---
name: academic-literature-search-strategy
description: >
  Builds a systematic, documented literature search from a research question:
  concepts and synonyms, Boolean strings, database selection, inclusion criteria,
  citation chasing, grey literature, and a reproducible search log. Use for
  "how do I search for literature", "I cannot find any sources on my topic",
  "build me a search string", "which databases should I use", "what are my
  keywords", "how do I document my literature search", "how do I know when to
  stop searching", "my search returns thousands of results".
category: 15 Academic University Research
ref: "15.02"
tier: 3
inherits: [K2, K3, K4, K5]
---

# Academic Literature Search Strategy

## 1. One-line description

A method for turning a research question into a documented, reproducible literature search: concept decomposition and synonym building, Boolean construction, database selection, pre-set inclusion criteria, citation chasing, and a search log that lets a marker or a colleague repeat exactly what you did.

## 2. What this skill is used for

**The research problem it solves.** Students search the way they search for anything else: type the topic into whatever box is nearest, read the first page, stop. Four failures follow predictably. The search returns either nothing or forty thousand items, and neither number is informative. Whole bodies of relevant work are missed because they use a different word for the same thing, particularly older work and work from an adjacent discipline. The sources that end up in the review are the ones that were easiest to find rather than the ones that matter most, so the review is shaped by search-engine ranking. And nothing is recorded, so the search cannot be repeated, cannot be defended, and cannot be extended six weeks later without starting again. Documenting the search is frequently an explicit marking criterion, and it is the criterion most often lost by students who did adequate searching and kept no record of it.

**Where it sits in the research lifecycle.** After the question is committed and before the literature review is written. It runs again, in reduced form, before submission, to catch work published during the project.

**Typical use cases.**
- Building the search for a dissertation or honours literature review.
- Rescuing a search that returns nothing, or returns thousands of irrelevant items.
- Finding the vocabulary a field actually uses, including the terms used before the current ones.
- Choosing databases for a question that sits between two disciplines.
- Producing the documented search log a marking rubric requires.
- Establishing, defensibly, that little or nothing has been published on a question.
- Updating a search before submission without redoing it from scratch.

**Who uses it.** Undergraduate and honours students building their first substantial review; supervisors diagnosing why a student's reading is thin or one-sided; academic librarians and skills staff working with students at scale. Assume no prior knowledge of Boolean operators, controlled vocabulary or database structure.

## 3. When to use it

- You have a committed research question and need to find what has been published on it.
- Your searching so far has been typing the topic into a search box and reading what came up.
- You are returning either almost nothing or an unmanageably large set, and cannot tell which is the real situation.
- You suspect you are missing literature, for example because sources you find keep citing work you never retrieved.
- Your rubric requires a described or documented search strategy, a search log, or a statement of inclusion criteria.
- Your question spans two disciplines and you do not know where the relevant work is indexed.
- You need to justify, in writing, that a topic is under-researched.
- You are near submission and need to check for work published since you searched.

