03 · What You Need to Know
The Difference Between Search Refinement and Conceptual Drift
A Research Question and a Search Strategy Are Not the Same Thing
The research question describes what you want to investigate. The search strategy is an information-retrieval representation designed to find literature relevant to that question.
They should correspond, but they do not need to contain exactly the same words or even exactly the same number of searchable elements.
Suppose your question concerns the effects of online learning on academic performance among university students. The question contains several substantive elements. Your database strategy may need multiple expressions for each concept, appropriate subject headings, and database-specific syntax. It may also omit an element that is poorly represented in searchable metadata and can be assessed more reliably during screening.
Research question
Defines the substantive phenomenon, population, relationship, intervention, outcome, context, or other scope of the inquiry.
Search strategy
Operationalizes enough of that scope in database-searchable form to retrieve potentially relevant records with an appropriate balance of sensitivity and precision.
The search is therefore an instrument for finding evidence, not a word-for-word transcription of the question.
Search Refinement Changes the Retrieval Representation
Consider a concept such as online learning. During exploratory searching, you discover that relevant literature also uses e-learning and online education.
Changing:
"online learning"
to:
("online learning" OR e-learning OR "online education")
does not necessarily change the substantive concept. It gives the same intended concept additional linguistic routes into the result set.
This is search refinement.
It reflects the process of discovering how researchers use different terms for the same concept.
Conceptual Drift Changes What Can Count as Relevant
Now imagine changing the online-learning block to:
("online learning" OR e-learning OR "online education" OR technology)
Technology is much broader. It could include virtual reality, robotics, learning management systems, classroom response systems, mobile devices, simulations, and many other technologies unrelated to online learning.
The new term does not merely improve access to the original concept. It allows a broader concept to satisfy the search.
That is conceptual drift.
Search refinement
Changes the retrieval mechanism while preserving the substantive information need.
Conceptual drift
Changes the substantive boundaries of what can qualify, often without an explicit decision to revise the research question.
The Most Useful Question Is: What New Studies Can Qualify Now?
When evaluating a search modification, do not ask only whether the result count increased or decreased.
Ask:
After this change, can a study qualify that would have been outside my original research scope?
If the answer is no, the modification may simply improve retrieval. If the answer is yes, investigate whether you have intentionally changed the scope.
The reverse question is equally useful:
After this change, can a study that still satisfies my research question no longer qualify for retrieval?
If yes, you may have introduced an unnecessary restriction.
Changing Synonyms Usually Does Not Change the Question, but It Can
Adding or removing synonyms is often ordinary search refinement. The important qualification is whether the terms are genuinely alternative representations of the same intended concept.
Suppose you search:
adolescent OR teenager
Adding youth may seem straightforward, but the age range represented by youth can vary across disciplines and policy contexts. If your eligibility criteria define adolescents narrowly, the additional term may retrieve broader populations.
That does not necessarily mean the term should be excluded. It means you should distinguish linguistic coverage from conceptual expansion.
A term can be useful for sensitive retrieval even when it is somewhat broader, provided screening can reliably remove records outside the intended population. The decision should be conscious rather than accidental.
Broader Search Terms Are Not Automatically Conceptual Errors
This point deserves care.
A search term can be broader than the eligibility concept yet still be useful because database records do not always express concepts with sufficient specificity.
Suppose your population is undergraduate students. Some relevant records may say only university students in the title and abstract, with undergraduate status described in the full text.
Searching university students may therefore be appropriate even though the term is broader than the final eligibility criterion.
The search is deliberately sensitive, and screening restores the intended boundary.
This differs from silently deciding that postgraduate students now belong in the research question.
Eligibility Criteria and Search Terms Operate at Different Levels of Specificity
A common mistake is expecting the search to enforce every eligibility criterion precisely.
Database metadata may not reliably report characteristics such as exact participant age, intervention duration, outcome definition, setting, diagnostic threshold, or subgroup eligibility.
Cochrane's current Handbook cautions against searching every element of a review question when doing so may reduce sensitivity. Outcomes and comparators, for example, may not be consistently represented in titles, abstracts, or indexing.
A search can therefore be broader than the final eligibility criteria without changing the research question.
| Stage |
Role |
Typical Level of Specificity |
| Research question |
Defines the substantive inquiry |
Conceptually precise |
| Search strategy |
Retrieves potentially relevant records |
May deliberately use broader or alternative representations to preserve sensitivity |
| Title/abstract screening |
Removes clearly ineligible records |
Applies more of the eligibility criteria using human judgment |
| Full-text screening |
Determines final study eligibility |
Applies detailed criteria using the complete report |
Adding Another AND Concept Is One of the Easiest Ways to Change the Search
Suppose your question concerns online learning among university students. The search retrieves 8,000 records, so you add academic performance:
(online learning terms) AND (university student terms) AND (academic performance terms)
The result count falls dramatically.
