The CitationLab Blog

Field notes and discoveries

Case files, guides and engineering notes on citation checking and reference integrity — everything we learn testing against a corpus of real theses, written up so you can use it.

All posts Case Files Guides AI & Integrity Inside CitationLab For Institutions Updating References All topics
THE THESIS [12] [13] [14] REFERENCE LIST 0 ENTRIES PARSED [12] READ AS VANCOUVER — MATCHED
Case Files · Featured

The checker said 90% of my citations failed. The checker was wrong.

A doctoral thesis, a couple of hundred references, and a report claiming nine in ten citations were missing. The true number of problems was two — and the difference was one broken assumption about citation style.

CitationLab Team · July 2026 · 8 min read
AI & Integrity

It missed 27 citations, then told me it had found the last one

A general-purpose assistant missed 27 citations in one thesis chapter and said each time that it had found the last. That failure is why CitationLab runs rules before models.

Read the post →
AI & Integrity

Ref[In] — naming the thing a citation check is actually for

Every tool describes an action: run a check. REFerence [IN]tegrity names the property a check is for — three claims about your document, and a ledger that has to balance.

Read the post →
Case Files

The APA default that broke every Vancouver thesis

Style was detected, then thrown away — why referencing style must be established first, not assumed.

Read the post →
Case Files

From 1 thesis to 5,000: how we test

The corpus story — and how running every change against real documents has caught our own mistakes.

Read the post →
Guides

How to find citations missing from your bibliography

The step-by-step check for cited-but-not-listed sources, before your examiner runs it for you.

Read the post →
AI & Integrity

ChatGPT gave you citations that don't exist. Now what?

How AI invents references that look flawless, the five tells that expose them, and the honest fixes once you find one in your thesis.

Read the post →
Inside CitationLab

Nothing gets lost: the conservation ledger

Every line the checker reads is either on your screen or in a named place you can open. Why a citation check should reconcile like a ledger.

Read the post →
For Institutions

What a Ref[In] Report tells a supervisor

One page before chapter one: how the Ref[In] Score is read, what the two tables mean, and why the report measures quality — never misconduct.

Read the post →
Case Files

The bare-year bug

'(Farah & Lindgren, 2019) … and in their follow-up (2020).' Whose year is that? How a checker should read a citation that arrives with no name.

Read the post →
Case Files

'Nelson SD, W.' — anatomy of a mangled author

Same characters, two grammars — only one has people in it. How author parsing goes wrong and why refusing with a reason beats guessing.

Read the post →
Case Files

One author, two spellings, zero matches

Fourteen 'missing' citations and one orphan entry — all the same person. Where spelling variants come from, and which ones a checker may fold silently.

Read the post →
Case Files

One orphan, one missing: a healthy report

What 'good' actually looks like on a citation report — and why chasing zero teaches worse habits than reading your two rows honestly.

Read the post →
Case Files

When a university becomes an author

The citation names two people; the entry names a building. Where affiliation-as-author comes from, and the evidence that puts the real authors back.

Read the post →
Updating References

How to update outdated references before submission

The currency pass every thesis needs: find superseded editions, published preprints, moved DOIs and retracted papers — and update them without breaking your citations.

Read the post →
Guides

How old is too old? Reference currency by discipline

A 1962 paper can be the strongest citation in your thesis or the weakest, and the year alone cannot tell you which. How to judge reference currency properly.

Read the post →
Guides

The 5-year rule myth

Your supervisor says nothing older than five years. They are not wrong — they are compressing three different concerns into one number. Here is how to unpack it.

Read the post →
AI & Integrity

Can AI invent fake references? Ours can't — by construction

The operation that produces fabricated references — generate a plausible source — is not implemented. What the model is actually asked, where references come from instead, and the one guard we removed on purpose.

Read the post →
For Institutions

Citation hygiene at scale: quality, not misconduct

A referencing defect that appears at a rate across a cohort describes the task, not the people doing it. What it costs an institution to read it the other way, and what to do with a report instead.

Read the post →
AI & Integrity

Deterministic first, AI second — and why the order is the whole design

Why CitationLab settles what it can with rules before any model is consulted, and what that ordering buys you in a reference check you can actually verify.

Read the post →
Guides

Did someone help with your thesis? Two checks before you submit

Supervisors, proofreaders, writing centres, AI tools — help with a thesis is normal and usually sanctioned. The accountability is not shared. Two checks to run on your references before you sign.

Read the post →
Guides

Et al. done right

How many authors et al. may stand in for, when it may first appear, and whether the reference list may use it at all are decided by your style. What the rules are, and why the abbreviation is the most information-destroying thing you can do to a citation.

Read the post →
Inside CitationLab

We read your extracts, not your thesis

What actually reaches an AI provider during a citation check: the citation string, the parsed authors and year, and a capped window of surrounding words. Every citation goes; no chapter does.

Read the post →
Guides

Match 10 — what the first ten pairs tell you, and what they don't

The first ten matched pairs are free. A sample of ten answers one question about a reference list well, hints at a second, and structurally cannot answer the third — which is the one most people assume it covers.

Read the post →
Guides

Orphan references: entries you never cite

A reference listed but never cited. Where orphans come from, why the heading on your list decides whether they are errors at all, and why a large share of them are not orphans but matching failures.

Read the post →
For Institutions

The Ref[In] Index — measuring REFerence [IN]tegrity across universities

Every thesis is screened for plagiarism; almost none is checked for whether its references resolve. The Ref[In] Index measures it, one score per institution, published annually from August 2027.

Read the post →
Guides

Audit your own paper's references before you submit it

A paper's reference list fails differently from a thesis: co-author merges, reference-manager exports, and a review cycle long enough for the literature to move. The five-step audit, in the order that pays.

Read the post →
Guides

Secondary citations ("as cited in")

A secondary citation names two works and puts only one in your reference list — which looks identical to the commonest defect in a bibliography. What the form claims, when it is the right answer, and how to remove one properly.

Read the post →
Updating References

How we find the newer version: seven engines, one verdict

Crossref, PubMed, Semantic Scholar, OpenAlex and the scraped tiers, asked in a fixed order. Why free-first is a metadata-quality rule, and what the engine ledger reveals.

Read the post →
Inside CitationLab

The seven steps of a Ref[In] annotation

Annotation runs as seven stages in a fixed order, on a frozen session, by one worker — and it will not begin until the ledger balances. What each stage produces, and why every constraint was added after something went wrong without it.

Read the post →
For Institutions

REFerence [IN]tegrity for universities — the check that stopped running

Every thesis is screened for plagiarism. Nobody checks whether its references resolve. Why the check stopped, what AI did to the problem, and which units inside a university already own the work.

Read the post →
Inside CitationLab

We check the references you already had, not just the ones we found

A citation check that only reports what is missing leaves the risky part untouched. Every reference in your bibliography is independently resolved — including the DOI you supplied, which is followed to see what it actually describes.

Read the post →
Guides

How to tell what citation style your thesis uses

Your thesis probably does not have one style. Identify the shape of each citation instead — six forms, decidable from the surface — and diagnose your own document in ten minutes.

Read the post →
AI & Integrity

Why every correction needs your click

Why CitationLab never applies an AI-formed correction on its own: the click is what turns a machine suggestion into a decision you can account for.

Read the post →