---
name: referencing-and-citation-management
description: >
  Handles academic referencing and citation properly: choosing and applying the
  required style family, quoting versus paraphrasing versus summarising, avoiding
  the too-close paraphrase, page numbers and locators, secondary citations,
  reference list reconciliation, and verifying every reference against the actual
  source. Use for "how do I reference this", "which citation style should I use",
  "is this paraphrase too close to the original", "do I need a page number",
  "how do I cite something quoted in another source", "check my reference list",
  "are these AI-generated references real", "how do I avoid accidental plagiarism".
category: 15 Academic University Research
ref: "15.03"
tier: 3
inherits: [K2, K3, K4, K5]
---

# Referencing and Citation Management

## 1. One-line description

A working method for citing and referencing accurately in any style family, distinguishing quotation from paraphrase from summary and their different plagiarism risks, and verifying every reference against the actual source, because a reference that has not been checked against the document it names is not a reference.

## 2. What this skill is used for

**The research problem it solves.** Referencing is taught as compliance and therefore learned badly. Students treat it as a formatting tax paid at the end, which produces four distinct failures. The reference list is assembled in the last two days from memory, browser history and half-recorded notes, and is wrong in ways nobody can now reconstruct. Paraphrases sit too close to the original, which is the commonest form of unintentional plagiarism and is penalised whether or not intent existed. Claims are attributed to sources that make a different claim, because the citation was captured from another author's summary rather than from the source itself. And, increasingly, references arrive fully formed, perfectly formatted and entirely fictional, produced by a system asked for sources rather than asked to find them. This last failure is well documented, it is trivially detectable, and it has ended academic and professional careers. This skill installs capture at the point of reading, a verification gate no reference passes without being opened, a paraphrase test, and a two-way reconciliation of citations against the reference list.

**Where it sits in the research lifecycle.** Alongside everything, from the first source opened to the final proofread. It is the only skill in this track that never stops running, and every hour it is postponed multiplies its cost.

**Typical use cases.**
- Setting up a citation practice at the start of a project rather than repairing one at the end.
- Applying an unfamiliar required style consistently across a long document.
- Deciding whether a passage should be quoted, paraphrased or summarised.
- Testing whether a paraphrase is far enough from the original.
- Handling a source you have only encountered inside another source.
- Reconciling in-text citations against the reference list before submission.
- Checking whether references in a draft, particularly AI-assisted drafts, correspond to real documents that say what is claimed.

**Who uses it.** Undergraduate and honours students, particularly those who have never been taught referencing beyond a one-hour induction; students working in a second language, where the too-close paraphrase risk is structurally higher; supervisors and markers diagnosing citation problems in a draft. Assume no prior familiarity with style conventions and no confidence in what counts as plagiarism.

## 3. When to use it

- You are about to read your first source for a project and have no capture system.
- You have been told to use a style you have not used before, or your document mixes two.
- You are unsure whether something needs a citation at all.
- You are unsure whether a paraphrase is your own writing or a lightly edited copy.
- You want to cite something you have only seen quoted in another author's work.
- You need a page number and do not know whether the source type has one or whether it is required.
- You are reconciling a reference list against a finished draft before submission.
- You have a draft containing references you did not personally retrieve, from any source including an AI assistant, and need to establish whether they exist and say what they are claimed to say.
- Your feedback said "referencing inconsistent", "over-reliance on quotation", or "check your paraphrasing".