Was the search improved?
Only if academic performance was genuinely required by the research question or review scope. If the project concerns online learning broadly, adding the outcome means studies of engagement, satisfaction, persistence, motivation, accessibility, or other outcomes can no longer qualify.
The search became smaller because the information need became narrower.
Watch Out
Do not add an AND concept merely to control workload. Every new concept creates another condition that records must satisfy. If the condition was not part of the intended evidence scope, you have changed the question rather than merely refined the search.
Removing an AND Concept Does Not Necessarily Broaden the Research Question
The opposite can also be misunderstood.
Suppose an outcome is part of the eligibility criteria but is inconsistently reported in titles and abstracts. Removing the outcome block from the database strategy may increase retrieval substantially.
That does not necessarily mean the review now accepts all outcomes. The outcome criterion can remain unchanged and be applied during screening.
The search became broader; the research question did not.
This distinction is central to sensitive literature searching.
Changing a Search Field Usually Changes Retrieval, Not the Concept
Suppose you initially search generative AI in titles only and discover that relevant studies use the term in abstracts but not titles.
Changing from Title to Title/Abstract generally does not change the concept. It changes where the database is allowed to find the same expression.
Likewise, restricting a highly ambiguous term from a broad field to titles may improve precision while preserving the intended meaning.
Field changes are therefore typically search refinements, provided they do not systematically make a substantive part of the intended literature impossible to retrieve.
As always, verify what the database's fields actually contain.
Phrase Searching Can Improve or Distort the Representation
Suppose your concept is social media. Searching the established phrase "social media" may represent the concept more accurately than allowing social and media to match independently.
That is a sensible refinement.
But an exact phrase can become too restrictive when legitimate wording varies. If authors write professional development for teachers, searching only "teacher professional development" may exclude relevant literature.
In that case, relaxing the phrase or using proximity does not necessarily broaden the research question. It improves linguistic coverage of the same concept.
Changing Proximity Distance Usually Changes Linguistic Flexibility
Suppose you require teacher within two words of burnout. Relevant articles frequently use expressions such as burnout among secondary school teachers.
Increasing the distance from two words to five may recover legitimate variations while preserving the same conceptual relationship.
However, widening proximity indefinitely eventually approaches a simple AND search, where the words can co-occur without a meaningful textual relationship.
So proximity refinement requires empirical testing. The question is whether the distance continues to represent the intended linguistic relationship.
Correcting Truncation or Wildcards Is Usually Technical Refinement
If randomi?ed is intended to capture randomized and randomised in a platform whose wildcard supports that behavior, the underlying concept has not changed.
Likewise, replacing several explicit word forms with a tested truncation stem can simplify the strategy without altering the information need.
But a truncation stem that retrieves unrelated words can introduce conceptual noise. Technical syntax is not conceptually neutral when its expansion reaches different meanings.
This is why phrases, truncation, and wildcards should be tested rather than treated as mere formatting.
Adding a Controlled-Vocabulary Term Usually Adds Another Route to the Same Concept
Suppose your free-text block represents depression and you identify the appropriate database subject heading for depression.
Adding that heading with OR generally creates another retrieval route to the same intended concept:
(subject heading for depression OR depression keywords)
This is ordinarily search refinement rather than a change in the research question.
But inspect the scope note and hierarchy. If the heading is broader than your intended concept and automatically includes narrower concepts outside your scope, the subject search may retrieve literature beyond the conceptual boundary.
Changing Subject-Heading Explosion Can Affect Conceptual Scope
Suppose a broad controlled-vocabulary term automatically includes several narrower terms. Some clearly belong to the research question; others may not.
Turning explosion on or off changes which indexed concepts can qualify through that route.
This can be a technical adjustment or a substantive scope decision depending on the hierarchy.
Inspect the narrower terms individually. Do not decide based solely on whether explosion produces a convenient number of records.
Changing Databases Does Not Change the Question
Searching an additional appropriate database usually expands information-source coverage rather than the research scope.
A study of educational technology might reasonably require databases covering education, psychology, computing, or multidisciplinary scholarship. The question remains the same even though each database contributes different literature.
What must change is the technical implementation. Controlled vocabularies, field codes, proximity operators, truncation, and automatic processing need to be adapted.
This is why search strategies should be translated between databases while preserving the underlying concepts.
Adding Another Database Can Reveal a Conceptual Problem
Although the database itself does not change the question, searching another disciplinary literature may expose terminology or conceptual distinctions that challenge your original formulation.
For example, education researchers and computer scientists may use the same term differently. You may discover that two concepts you treated as synonyms are distinct in one field.