## 4. When NOT to use it

- **The project is a formal systematic review.** This skill produces a rigorous, documented and repeatable search for a normal narrative or thematic literature review. It is not a systematic review method. A systematic review requires a protocol written and ideally registered before searching, pre-specified eligibility criteria and outcomes, exhaustive rather than sufficient searching, dual independent screening with an agreement measure, formal risk-of-bias assessment, and reporting to a recognised standard. If the work will be described as systematic, hand over to **15.06 Systematic Literature Review**, or to **15.14 Advanced Systematic Review and Meta-Analysis** where quantitative pooling is intended. In the other direction, if you are running 15.06 and only need the search mechanics, they are here, but 15.06 governs the protocol, the screening design and the reporting. Do not retrofit the word "systematic" onto a search built with this skill; the log will not support the claim, and the claim is checkable.
- **The question is not yet fixed.** Searching against a moving question wastes the search and produces a log nobody can interpret. Go back to **15.01 Academic Research Topic Selection**, commit, and then search. The exception is a deliberate scoping search to test whether a question is viable, which is short, time-boxed, and labelled as scoping in the log.
- **You need the evidence appraised and synthesised, not found.** This skill finds and documents; it does not judge source quality or resolve disagreement between sources. Appraisal and synthesis for an applied or commercial review is **10.01 Literature Review and Desk Research**; writing the review as an argument rather than a list is **15.05 Academic Writing Structure and Argumentation**.
- **You are looking for a specific known item.** If you already have an author and title, this is retrieval, not searching, and building a concept structure around it wastes an hour. Go to the source directly, then use it as a starting point for citation chasing.
- **The literature genuinely does not exist in indexed form.** For very local, very recent, or practice-based questions, the relevant material may sit in policy documents, professional guidance, institutional archives, community records or practitioner writing. A database search will return nothing and the correct conclusion is not "there is no literature". Shift the source map to grey literature and documentary sources, and say what you did in the method.
- **The task is verifying references in a document you already have.** That is citation checking, and it belongs to **15.03 Referencing and Citation Management**.
- **The user wants a list of sources produced for them without searching.** No search means no sources. See §12.1 and §12.2. A reference list assembled without retrieval is fabrication in bulk regardless of how many entries happen to be real.

## 5. Required inputs

**Required.**
- **A committed research question**, with its population, phenomenon and context visible. The concepts are extracted from these; without them there is nothing to decompose.
- **Access to at least one indexed academic database.** Almost every institution provides these through its library. If the only available tool is a general web search engine, say so in the method, expect a different and weaker evidence base, and adjust the claims the review makes accordingly.
- **The discipline**, because it determines vocabulary, which databases matter, and what counts as a legitimate source type. Do not assume it.

**Optional, and what each one adds.**
- **Two or three relevant items already known to be good.** The single most useful optional input. They supply real vocabulary, they let you test whether a draft string retrieves what it should, and their reference lists start the backward chase.
- **The institution's database list or a subject guide.** Turns database selection from guesswork into a short list, and reveals subscriptions the student did not know they had.
- **A date limit required by the discipline or the supervisor.** Lets you exclude on age at the search stage rather than discovering the scale of the older literature at screening.
- **Language capability.** Determines the honest coverage statement. A search in one language on a question studied in several has a coverage limitation (K3 §6) and it belongs in the method, not hidden.
- **The rubric wording about the search.** Some rubrics require a described strategy, some a full log, some a flow of counts. Knowing which prevents both under-documenting and spending days on a diagram nobody asked for.
- **Whether a librarian appointment is available.** Frequently the highest-value hour in the whole project, particularly for controlled vocabulary and for cross-disciplinary questions.

## 6. Questions to ask before starting

1. **What discipline, and does the question cross into another?** Determines vocabulary, database choice and what counts as evidence. A cross-disciplinary question needs at least one database from each side, because indexing is discipline-bound. Default: ask. Never assume a discipline from the topic, because the same topic is studied differently in several.
2. **Is this search for a narrative review or a systematic one?** This decides which skill governs and how much is enough. Default: assume narrative, state it, and say explicitly that the search is documented but not exhaustive.
3. **What does the rubric require you to show about the search?** Determines the artefact: a paragraph, a table, or a full log with counts. Default: keep the full log regardless, since a log can be summarised into a paragraph but a paragraph cannot be expanded into a log.
4. **What time window is defensible for this question?** Determines the date limit. Default: no date limit at the first pass, then apply one only if volume demands it and say why. A default of "last five years" applied without reason routinely excludes the foundational work the field is built on.
5. **What languages can you read, and is the relevant work published in others?** Determines the coverage statement. Default: state the languages searched as a limitation.
6. **How many weeks until the review must be drafted?** Determines the depth achievable and where to stop. Default: assume the search must be substantially complete in two to three weeks, because the writing takes longer than students expect.
7. **What is your institution's AI use policy for this assessment?** Governs how this skill may be used at all, and what must be declared. Default: assume assistance must be declared and no generated text may be submitted (§12.1).