## 4. When NOT to use it

- **Your institution or department publishes its own referencing guide.** Then that guide governs, in full, wherever it differs from any general account of the style, including this one. Institutional guides routinely diverge from the published style manual on punctuation, on how many authors are listed before an abbreviation, and on electronic source formats, and markers apply the institutional version. Use this skill for the reasoning and the discipline, and the institutional guide for every detail of form.
- **The citation regime is specialised and statutory.** Legal citation, standards, patents, statutes, case law, archival manuscripts, ancient and religious texts, and some clinical and technical conventions have their own authorities and are not covered by a generic account of style families. Use the discipline's own authority, and use this skill only for the verification discipline, which still applies.
- **The problem is what to cite, not how.** Deciding which sources support a claim, how many are enough, and how a source is integrated into an argument is **15.05 Academic Writing Structure and Argumentation**. Citation mechanics cannot fix an argument that cites the wrong things.
- **The problem is that you cannot find sources.** That is **15.02 Academic Literature Search Strategy**. Referencing operates on sources you have; it does not obtain them.
- **The real issue is that the sources have not been read.** No citation method rescues a paragraph built from abstracts. If you cannot state what the source's method was and what it actually found, do not cite it for a substantive claim, and do not resolve this with a more careful reference format. Read it, or drop the claim (K4 §6.4).
- **You are auditing a large existing document's citations as a quality process rather than writing one.** For a systematic verification pass over a finished output, use **13.02 Source and Citation Verification**, which is built for auditing at scale.
- **The question is whether AI assistance itself must be declared.** That is governed by your institution's policy and its declaration requirements, not by a referencing convention, and it varies by institution and by assessment. See §12.1. Ask your institution; do not infer a rule.
- **No source can be opened.** If a reference cannot be retrieved in any form, it does not enter the reference list, is not offered as a lead, and is not written in citation format. There is no fallback here and no partial version. See §12.2.

## 5. Required inputs

**Required.**
- **The required referencing style, named**, or confirmation of which style family applies. Ask. Do not assume a style from the discipline, the country or the topic, because programmes within one department frequently differ. If it genuinely cannot be established before work must proceed, capture full bibliographic detail for every source (which is style-independent) and format later, and state the assumption.
- **The actual sources**, in a form that can be opened. A record, an abstract, a search result or a citation in another document is not the source.
- **The draft or the passage**, where the task is checking rather than setting up.

**Optional, and what each one adds.**
- **The institution's own referencing guide.** Overrides everything else on matters of form. Its absence is the single largest cause of technically correct references being marked down.
- **The marking criteria wording on referencing.** Distinguishes a criterion about accuracy from one about integration and range of sources, which are different problems with different fixes.
- **A reference management tool the student already uses.** Automates capture and formatting, and moves the work from producing references to checking them. It does not remove the checking, because tools import whatever metadata a publisher supplied, which is frequently wrong.
- **The word count and whether quotations count toward it.** Materially changes the quote-versus-paraphrase decision, and is an institutional rule that varies.
- **The list of sources already cited in the draft.** Enables the two-way reconciliation in Step 9 without reconstructing it.
- **Any feedback on referencing from a previous assignment.** The same student usually makes the same three errors, and naming them once fixes them faster than a general review.

## 6. Questions to ask before starting

1. **Which referencing style is required, and does your institution publish its own version of it?** Determines every formatting decision and which authority wins in a conflict. Default if unanswered: work in the style family the student names, capture full detail, and flag prominently that the institutional guide must be checked before submission.
2. **Which style family is it: author-date, numeric, or note-based?** Determines the shape of in-text citation, whether page numbers appear in the text or in a note, and how the end list is ordered. Default: ask which of the three the examples in the handbook look like, which a student can answer without knowing the style's name.
3. **Do quotations count toward your word limit, and is there a stated expectation about quotation density?** Changes the quote-versus-paraphrase decision throughout. Default: assume quotations count, and prefer paraphrase, which is the safer default academically in most disciplines regardless.
4. **What is your institution's policy on AI assistance for this assessment, and what must be declared?** Governs whether this skill may be used at all and what must appear in a declaration. Default: assume assistance must be declared and no generated text may be submitted (§12.1).
5. **Are any of your sources ones you have not personally opened?** Determines whether a verification pass is needed before anything else happens. Default: assume any reference whose provenance the student cannot state has not been verified, and verify it.
6. **Which source types are in your set: journal articles, books, chapters, theses, reports, standards, datasets, web pages, media?** Each has a different required pattern and a different locator problem, and the unusual ones are where errors concentrate. Default: handle journal articles and books first, and treat everything else as needing an explicit check against the guide.