At that point, you face a substantive decision. Either preserve the original scope and adjust the terminology accordingly, or revise the conceptual framework of the project.
Do not hide that decision inside a new OR term.
Search Results Are Evidence About the Search, Not Automatic Evidence About the Question
Researchers often refine searches based on what they see in the results. That is appropriate. Relevant and irrelevant records provide empirical information about terminology and indexing.
The danger is allowing the retrieved literature to redefine the question merely because some topics are easier to find.
Suppose your intended concept retrieves few studies while a neighboring concept has a large literature. It can be tempting to broaden the term until the results become more satisfying.
The availability of literature is relevant to feasibility, but it should not silently determine the construct you claim to study.
Too Many Results Should Trigger Diagnosis, Not Scope Reduction
When a search returns thousands of records, inspect why.
One broad OR term, an ambiguous acronym, excessive truncation, full-text searching, incorrect parentheses, or a loose AND relationship may be responsible.
Correcting those problems can improve precision without changing the substantive question.
Adding a new population, outcome, setting, date restriction, or other concept solely to reduce the count is different.
The distinction is central when deciding what to do with an overly broad search.
Too Few Results Should Trigger Retrieval Expansion Before Scope Expansion
The same principle applies in reverse.
If a search returns almost nothing, first check synonyms, subject headings, historical terminology, fields, exact phrases, proximity distances, filters, Boolean structure, and database coverage.
Do not immediately replace the specific concept with a broader one simply to generate results.
As discussed when a search returns too few results, broaden the retrieval route before broadening the research question.
Date Limits Can Quietly Change the Evidence Question
Suppose your original question concerns the effectiveness of an intervention without a temporal restriction. You search all available years and retrieve a large literature. You then restrict the search to studies published since 2020 because screening is difficult.
The revised search now answers something closer to:
What evidence has been published since 2020?
That may be a legitimate question, but it is not necessarily the original one.
Date restrictions are appropriate when they follow from a substantive rationale, such as the introduction of a technology, policy, diagnostic definition, or the scope of an update review. They should not be invisible workload controls.
Language Limits Can Also Alter the Accessible Evidence Base
Restricting a search to English does not change the conceptual topic, but it changes which evidence is eligible to enter the review process.
For some systematic review methodologies, such restrictions are explicitly discouraged because they can introduce bias. Cochrane's current standards require searches for intervention reviews not to be restricted by language or publication status.
For other projects, language constraints may be unavoidable. If so, describe them as limitations or eligibility decisions rather than treating them as neutral search refinements.
Full-Text Availability Is Usually an Access Constraint, Not a Research Concept
Limiting results to "full text available" can make a project easier to complete. It can also remove relevant evidence simply because access arrangements differ.
Unless accessibility itself is part of the research question, this filter does not improve conceptual precision. It changes the accessible evidence base according to licensing or repository availability.
This is a particularly clear example of a practical convenience masquerading as a search criterion.
Publication-Type and Study-Design Restrictions Can Be Legitimate but Consequential
If the research question is explicitly about randomized controlled trials, applying an appropriate validated trial filter is consistent with the intended evidence design.
If you initially intended to include multiple quantitative designs but later restrict the search to randomized trials because there are too many results, the evidence question has changed.
The same distinction applies to qualitative studies, conference proceedings, dissertations, reviews, and other document or design types.
Do Not Confuse Search Feasibility With Conceptual Validity
A question can be conceptually sound but produce a search that is expensive to screen. Another question can be easy to search but scientifically uninteresting.
Practical constraints matter. Time, staffing, database access, language capability, and screening resources can legitimately influence project design.
The important point is to make those trade-offs explicitly.
If feasibility requires narrowing the research question, revise the protocol or research plan accordingly. Do not leave the question unchanged in the manuscript while quietly narrowing it through database filters.
Keep an Invariant Statement of the Information Need
One practical way to prevent drift is to write a short statement describing what must remain conceptually true throughout search refinement.
For example:
Search Invariant
What must remain true?
Retrieve research concerning the use of generative artificial intelligence in assessment within higher education, regardless of the specific generative-AI tool, academic discipline, or measured educational outcome.
Now evaluate modifications against that statement.
Adding ChatGPT as another generative-AI term is consistent with the invariant. Adding academic performance with AND would violate it if the search is supposed to include studies regardless of outcome. Restricting to nursing students would violate it if all higher-education disciplines remain eligible.
The invariant does not replace formal eligibility criteria. It is a compact guardrail against accidental conceptual drift during iterative search development.