## 7. Step-by-step methodology

**Step 1. Extract the search concepts from the question, and expect two to four.**
Underline the substantive ideas in your question. Each becomes a concept block. A question about the influence of workload on teacher retention in rural secondary schools yields three blocks: workload, retention, and rural schools. Two rules govern the count. Fewer than two blocks means you are searching a topic rather than a question and will drown. More than four almost always over-constrains, because every added block multiplies the ways an otherwise relevant item can fail to be retrieved. Where a question has four or more, drop the least essential from the search string and apply it at screening instead, which is far more forgiving than making it a retrieval requirement. *Correct result: two to four named concept blocks, with any deliberately deferred concept recorded as a screening criterion.*

**Step 2. Build the synonym set for each block, across communities and across time.**
For each block list: everyday terms, the technical term in your discipline, the technical term in adjacent disciplines, plural and spelling variants including national spelling differences, acronyms and their expansions, and the terms used in older literature. This last item is the one students skip and the one that costs most: fields rename their constructs, and searching only the current term silently excludes the foundational decade. Sources of real vocabulary, in order of usefulness: the keyword lists of the two or three good items you already have; the controlled vocabulary or subject headings of the database itself, which is a curated list of the terms the index actually uses; and the titles of recent review articles. Never invent a synonym set from general impression; harvest it from real records. *Correct result: a table with one row per block and a populated synonym list per row, each term traceable to where you found it.*

**Step 3. Set inclusion and exclusion criteria before you search, and write them down.**
Criteria set after searching are criteria fitted to what you found, which is the mechanism by which a review quietly excludes the inconvenient literature. Specify: population, phenomenon or intervention, setting or context, study designs admitted, publication types admitted (peer-reviewed articles, book chapters, theses, conference papers, reports), date range with a reason, and language. State each as a decision rule a second person could apply to the same record and reach the same verdict. *Correct result: a written criteria list, dated before the first search, with a reason recorded for any restriction that narrows the evidence base.*

**Step 4. Construct the string, generically, then adapt it per database.**
Four mechanics apply across almost every academic database, though the exact syntax varies and must be checked in that database's own help pages.
- **Boolean.** Combine synonyms within a block with OR, and combine blocks with AND. This is the whole architecture: OR widens, AND narrows. NOT is a last resort, because it removes any record containing the excluded term, including records where it appears incidentally, and it is a common cause of silently losing relevant work.
- **Phrases.** Enclose multi-word concepts so they are searched as a unit rather than as separate words, otherwise a two-word concept matches documents where the words appear pages apart.
- **Truncation and wildcards.** A truncation symbol at a word stem retrieves its variants in one term. Truncate carefully: a stem cut too short retrieves a large unrelated set, and this is the commonest cause of a search returning tens of thousands of irrelevant records.
- **Proximity and field limits.** Proximity operators require two terms within a specified distance, which is more precise than AND and less rigid than a phrase, and they are the single most effective tool when a search is too broad. Field limits restrict matching to title, abstract or subject headings rather than the full text; searching title and abstract is the usual default for a review, because full-text matching retrieves anything that mentions your term once in passing.
Write the string with the blocks visible, one per line, so it can be edited and logged. *Correct result: a readable block-structured string, plus the adapted version actually run in each database.*

**Step 5. Select databases deliberately, and use more than one.**
Databases differ in what they index, and no database indexes everything. Choose by discipline: at minimum one broad multidisciplinary index and one or two subject-specific indexes for your field, plus one from the adjacent discipline if the question crosses. Understand the difference in kind between a general web search engine and an indexed academic database. A general search engine ranks by relevance signals you cannot inspect or reproduce, has no controlled vocabulary, no field structure, and no stable result set, so the same search run twice can return different results and the search cannot be documented in a way anyone can repeat. An indexed database has a defined scope, curated subject headings, structured fields, and a result count that means something. Use general search tools for citation chasing, for grey literature and for finding a known item, and use indexed databases for the searching you will report. Record which databases you used and, just as importantly, which you did not and why. *Correct result: a named database list with a one-line reason for each, and the coverage gap acknowledged.*