## 7. Step-by-step methodology

**Step 1. Establish what citation is for, because the rest follows from it.**
A citation does three jobs, and only the third is compliance. It gives the reader a route to the evidence, so that any claim in your work can be checked at source. It locates your work in a conversation, showing what you are building on, responding to or disputing. And it distinguishes your contribution from other people's, which is what protects you. A student who understands the first job stops writing citations that gesture at a body of work and starts writing citations that would let a sceptical reader arrive at the exact page. It decides several later judgements, including when a locator is required and why a secondary citation is a weak move. *Correct result: the student can say, for any citation in their draft, what a reader would find if they followed it.*

**Step 2. Identify the style family, and let the family govern the mechanics.**
Three families cover almost all requirements, and recognising which one you are in tells you more than memorising a template.
- **Author-date.** The in-text citation carries author surname and year, with a page or other locator added for quotation and for specific claims. The end list is alphabetical by author. The reader identifies the source at the point of reading, which supports critical reading and is why most social science and science disciplines use it.
- **Numeric.** The in-text citation is a number, assigned in order of first appearance, with the end list ordered by that number. Compact and unobtrusive, favoured where citation density is high, and it costs the reader the ability to see whose work is being cited without looking away.
- **Note-based.** The in-text marker is a superscript number pointing to a footnote or endnote carrying the full reference, often with a bibliography as well. Used where sources are discussed as well as cited, and where commentary belongs alongside the reference, which is why humanities and legal writing favour it.
Establish the family, then get the exact form from the institution's guide. Where that guide and a general account differ, the institutional guide wins, without exception. Say this to the student explicitly, because they will find contradictory advice online and will otherwise follow whichever source looks most authoritative. *Correct result: the family named, the governing guide identified, and one worked example of each source type the project will use.*

**Step 3. Capture at the point of reading, never afterwards.**
When you open a source, record before you read: authors in full, year, exact title, container (journal, book, publisher, institution), volume, issue, pages, edition, and a persistent identifier or the location and date of access for online material. Then, as you read, record for every idea you might use: the claim in your own words, the exact page or section, and whether the note is quotation, paraphrase or summary. Mark quotations unmistakably at the moment you write them down, with quotation marks and a page number, because the single most common route to accidental plagiarism is a verbatim sentence copied into notes and later mistaken for the student's own paraphrase. Reconstructing references at the end from memory or browser history is the practice this step exists to eliminate; it is where wrong years, wrong journals and unverifiable claims enter a document. *Correct result: a growing record where every entry has complete bibliographic detail, a locator, and a note type.*

**Step 4. Verify every reference against the actual source. This is the gate.**
Nothing enters the reference list on plausibility, on a database record, on another author's citation of it, or on a formatted string produced by any tool including this one. For each reference, confirm three things from the document itself: that it exists and you have opened it; that the bibliographic details you hold match what the source says, not what a citing document said; and that the specific claim you attribute to it appears in it and means what you are using it to mean. The third check is the one most often skipped and it catches the most damaging error, which is a real source cited for a claim it does not make. Where a detail cannot be established, mark it incomplete and say what is missing. Never complete a partial reference by inference: a guessed year, volume, page range or identifier is worse than a visible gap. *Correct result: every reference traceable to a document you opened, with any incomplete field flagged rather than filled.*

**Step 5. Decide, for each use, whether to quote, paraphrase or summarise, and know what each risks.**
- **Quotation** reproduces exact words in quotation marks or as a set-off block, with a locator. Use it only when the wording itself is the point: a definition you will interrogate, a distinctive formulation, a claim you are about to dispute in its author's own terms, or primary data. Risk: over-quotation, which reads as an inability to process the material and, in most rubrics, scores as a weakness rather than as evidence of reading.
- **Paraphrase** restates a specific passage in your own words and structure, at roughly the same length, with a citation and usually a locator. Use it as the default. Risk: the too-close paraphrase, covered in Step 6.
- **Summary** compresses an argument or a whole study into much shorter form, with a citation. Use it for context and positioning. Risk: distortion, because compression drops the qualifications that made the original claim defensible, and a summarised finding routinely arrives stronger and more general than the author stated it.
All three require a citation. The belief that only quotation needs one is the commonest misconception about plagiarism, and it produces uncited paraphrase, which is plagiarism. *Correct result: each use tagged in the notes, with quotation reserved for cases where wording matters.*

