term earns a low confidence level when it is technically acceptable but disconnected from the manuscript's evidence base or the journal's expected vocabulary. Low-confidence terms are usually cut unless the journal requires a fixed number of keywords, in which case the team documents the weakest remaining term as a compliance choice rather than a discovery choice.
The confidence level matters because keywords are a submission artifact that a team lead or co-author may need to approve. If the memo says only "keywords: diabetes, adherence, older adults," the approver cannot tell whether those were tested or guessed. If the memo says "diabetes mellitus type 2 (MeSH, retrieved 14 of 20 core references, high confidence)," the approver can verify the work without opening every source.
When Do I Need a Broader Literature Check?
Some keyword decisions cannot be settled by the manuscript's own reference list alone. The manuscript may be interdisciplinary, bridging two vocabularies that each have their own canonical terms. Or the target journal may be a general-interest venue where the reader's search vocabulary differs from the author's subfield vocabulary. In these cases, the keyword set needs a broader retrieval and terminology check before the decision closes.
A broader check starts with the same candidate fleet but widens the search scope. Run each candidate across the discipline's main search surfaces and record the first-page relevance, not just the raw count. If the manuscript bridges behavioral economics and medication adherence, the keywords should probably include one term from each discipline's core vocabulary. The selected terms should each pull in relevant work from their own side of the bridge. The wider check also reveals when a candidate is a near-miss: the phrase retrieves the right topic but the wrong study design, population, or setting.
For teams running this wider check as part of a longer research task, Scholar Agent fits the literature-review style use case. For teams that want to structure the keyword check alongside other submission work, Research Projects provides a project container for the manuscript, the candidate list, and the retrieval notes together.
What Does a Complete Submission-Ready Keyword Package Look Like?
The final deliverable is not a comma-separated list. It is a short decision memo that contains the keyword list and enough provenance for another researcher to audit. The memo has four sections.
First, the final keyword set in the exact order and format the target journal requires. Second, a one-line rationale for each keyword tying it to the manuscript's claim, the cited references, and the retrieval test. Third, the controlled-vocabulary mapping where applicable, including the MeSH heading and any entry terms checked. Fourth, the confidence level for each keyword using the high-medium-low scale defined above.
The memo should also record the cut list: candidates that were considered and rejected, with the reason. This is the section that makes the package reusable. When the journal asks for a revision, or when a co-author questions why a particular synonym was not included, the cut list answers the question without reopening the whole analysis.
Worked Example: A Decision Memo for an Adherence Study
Suppose a clinical research team is submitting a manuscript on the effect of a text-message reminder program on antihypertensive medication adherence in adults with type 2 diabetes. The draft team proposed six keywords off the top of their heads: reminders, blood pressure, pills, text messages, chronic disease, and compliance. The research lead runs the structured workflow and produces the following decision memo.
Final keyword submission (with provenance):
| Keyword | Source | Retrieval result | Confidence |
|---|---|---|---|
| Medication adherence | Cited references and MeSH map | Retrieves 16 of 20 core references | High |
| Diabetes mellitus, type 2 | MeSH heading mapped from abstract | Retrieves relevant clinical trials | High |
| Antihypertensive agents | MeSH heading mapped from cited set | Retrieves the medication class literature | High |
| Text messaging | Abstract and journal instruction | Retrieves the mHealth intervention set | Medium |
| Older adults | Cited references agreed on the term | Moderate overlap with core set | Medium |
Cut list: Reminders (too generic, retrieved appointment-reminder studies unrelated to this intervention), blood pressure (captured by the MeSH medication heading), pills (not the field's canonical term, cited sources consistently said medication), compliance (reviewer-facing connotation mismatch; adherence is the accepted terminology in the cited work), chronic disease (too broad, no retrieval advantage).
The memo shows why the team replaced five of the six draft terms. The one retained term, text messaging, moved from the draft list into final with a documented medium-confidence level because the retrieval test was strong but the manuscript's cited references used the landing-term mobile health in some sources. The team flagged that mapping note for the journal's keyword field or the cover letter if the journal allows a short note.
This memo is now reusable. If the journal asks for one vocabulary switch, the team can see exactly which terms were tested, which were cut, and what confidence level each surviving term carried. The provenance survives the decision.
How Do I Know the Keyword Set Is Ready to Submit?
The completion criteria are specific and checkable. The keyword set is ready when all of the following are true:
- Every keyword maps to a concept that is present in the manuscript's claim, not merely adjacent to it.
- The set follows the target journal's keyword instructions exactly, including required vocabulary lists, formatting, order, and capitalization.
- At least one search test has been run for each candidate, with the retrieval result recorded against the manuscript's own reference neighborhood.
- Each kept keyword has a documented source and a confidence level.
- The cut list exists and records at least one serious rejected alternative for a noncompliant set; a set with no cuts is a signal that the team did not actually compare alternatives.
- If a controlled vocabulary is applicable, the heading and entry-term mapping is recorded.
- The final list has been read once in its intended journal format by someone other than the drafter.
If any criterion fails, the submission package is not ready even if the manuscript itself is finished. The keyword field is the last decision that connects the article to its future readers, and it deserves the same audit trail as the methods section.