**Step 6. Pilot the search before running it properly.**
Run the string in one database and inspect it rather than harvesting it. Three checks. First, known-item retrieval: do the two or three items you already know are relevant appear? If not, find out which block excluded them and fix that block, because if your string cannot retrieve the papers you already have, it will not retrieve the ones you do not. Second, precision: scan the first fifty titles and estimate what proportion are plausibly relevant. Under roughly a tenth means the string is too broad and needs a proximity operator, a field limit, or a tighter truncation. Third, volume: a result set in the low tens for a question you believe is studied usually means an over-constrained string, most often a fourth concept block or a NOT. *Correct result: a string that retrieves the known items and returns a set you can actually screen, with each adjustment recorded.*

**Step 7. Run the searches and log every one, including the ones that fail.**
Open the log before the first search, not after the last. For every search record: database, exact string as run, any limits applied, date run, number of results, and the number taken forward. Record searches that returned nothing, because those rows are what make an absence claim defensible. This log is the artefact that turns "I could not find much" into a professional statement about a specific set of searches, and it is frequently marked directly. It is also what lets you re-run the search before submission in an hour instead of a week. *Correct result: a complete log, in one place, updated as you go rather than reconstructed afterwards.*

**Step 8. Screen in two passes and record the reason for every exclusion.**
First pass on title and abstract against the criteria from Step 3, which is fast and errs toward inclusion, because a wrongly excluded item is never recovered while a wrongly included one is caught in the second pass. Second pass on the full text, where most real exclusions happen, because titles routinely conceal a different population, a different construct, or a study design that does not answer your question. Use a small fixed set of exclusion reasons (wrong population, wrong context, wrong construct, wrong design, not retrievable, duplicate, not in a language searched) and record counts at each stage. Categorical reasons let you see the shape of what you removed, and expose the case where most of the disconfirming literature has been excluded on a criterion that will not survive a supervisor's question. *Correct result: counts of identified, screened, full-text assessed, included and excluded by reason.*

**Step 9. Chase citations backwards and forwards.**
Database searching finds what the index and your vocabulary allow. Citation chasing finds what the field itself considers connected, and it routinely retrieves work no string reaches. **Backwards**: take your best five or six included items and mine their reference lists, which is the most efficient route to the foundational literature and to older vocabulary. **Forwards**: find what has cited those items since, which is how you reach the recent work and the critiques. Run both to at least one full round, and record what each round added. When a round of chasing produces only items you already have, that is meaningful evidence about coverage. *Correct result: a record of which items were chased, in which direction, and what each round yielded.*

**Step 10. Decide on grey literature deliberately, and say what you decided.**
Grey literature is material published outside commercial or academic channels: government and agency reports, policy documents, dissertations and theses, conference papers, working papers, professional and standards-body guidance, and organisational reports. Include it where the question is applied, recent, policy-related, geographically specific, or where the practitioner literature leads the academic one, which is common in professional and applied fields. Exclude it where the discipline's convention is peer-reviewed sources only, but exclude it explicitly, not by omission. Grey literature is not lower-quality by definition, it is unrefereed, which means the appraisal burden shifts to you. Search it separately, log it separately, and mark it in the source list. *Correct result: a stated decision with a reason, and grey sources identifiable as such.*

**Step 11. Recognise saturation, then stop and say how you knew.**
Stop when three things are true together: new searches with new vocabulary return items you have already seen; a round of citation chasing adds nothing new; and each of your concept blocks has been searched in at least two databases and in at least two vocabularies. One of these alone is not saturation. Repeatedly retrieving the same items from the same string is a fact about the string, not about the literature, and it is the most common false stopping signal. Where saturation is reached with very few items, that is a finding: the honest statement is what was searched and what was not found, phrased against the log, never as "there is no research on this". Then write the method paragraph directly from the log: databases, date of searching, concepts and synonyms, string structure, limits, inclusion criteria, screening counts, citation chasing, grey literature decision, and stated limitations. *Correct result: a stopping decision with its evidence named, and a method paragraph a marker can follow and a colleague could repeat.*

## 8. Analytical framework

Two structures, one for building the search and one for accounting for it.