**Step 6. Apply the paraphrase test, because word-swapping is not paraphrase.**
A paraphrase that keeps the original's sentence structure and substitutes synonyms is not your writing, and it is penalised whether or not you intended to copy. The reliable method is procedural rather than judgemental. Read the passage until you understand it. Close it, or turn it face down. Write what it said from your own understanding, in your own sentence shapes, without looking. Then reopen the original and compare. If any phrase of more than about three consecutive words matches, either quote it properly or rewrite it. If your sentence order mirrors the original's sentence order, restructure, because structure is copied as surely as wording is. Two further points matter. Technical terms and standard terminology do not need to be changed and should not be, since replacing them with near-synonyms introduces error. And a correctly cited paraphrase that is too close is still an offence in most institutional definitions, so the citation does not rescue it. This test also protects students writing in a second language, for whom staying close to the source's phrasing is a natural strategy and a structurally higher risk. *Correct result: a paraphrase you produced without the source in view, checked against it afterwards, with any surviving phrasing either quoted or rewritten.*

**Step 7. Attach locators, and know when they are required.**
A page number or equivalent locator is required for every direct quotation, without exception. It is required in practice for any specific claim: a particular finding, a figure, a definition, a table, a specific argument. It is not required when you are citing a whole work for its overall argument or as an example of a position. The governing test is the one from Step 1: could a reader find what you are pointing at without reading the whole source? For sources without stable pages, use whatever the source provides as a stable locator: a section heading, a paragraph number, a chapter, a timestamp for audio or video. Do not invent a page number for a source that has none, and do not carry over a page number from a different edition, which is a common error with books. *Correct result: locators on every quotation and every specific claim, in the form the source actually supports.*

**Step 8. Prefer the original over the secondary citation, and mark it honestly when you cannot.**
When you encounter a claim attributed to author A inside a work by author B, the correct action is to find and read A. Two reasons, and the second is the one students underestimate. First, B may have misread A, or compressed A in a way that changes the claim, which happens often enough that markers check it. Second, citing A through B tells the reader you have not read A, and a reference list built substantially on secondary citation is a visible signal about the depth of reading. Only where the original is genuinely unobtainable (out of print, untranslated, an unpublished source referenced in a published one) is a secondary citation acceptable, and then it is marked with the convention your style uses for "cited in", with only the source you actually read appearing in the reference list, or both, exactly as your guide specifies. Never present a secondary citation as though you had read the original. *Correct result: originals obtained wherever possible, and every remaining secondary citation explicitly marked.*

**Step 9. Reconcile in both directions before submission.**
Two passes over the finished draft, because they catch different failures. Forward: take every in-text citation in order and confirm a matching entry exists in the list, with the same author spelling and year. Backward: take every list entry and confirm it is cited at least once. Orphan entries are usually sources dropped during editing, and they signal a list assembled separately from the writing, which is what a marker suspects on seeing one. Then check the list's order (alphabetical for author-date, order of first appearance for numeric) and its consistency of form. Do this manually or with a tool, but do not assume a tool has done it: reference managers import publisher metadata that is frequently wrong, and they cannot detect a reference that is correctly formatted and factually fictitious. *Correct result: zero unmatched citations, zero orphan entries, consistent ordering, and a spot-check of at least a handful of entries against the source documents.*

**Step 10. Run the fabrication audit, especially on any AI-assisted draft.**
This is separate from Step 4 and it is run last, on the finished document, because references can enter a draft late. For every reference, ask: did I personally retrieve this document, and can I open it right now? Any reference that fails is removed, not investigated later. The characteristic signature of a fabricated reference is worth teaching explicitly: plausible authors who publish in the area, a real journal, a plausible year, a title that sounds exactly like something that would exist, and no such article. Fabricated references are not rare, they are not subtle, and they are trivially detectable by anyone who tries to open one, which markers increasingly do as routine. The consequence is not a lost mark for referencing; it is an academic misconduct case, and in professional and published contexts it has been career-ending. Where a reference is genuinely unobtainable through your library, treat it as unverified: it does not carry a claim. *Correct result: a signed statement to yourself that every entry was opened, and an empty list of unverified references.* `RESEARCHER SIGN-OFF REQUIRED` on the final reference list of any submitted work (K5 §2.8).