Use a Concept Table to Separate Stable Concepts From Changing Terms
A working concept table can make search evolution easier to track.
| Stable Concept |
Current Retrieval Terms |
Reason for Change |
Question Changed? |
| Generative AI |
generative AI, ChatGPT, large language model terms |
Relevant literature revealed additional terminology |
No, if all terms represent the intended concept |
| Higher education |
higher education, university students, college students |
Population terminology varies |
No, subject to eligibility boundaries |
| Assessment |
assessment, evaluation, formative assessment terms |
Terminology refined after record inspection |
Needs checking if evaluation is broader than intended assessment concept |
| Outcome |
None required in search |
Outcome assessed during screening |
No, if outcome eligibility remains unchanged |
The final column is particularly useful. It forces you to ask whether a search change is merely technical or whether the substantive scope moved.
Keep a Search Change Log for Consequential Revisions
For a substantial review, record important search changes during development.
A simple log might contain:
- what changed;
- why it changed;
- what effect it had on retrieval;
- whether known relevant records were gained or lost;
- whether the underlying eligibility scope changed.
This does not require documenting every typo or experimental query. Focus on decisions that materially affect retrieval.
The log becomes particularly valuable when several people are developing the search or when the final strategy is revisited months later.
Change One Important Element at a Time
If you simultaneously remove an acronym, add a subject heading, widen a proximity distance, change the field, and apply a date limit, you cannot tell which modification caused the change in results.
A more disciplined workflow is:
Preserve the baseline Save the current strategy and result count.
State the problem For example: one acronym is producing large amounts of irrelevant retrieval.
Make one targeted change Restrict the acronym to a more informative field.
Compare Inspect the records lost and retained.
Check the invariant Determine whether the same substantive studies remain eligible for retrieval.
Document and continue Keep the change if it improves retrieval, then move to the next issue.
This makes refinement interpretable rather than merely iterative.
Known Relevant Records Can Detect Accidental Narrowing
If you have several articles that clearly fit the intended scope, use them as test records.
After a consequential refinement, ask whether those papers remain retrievable in databases that index them.
If one disappears, investigate why. Perhaps the new title restriction is too narrow, the proximity distance is too strict, or a synonym you removed was its only retrieval route.
Known-item testing does not prove completeness, but it can expose obvious divergence between the search and the question.
Inspect Newly Retrieved Records for Accidental Broadening
The opposite test is equally important.
After adding a new synonym or broadening a field, inspect records that were not retrieved before. Do they represent legitimate alternative expressions of the original concept, or has the search entered a neighboring literature?
This gives you evidence about whether the change improved sensitivity or altered scope.
Search Refinement Is Expected in Iterative Searching
A search strategy need not emerge perfectly formed before you inspect the literature.
Cochrane's guidance describes search development as an iterative process in which terms can be modified based on what has already been retrieved. Relevant records can reveal new terminology, indexing, and retrieval problems.
The methodological concern is not that the search changes. It is whether changes are deliberate, defensible, and consistent with the intended information need.
Protocol Amendments Are Different From Search Refinements
For a preregistered or protocol-driven review, changes to the substantive scope may require formal documentation as amendments.
Correcting a database field code or adding a legitimate synonym generally does not constitute the same kind of change as revising the population, intervention, outcome, date range, or eligible study designs.
The exact reporting requirements depend on the review methodology and registration platform. The useful principle is to distinguish technical search development from substantive protocol modification.
Do Not Pretend a Necessary Scope Change Never Happened
Sometimes the original question genuinely needs revision.
Perhaps the intended population is not represented in enough studies to support the planned synthesis. Perhaps the concept was defined incorrectly. Perhaps exploratory searching reveals that two constructs assumed to be equivalent are meaningfully different.
Changing the question can be the correct methodological decision.
The problem is not revision. The problem is allowing the revision to happen invisibly through search terms while continuing to describe the original question as though nothing changed.
Ask Three Questions Before Keeping a Major Search Change
The three-question drift check
1. Does this change represent the same concept more effectively?
If yes, it is likely ordinary search refinement.
2. Can studies outside the original scope now qualify, or can studies inside the original scope no longer qualify?
If yes, examine whether the change has altered substantive boundaries or merely adjusted sensitivity around imperfect metadata.
3. Would I need to change the eligibility criteria or wording of the research question to justify this modification?
If yes, treat it as a potential scope or protocol change rather than merely a search refinement.
These questions are more useful than asking whether the result count moved in the desired direction.
The Best Search Is Not Necessarily the Most Stable-Looking Search
A search that never changes may simply be a search that was never tested properly.
Terminology discovery, thesaurus inspection, known-item testing, database translation, and examination of irrelevant retrieval should all be capable of improving the strategy.
Stability should exist at the conceptual level. The technical search should be allowed to evolve until it represents that stable information need as effectively as practical.