**The concept grid**, which is the search:

    Block A (synonyms joined by OR)
      AND
    Block B (synonyms joined by OR)
      AND
    Block C (synonyms joined by OR)

Applying it: OR is your recall control and AND is your precision control, and every problem with a search is a problem with one of them. Too few results means a block is too thin, so add synonyms, or a block is unnecessary, so remove it. Too many results means the blocks are too loose, so tighten with field limits, proximity or a truncation that was cut too short. Diagnose which block is responsible by running the blocks separately and looking at the counts, rather than adjusting the whole string at once. A search you cannot debug block by block is a search you cannot defend.

**The retrieval funnel**, which is the account of the search:

    Identified → De-duplicated → Title and abstract screened →
    Full text assessed → Included → Added by citation chasing → Final set

Each transition carries a number and a reason. The funnel is what makes the search reproducible: a reader who has the log and the funnel can see not only what you found but what you discarded and on what grounds. The most informative point in the funnel is the full-text stage, because a large drop there means the search string is retrieving the wrong construct, and the fix is upstream in the vocabulary, not downstream in the screening.

## 9. Output format

**1. Question and concept table.** The question, with the concepts underlined, and one row per block listing the synonyms and where each came from.

**2. Inclusion and exclusion criteria.** Written as decision rules, dated before searching, with a reason for every restriction.

**3. Search log.**

| # | Database | Exact string as run | Limits | Date run | Results | Taken forward |
|---|---|---|---|---|---|---|

Every search appears, including those returning zero.

**4. Database rationale.** Databases searched with a reason each, and databases not searched with a reason each.

**5. Screening record.** Counts at each funnel stage, and exclusions grouped by standard reason.

**6. Citation chasing record.** Items chased, direction, and what each round added.

**7. Grey literature decision.** Included or excluded, with the reason, and where grey sources were sought.

**8. Included source list.** With grey literature marked, and every entry retrieved and opened.

**9. Method paragraph.** The prose version for the dissertation, written from the log.

**10. Limitations.** Languages searched, databases unavailable, date restrictions, and an explicit statement that the search is documented and repeatable but not exhaustive.

**When the search returns little, the format must not force fabrication (K4 §1).** A thin included list stays thin. The log grows instead, because the value of a near-empty result set lies entirely in the record of what was searched to establish it. Never pad an included list with items that failed the criteria, never list a source that was not retrieved and opened, and never write "the literature suggests" where the literature you actually retrieved does not.

## 10. Quality checks

Run before the review is drafted. These sit on top of K4 §8.

1. Does the string retrieve the items you already knew were relevant?
2. Has each concept block been tested separately, so you know which one is controlling the result count?
3. Does every block contain terms from more than one vocabulary, including older terminology?
4. Were the inclusion criteria written before searching, and are they dated?
5. Was more than one database searched, and is the reason for each choice recorded?
6. Is the exact string as run recorded for each database, rather than a tidied version?
7. Are zero-result searches in the log?
8. Has citation chasing been run in both directions, with at least one full round?
9. Is the grey literature decision explicit rather than implied by absence?
10. Has every included item actually been retrieved and opened, not just seen as a record?
11. Is every exclusion assigned one of the standard reasons, and do the counts add up?
12. Are absence claims phrased against the log, naming the searches that support them?
13. Is the search described as documented rather than systematic, unless 15.06 governs it?
14. Could a second person run your log and arrive at approximately your included set?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **Topic searching** | The search box contains the whole question as a sentence | Decompose into concept blocks and combine with AND (Step 1) |
| **Single-vocabulary search** | Every retrieved item uses the same term for the construct | Harvest synonyms from real records and subject headings (Step 2) |
| **Missing the older literature** | Nothing retrieved predates the current terminology | Search the terms the field used before (Step 2) |
| **Too many blocks** | A precisely phrased search returning almost nothing | Three blocks maximum in the string; defer the rest to screening |
| **Over-truncation** | Tens of thousands of unrelated records | Lengthen the stem and check what the wildcard is matching |
| **NOT used casually** | Relevant items missing for no visible reason | Avoid NOT; exclude at screening instead |
| **One database only** | The reference list is dominated by one publisher or journal family | Search a subject index and a multidisciplinary index (Step 5) |
| **General search engine treated as a database** | The method cannot state a result count or be repeated | Use indexed databases for reported searching (Step 5) |
| **Criteria fitted after the fact** | Exclusion criteria that neatly remove the awkward studies | Write and date the criteria before searching (Step 3) |
| **No log** | The search cannot be described, repeated or updated | Open the log before the first search (Step 7) |
| **False saturation** | The same string keeps returning the same items | Saturation requires new vocabulary and new databases too (Step 11) |
| **Absence overstated** | "There is no research on this" after two searches | Phrase absence against the log and name where you looked |
| **Records mistaken for sources** | A citation drawn from an abstract the full text of which was never opened | Nothing enters the included set unretrieved (Step 8, and 15.03 §12) |
| **AI-supplied source list** | Plausible references that no database returns | Every source comes from a logged search; see §12.2 |