## 8. Analytical framework

**The citation chain**, which is the structure every reference has to support:

    Claim in your text → Citation → Reference list entry →
    Retrievable document → Specific location in it → The claim, as the source states it

Applying it: the chain is only as strong as its weakest link, and each link fails in a characteristic way. A claim with no citation is unattributed. A citation with no matching list entry is unresolvable. A list entry naming no retrievable document is fabrication. A retrievable document with no locator is uncheckable in practice, because a reader will not read forty pages to verify one sentence. And a locator pointing to text that does not support the claim is misattribution, which is the most serious of the five because it survives every formatting check and is invisible until someone reads the source. Test any reference by walking the chain forward, and you will find the break.

**The use ladder**, which decides how a source enters your sentence:

    Quotation (exact words, locator, wording is the point)
      → Paraphrase (your words and structure, same passage, locator)
        → Summary (compressed, whole argument or study, citation)
          → Synthesis (multiple sources, your claim, all cited)

Move down the ladder as your control of the material grows. Strong academic writing sits mostly at the third and fourth levels, uses the second where precision about a specific passage matters, and quotes rarely and deliberately. A draft dominated by the first level is a draft that has collected material rather than digested it, and that is what markers mean when they write "over-reliant on quotation".

## 9. Output format

**1. Style declaration.** The style family, the specific style, and the governing guide, with a note that the institutional guide overrides all other accounts of it.

**2. Source register.** One row per source, maintained from first reading.

| ID | Full bibliographic detail | Type | Retrieved and opened (date) | Verified against source | Locator convention | Notes on what it actually claims |
|---|---|---|---|---|---|---|

**3. Note record.** Per extracted idea: source ID, locator, note type (quotation, paraphrase, summary), and the note itself, with quotations unmistakably marked.

**4. Formatted reference list.** In the required style, ordered as the style requires, with any incomplete field visibly flagged rather than filled.

**5. Reconciliation report.** Unmatched in-text citations, orphan list entries, year and spelling mismatches, and ordering errors, each with its location in the draft.

**6. Paraphrase check log.** Passages tested, and the action taken on each (accepted, rewritten, converted to quotation).

**7. Secondary citation register.** Every claim cited through an intermediary, with a note on whether the original was sought and why it could not be obtained.

**8. Fabrication audit.** A per-reference confirmation of personal retrieval, and an explicit list of any reference that could not be opened, which do not carry claims.

**When a source cannot be verified, the format must not force fabrication (K4 §1).** The reference list shrinks; it is never padded, and a missing field is never inferred. An unverifiable reference is removed and the claim it supported is either re-sourced or restated as unsourced. A reference list with fourteen verified entries is stronger work than one with twenty-two of which six cannot be opened, and it should be said plainly that the shorter list is the better outcome.

## 10. Quality checks

Run before submission. These sit on top of K4 §8.

1. Has every reference in the list been personally opened, and could it be opened again right now?
2. Do the bibliographic details come from the source itself rather than from a document citing it?
3. Does every cited claim actually appear in the cited source, in the sense you are using it?
4. Does every in-text citation have a matching reference list entry, with the same author and year?
5. Does every reference list entry appear at least once in the text?
6. Does every direct quotation carry quotation marks or block formatting and a locator?
7. Has every paraphrase been produced without the source in view and then compared against it?
8. Does any paraphrase retain more than about three consecutive words, or the original's sentence order?
9. Are technical terms left as they are rather than replaced with near-synonyms?
10. Does every specific claim, figure or definition carry a locator?
11. Is every secondary citation marked as such, and was the original genuinely unobtainable?
12. Is any field in any reference inferred rather than read, particularly year, volume, pages or identifier?
13. Is the list ordered and punctuated consistently, according to the institution's own guide?
14. Have you spot-checked entries produced by a reference management tool against the actual sources?
15. Is the institution's AI declaration requirement satisfied for this assessment?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **Fabricated references** | Perfectly formatted, plausible, and no such document exists | Retrieval before entry, and the audit in Step 10. See §12.2 |
| **Real source, wrong claim** | The reference opens, and does not say what it is cited for | Verify the claim, not only the existence (Step 4) |
| **Too-close paraphrase** | Same sentence order, synonyms substituted | Write from understanding with the source closed, then compare (Step 6) |
| **Uncited paraphrase** | Paraphrased material with no citation because it was not quoted | All three use types require a citation (Step 5) |
| **Quotation drift in notes** | A verbatim sentence in notes, later used as if paraphrased | Mark quotations at the moment of capture (Step 3) |
| **Reconstructed references** | The list is built in the final two days from history and memory | Capture at the point of reading (Step 3) |
| **Missing locators** | Quotations with no page number; a definition attributed to a whole book | Locator rule in Step 7 |
| **Invented page numbers** | A page number for a source with no stable pagination | Use the locator the source actually provides |
| **Undeclared secondary citation** | The original is cited but only the intermediary was read | Find the original, or mark it as cited in (Step 8) |
| **Orphan list entries** | References nothing in the text cites | Backward reconciliation (Step 9) |
| **Style mixing** | Two conventions in one document, usually from copied text | Declare the style once and check the whole list against it |
| **Tool trust** | Formatting is immaculate and the metadata is wrong | Tools format, they do not verify. Spot-check against sources |
| **Over-quotation** | Long stretches of quoted text stitched with linking sentences | Move down the use ladder (§8) |
| **Summary inflation** | The source's hedged finding appears in your text as a general truth | Summarise the qualification along with the claim |

## 12. AI guardrails

Skill-specific. The universal prohibitions in K4 apply in full and are not repeated. This skill is uncompromising on the second item and there are no exceptions to it.

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: formatting a reference the student has retrieved, checking a paraphrase the student wrote, and reconciling a list against a draft are legitimate assistance. Producing paraphrased prose for insertion into the student's text is not, because the paraphrase is itself the intellectual work being assessed.

2. **Never generate a reference from memory, plausibility or pattern. There is no exception to this, and no softened form of it.** K4 §2.4 prohibits fabricating sources. The operational rule here: a reference exists only if the document has been retrieved and opened. Not if it is probably real. Not if the author certainly works in this area. Not if it is offered as a starting point, a lead, an example of the kind of thing to look for, or an illustration of the format. AI-fabricated citations are a well-documented failure with career-damaging consequences: they have produced academic misconduct findings, retractions, and professional sanctions, and they are detected by the simplest possible test, which is someone trying to open one. Every reference must be verified against the actual source, by the student, before it enters a submitted document.

3. **Never render an unverified source in citation format.** Formatting is what makes a fabrication dangerous. An unformatted description ("there is likely to be work on this in occupational health journals from the 1990s") reads as a hunch and is used as one. The same content set in the required style, with a year and a volume number, reads as evidence and is copied into a reference list. Where you believe relevant work exists, describe the kind of work and where to search for it, in prose, never as a reference.

4. **Never complete a partial reference.** Do not supply a missing year, volume, issue, page range, publisher, edition, place or persistent identifier by inference. Leave the field marked as missing and say what is missing and where to find it. An invented identifier is more harmful than a blank one, because it will be checked and will fail in a way that looks like carelessness rather than absence.

5. **Never confirm that a reference is real.** You cannot verify a document's existence, and stating that a reference "appears to be a genuine publication" gives a student false assurance about the one thing they must check themselves. The correct response to "is this reference real?" is to say that it must be verified by retrieving it, and to explain how.

6. **Never write the paraphrase.** Where a student asks for a passage to be put in their own words, the words produced are not their own. Instead: explain what the passage claims, ask them to write it from understanding with the source closed, and then test their version against the original using Step 6. This is slower and it is the point, because the paraphrase is where comprehension is demonstrated and assessed.