## 12. AI guardrails

Skill-specific. The universal prohibitions in K4 apply in full and are not repeated.

1. **Academic integrity.** These skills assist a researcher's thinking, structure and rigour. They do not produce work to be submitted as the student's own unaided output. The user must comply with their institution's AI use policy and its declaration requirements, which vary by institution and by assessment. Where an institution prohibits AI assistance for a task, this skill must not be used for it. The skill never writes a passage for submission as though the student wrote it; it interrogates, structures, critiques and teaches. Operationally in this skill: search strings, synonym sets and screening criteria are legitimate scaffolding and are the student's to test, adapt and defend. The searching itself, the reading and the method paragraph are the student's work, and a student who cannot explain why a term is in their string does not have a search strategy.

2. **Never supply sources, references or a reading list from model knowledge.** Every item in an included list comes from a search that was actually run and a document that was actually opened. A list of plausible-looking references is fabrication in bulk however many entries happen to exist, and it is the single most damaging failure available in this task (K4 §2.4, and **15.03** §12).

3. **Never report a search you did not run.** Do not state that a database was searched, do not report a hit count, and do not narrate a screening process unless it happened. Where you cannot access a database, record it in the log as not searched, with the reason. A fabricated log is worse than no log, because it converts an honest gap into a false claim about method.

4. **Never claim that a topic is under-researched, unexplored or a gap.** You can state what a named set of searches did or did not retrieve. That is a fact about the searches. Whether a gap exists is a claim about the literature that only an executed and documented search can support, and even then only in the terms the log allows.

5. **Never describe a search built with this skill as systematic.** Documented, repeatable and transparent are accurate. Systematic is a technical term with protocol, dual-screening and reporting obligations, and misusing it in a method section is a substantive error that a marker will catch (§4, and **15.06**).

6. **Never present a database record, abstract or search snippet as though the source had been read.** Retrieval is not reading, and an abstract routinely misdescribes what a paper actually did. Where the full text could not be obtained, the item is either excluded with the reason "not retrievable" or carried explicitly as abstract-only, never silently.

7. **Never invent controlled vocabulary or subject headings for a specific database.** Subject heading systems are real, curated and checkable, and a plausible-sounding heading that does not exist produces a search that returns nothing and a student who thinks the literature is empty. Describe how to find the real headings in the database's own thesaurus instead.

8. **Where the search was constrained by access, language or database availability, say so next to the conclusions it affects** (K3 §6, K4 §4.3). A coverage limitation reported only in an appendix has been hidden.