7. **Never assume the referencing style, the institution's variant of it, or the country's conventions.** Ask. Where you cannot ask, work in the family the student describes, state the assumption prominently, and direct them to the institutional guide, which overrides any general account including yours.

8. **Never advise that a citation is unnecessary because material is "common knowledge" without saying that the boundary is discipline-specific and institution-specific.** What counts as common knowledge in a field is not a general rule, and the safe default for an undergraduate is to cite.

9. **Where a claim rests on a source that could not be retrieved, the claim is restated as unsourced or removed** (K4 §2.4, K3 §3.4). It is never carried with a note promising verification later.

## 13. Best-practice principles

- **Reference as you read, not as you finish.** The cost of capture at the moment of reading is under a minute; the cost of reconstruction at the end is hours, and the reconstruction is where errors and fabrications enter.
- **A citation is a route, not a badge.** Write every citation so that a sceptical reader could arrive at the exact sentence you drew on. That single standard resolves most questions about locators, secondary citations and specificity.
- **The reference list is checked, and increasingly checked mechanically.** Assume every entry will be opened by someone. Write the list you would be comfortable having audited line by line, because that is now a routine event rather than a rare one.
- **Paraphrase with the source closed.** It is the only reliable protection against the too-close paraphrase, and it doubles as a comprehension test: material you cannot restate without looking is material you have not understood well enough to use.
- **Read the original.** A reference list built on secondary citation is visible, and beyond the risk of inheriting someone else's misreading, it tells the marker how far the reading actually went.
- **Prefer paraphrase and summary to quotation.** Quotation shows you found the passage. Paraphrase and synthesis show you understood it, and that is what carries marks.
- **Fewer verified sources beat more unverified ones.** A shorter, wholly checkable reference list is stronger work, and it is safer, than a longer one containing entries nobody has opened.
- **Tools format, humans verify.** Reference managers eliminate the tedious part and none of the risk. They import wrong metadata routinely and they cannot detect a reference that does not correspond to a real document.
- **The institutional guide beats every other authority.** Including the official style manual, including any general account, including this one. Markers apply the local variant, and disputes about who is technically right are not won by students.
- **When you are unsure whether to cite, cite.** The cost of an unnecessary citation is negligible. The cost of a missing one is an academic misconduct process.
- **A gap in a reference is a fact worth showing.** A visibly incomplete entry invites a correction. A silently completed one invites a finding.

## 14. Worked example

Generic fictional scenario, academic, health sciences.

**INPUT**

A final-year student on a health-related programme submits a draft chapter for supervisory comment. It contains thirty-one references in an author-date style. The supervisor's feedback says: referencing inconsistent, some paraphrasing too close, and two references could not be located.

**PROCESS**

*Step 2.* The style family is confirmed as author-date, and the institution turns out to publish its own guide, which differs from the general published version in two respects the student had not noticed: it requires all authors to be listed up to six rather than abbreviating after three, and it requires a specific format for organisational reports. Twelve of the thirty-one entries are wrong against the institutional guide while being correct against the general account the student had followed. The institutional guide governs, and this is recorded once so that the whole list is corrected consistently rather than one entry at a time.

*Step 4 and Step 10.* The verification pass is run on all thirty-one. Twenty-six open immediately. Three are behind access the library does not hold and are marked unverified. Two cannot be found at all. Investigating those two produces the diagnosis that matters: both had been produced by an AI assistant early in the project when the student asked for "sources on the topic", both are perfectly formatted, both name real and plausible researchers in the field, both name real journals, and neither article exists. The student had not noticed because both had never been opened. Both are removed. The claims they supported are re-sourced, one successfully and one not, and the unsupported claim is rewritten as a statement of the author's own reasoning rather than an evidenced claim.

*The judgement call.* One of the three inaccessible references is central to the chapter's argument, and the student wants to keep citing it on the strength of a detailed description in a review article that discusses it at length. This is the point where the temptation is strongest and the reasoning must be explicit. Two options are legitimate. The claim can be attributed to the review article, cited as such, which is honest and weaker. Or the original can be requested through interlibrary access, which takes two weeks the timeline can absorb. What is not legitimate is citing the original as though it had been read, which would be a misattribution invisible to every formatting check and detectable by anyone who compared the claim against the source. The student requests the original, and cites the review article in the interim with the "cited in" convention, replacing it when the original arrives.