## 13. Best-practice principles

- **Search the vocabulary, not the topic.** The single largest determinant of what a search finds is the synonym set, and the single most common reason a student finds nothing is that the field calls it something else.
- **Harvest terms, never invent them.** Take vocabulary from the keyword fields of real records and from the database's own thesaurus. Invented synonyms feel comprehensive and retrieve nothing.
- **Three concept blocks is usually the ceiling.** Each additional block is another way for a relevant item to be missed. Extra constraints belong at screening, where a human is judging, not in the string, where a machine is matching.
- **Debug block by block.** When a search misbehaves, run each block alone and look at the counts. The problem is almost always in one block, and adjusting the whole string at once hides which.
- **Citation chasing finds what strings cannot.** The reference list of one excellent paper is often worth more than a day of database work, because it encodes what the field itself thinks is connected.
- **The log is the deliverable, not the by-product.** It is what is marked, what makes absence defensible, and what makes the pre-submission update cheap. Keeping it costs about a minute per search and there is no way to reconstruct it later.
- **Write the criteria before you know what you will find.** Criteria written afterwards are indistinguishable from criteria fitted to the desired conclusion, and the difference is visible to an experienced marker.
- **Date limits need a reason.** "The last five years" is a habit, not a rationale, and in most fields it excludes the work everything else is responding to.
- **Grey literature is unrefereed, not inferior.** In applied and professional fields it is frequently the most current and most specific material available, and excluding it silently is a coverage decision made without acknowledgement.
- **A search that finds almost nothing is a result, if it is documented.** Undocumented, it is an admission. The log is the entire difference.
- **Re-run before you submit.** A short repeat of the recorded searches, filtered to the period since you last ran them, protects against the examiner who knows the paper published in the intervening months.

## 14. Worked example

Generic fictional scenario, academic, education.

**INPUT**

An honours student in education has a committed question: "How does administrative workload relate to teacher retention in rural secondary schools?" They have searched the phrase in a general web search engine, retrieved a mixture of news articles and two policy blogs, concluded that "nobody has researched this", and arrived believing the topic must be changed.

**PROCESS**

*Step 1.* Three concept blocks are extracted: administrative workload, teacher retention, rural secondary schools. A fourth candidate block, secondary phase specifically, is deliberately deferred to screening, because making the school phase a retrieval requirement would exclude studies covering multiple phases that report phase-level results.

*Step 2.* Synonym building is where the diagnosis happens. The retention block turns out to be the problem. The field writes about retention, but also about turnover, attrition, intention to leave, intention to stay, teacher mobility, and, in the older literature, wastage. The student's single term was retrieving perhaps a fifth of the relevant work. The workload block similarly expands to administrative burden, paperwork, non-teaching duties, accountability demands and bureaucratic load. Rural expands to remote, non-metropolitan, and, in some national literatures, hard-to-staff schools, which is a term with a different meaning that will need care at screening. Terms are harvested from the keyword fields of two known-good articles and from a subject thesaurus, not composed.

*Step 3.* Criteria are written and dated: empirical studies of qualified school teachers; any country, because restricting to one would leave almost nothing; published 1995 onwards, with the reason recorded that accountability-driven administrative burden is the phenomenon of interest and it is largely a post-1995 development; English language, recorded as a limitation; peer-reviewed articles plus government and agency reports, since teacher workforce data is often published by education departments rather than in journals.

*Step 4.* The string is built with the three blocks on three lines, phrases enclosed, "teach" truncated at a stem long enough to catch teacher and teachers and teaching without pulling in unrelated words, and matching limited to title, abstract and subject headings rather than full text.

*Step 5.* Three databases are selected: one subject index for education, one broad multidisciplinary index, and one covering public policy and administration, because the accountability side of the question sits partly in that literature. A general web search engine is retained for grey literature and citation chasing only, with the reason recorded that its results cannot be reproduced or counted.

*Step 6.* The pilot fails informatively. The two known-good articles are not retrieved. The cause is traced to the rural block: one of them describes its schools as "remote and regional" and the other as "non-metropolitan", neither of which was in the block. Both terms are added, and the known items now appear.

*The judgement call.* The rural block is by far the weakest, because there is no consistent international vocabulary for it and every added synonym reduces precision. The choice is between a tight rural block that will miss relevant work, and a loose one that will return several thousand records the student cannot screen in the time available. The resolution is to remove the rural block from the string entirely in the subject index, where the resulting set is large but tractable, and to apply rurality as a screening criterion instead. This is recorded in the log with its reason, so that the method paragraph can explain why the search string does not contain a concept that appears in the question, which would otherwise look like an error.