*Step 6.* The paraphrase check is run on eight passages the supervisor flagged. Five are genuinely too close: same sentence order, technical terms retained (correctly), and connecting language substituted with synonyms. They are rewritten with the source closed. One turns out to be a verbatim sentence that had lost its quotation marks between the note record and the draft, which is the classic quotation-drift failure and the one with the most serious consequences; it is restored as a properly marked quotation with its page number. Two are acceptable and left alone, which matters, because the exercise is calibration rather than blanket rewriting.

*Step 7.* Locators are audited. Four direct quotations have no page numbers. Six specific findings are cited to whole articles, and are given page or section locators. Two references to a book use page numbers from a different edition than the one in the reference list, which is corrected by checking the edition actually consulted.

*Step 9.* Reconciliation finds three orphan entries, all sources cut during an earlier revision, and two in-text citations with a year that does not match the list, both traceable to a source with an online-first date and a later issue date.

**OUTPUT**

A corrected list of twenty-nine verified references, two removed fabrications with their claims re-sourced or rewritten, one secondary citation properly marked pending arrival of the original, five rewritten paraphrases and one restored quotation, complete locators, and a reconciliation showing no unmatched citations and no orphan entries. The list is shorter than the draft's and the work is stronger, which is the point to make explicitly to the student.

`RESEARCHER SIGN-OFF REQUIRED` on the final reference list, and the student, not the supervisor and not a tool, is the person who confirms that every entry was opened (K5 §2.8).

## 15. Advanced usage

**Setting up before the first source.** The highest-value use of this skill is at week one, not week twelve. Ten minutes establishing the style, the institutional guide and a capture format saves a week later and eliminates the entire class of end-of-project reconstruction errors. Where a reference manager is available, configure it then, and configure the note-type field, which is what prevents quotation drift.

**Auditing an inherited draft.** When taking over a document written by someone else, or returning to your own work after months, run Step 10 before reading anything else. A reference set of unknown provenance is unusable until it has been checked, and the check is fast: attempt to open every entry and sort into verified, inaccessible and non-existent.

**Working across two styles.** Students on joint programmes or writing for publication alongside a dissertation frequently need two. Keep the source register style-neutral, with complete detail captured once, and generate each style from it rather than maintaining two formatted lists that will silently diverge.

**High-volume verification.** Where a document carries a large reference set and a systematic audit is needed rather than an individual check, the task changes character and belongs to **13.02 Source and Citation Verification**, which is built for scale and for recording an audit trail across an entire output.

**Where the standard approach does not fit.** For archival, manuscript and unpublished material, the locator problem dominates and the convention comes from the archive or the discipline rather than from a general style. For datasets, software and preregistrations, cite the versioned object with its persistent identifier and the access date, since the content behind a plain link changes. For sources that no longer exist online, cite the archived capture and give the capture date, and do not silently cite a dead link.

## 16. Skill chain

**Recommended previous skills:**
- **15.02 Academic Literature Search Strategy.** Hands over retrieved sources, at which point capture begins. The two skills interlock: nothing enters the search's included set unretrieved, and nothing enters the reference list unopened.

**Recommended next skills:**
- **15.05 Academic Writing Structure and Argumentation.** Takes verified sources and integrates them into sentences that argue, which is where citation stops being mechanical and starts carrying the argument.
- **15.04 Research Proposal Writing.** Requires an accurate reference list at proposal stage, where fabricated references are frequently first caught.

**Runs well alongside:**
- **15.01 Academic Research Topic Selection**, from the first source opened during the prior-answer check.
- **13.02 Source and Citation Verification**, for auditing a finished reference set at scale.
- **13.03 AI Output Verification**, where a draft has been produced with substantial AI assistance and requires a systematic check before submission.

---
A Yazi Supplied Skill and resource.