*Steps 7 to 9.* Searches are logged, including two that return nothing. Screening runs in two passes, with the second pass excluding a substantial number of items on "wrong construct", because a large literature on teacher workload measures total hours rather than administrative burden specifically. That pattern is itself a finding and is noted for the review. Backward chasing from four core papers adds six items, including two pre-2000 studies using the older term. Forward chasing adds three recent items, one of which directly critiques the main study the student had planned to build on.

*Step 10.* Grey literature is included, with the reason that national teacher workforce statistics and departmental retention reports carry data no journal article holds. It is searched separately and marked in the source list.

*Step 11.* Saturation is judged reached when a new vocabulary variant in the third database returns only known items and forward chasing adds nothing further. The original conclusion, that nobody had researched this, is replaced by a documented set of items, and the actual gap turns out to be narrower and more defensible: administrative burden is rarely separated from total workload in rural settings specifically.

**OUTPUT**

A concept table with sourced synonyms, dated criteria, a log of eleven searches across three databases with two zero-result rows, a screening funnel showing the large full-text drop and its reason, a citation-chasing record adding nine items, an included set of twenty-eight sources with grey literature marked, and a method paragraph that explains, defensibly, why the rural concept was applied at screening rather than in the string.

`RESEARCHER REVIEW RECOMMENDED` on the date limit and on the decision to move rurality to screening: both are defensible, both narrow what the review can claim, and both should be confirmed with a supervisor before the review is written (K5 §2.7).

## 15. Advanced usage

**Cross-disciplinary questions.** Where a question sits between fields, build two vocabulary sets rather than one merged set, and run the search separately in each discipline's index using that discipline's own words. Merging vocabularies into one string produces a search that is native to neither and retrieves the intersection rather than the union. The most interesting material in cross-disciplinary reviews usually comes from the field the student is not enrolled in.

**Search updating.** Keep the log in a form that supports re-running: exact strings, date last run, and result counts. Before submission, re-run each recorded search with a date filter starting from the last run date. This takes about an hour with a good log and is uninsurable without one.

**Diagnosing a thin review.** Where a supervisor says the reading is thin or one-sided, the fastest diagnostic is not more reading but an inspection of the concept table and the log. One-sidedness almost always traces to a single vocabulary, a single database, or a missing round of citation chasing, and all three are visible in the artefacts within minutes.

**Escalating to systematic.** Where a review turns out to matter more than expected, the upgrade path is explicit: write the protocol before further searching, make the search exhaustive rather than sufficient, add a second independent screener with an agreement measure, and report to a recognised standard. That is a different method with a different cost, and it is **15.06 Systematic Literature Review**. It cannot be applied retrospectively, because the protocol must precede the search.

**Where the standard approach does not fit.** For very recent phenomena, shift weight to forward citation chasing and preprint and conference material, and lower the expectation of peer-reviewed coverage explicitly. For questions rooted in a specific national or local context, expect the relevant material to sit in government, agency and institutional publications, and treat the database search as supporting rather than primary.

## 16. Skill chain

**Recommended previous skills:**
- **15.01 Academic Research Topic Selection.** Hands over the committed question, from which the concept blocks are extracted, and the targeted prior-answer check that this search replaces.

**Recommended next skills:**
- **15.05 Academic Writing Structure and Argumentation.** Takes the included set and turns it into a review that argues rather than lists, which is where most of the marks sit.
- **15.04 Research Proposal Writing.** Takes the search and its findings to establish the gap the proposal is built on.
- **15.06 Systematic Literature Review.** Takes over entirely where a protocol-driven, exhaustive, dual-screened method is required.

**Runs well alongside:**
- **15.03 Referencing and Citation Management**, which captures every retrieved item at the moment of retrieval and verifies every reference against the source.
- **10.01 Literature Review and Desk Research**, which supplies source appraisal and synthesis where the review must also serve an applied decision.
- **13.02 Source and Citation Verification**, which audits a finished reference set.

---
A Yazi Supplied Skill and resource.