How to mark reference lists faster, in five minutes a paper

The reference list is marked last, at the bottom of the script, when the mark is already half decided. That is where the time goes on the wrong thing: a marker corrects italics and full stops, and never finds out whether the third source exists.

To mark reference lists faster, reverse the order. Start with the check that can move a mark by a class and end with the one that separates neighbouring bands. Five minutes a paper is the budget, and a lookup in EdCitation's free Verify references can do the first two of them.

What should I check first in a reference list?

Check first whether the sources exist and whether any has been retracted, because that is the only check that catches an invented reference, and an invented reference is the one referencing fault that moves a mark by a class rather than a band.

The reason is the fabrication rate in work written with a language model. Walters and Wilder (2023) went through the 636 references in 84 short literature reviews ChatGPT had written: 55% of GPT-3.5's and 18% of GPT-4's did not exist. Linardon et al. (2025) ran GPT-4o on six mental health reviews and found 35 of 176 references fabricated, 19.9%, and 46% in a specialised review of digital interventions for binge eating disorder.

Retraction is the same check from the other side: the paper exists, and the record says it should no longer be relied on. The publisher's page carries the notice, and what a retraction is and how to check for one shows what to look for.

This check is not a hunt for typos. Across 105 studies of reference accuracy in peer-reviewed journals, Logan et al. (2024) found that 32.7% of references carried at least one error, most often in an author's name. A paper that cannot be found where it should be is a different finding, and your notes should keep the two apart.

Which references to look up by hand when you cannot look up all of them

Sample five and weight the sample: the two sources the argument leans on most; any book chapter, since GPT-4 fabricated 70% of the chapters it cited in the Walters and Wilder (2023) test, against 18% of its references overall; anything without a DOI; and sources on a narrow topic. How to check whether a reference is real gives the hand method.

How do I check that citations and the reference list match?

Read the list against the text in both directions: every in-text citation needs an entry and every entry needs a citation. In a 2,000-word essay this takes about a minute.

1. Search the text for "(19" and "(20", which finds nearly every author-date citation, and tick each against the list.
2. Mark any entry you did not tick: padding, a leftover from a deleted paragraph, or a reading the student meant to use.
3. Check that surname and year agree exactly. "Smith (2019)" in the text and "Smith, J. (2020)" in the list is a citation the reader cannot follow.
4. In a numbered style, check that the numbers run in order of first mention.

This is also the check a tool does in the same pass as the lookup, which is why it sits second and not fourth. Upload the whole paper to Verify references, rather than pasting the list alone, and it ticks every citation in the text against the list, and every entry in the list against the text.

Do the sources fit the brief?

Compare the list with what the assignment asked for: how many sources, how recent, what kind, and whether the set readings appear. This is a judgement: a minute of thought, not of searching. It is quicker with the brief's requirements in front of you as a list; EdCitation's Check your paper turns the brief into a checklist (word limits, required sections, the style, how many sources), and that is the list to hold each script against.

De Montfort University's generic descriptors expect work at 60 to 69% to draw on "an appropriate range of properly referenced sources" (De Montfort University, n.d.), and the University of Essex's Department of Government asks third-year essays for "evidence of independent literature searches" (University of Essex, n.d.). Neither gives a number, and no style guide does; how many references an essay needs sets out what university guides say.

The signs take seconds: no set reading; every source older than the module's own readings; web pages where the brief asked for peer-reviewed work; a list far longer than the essay used.

One check here no tool performs: open the source behind the central claim and set its abstract beside the sentence that cites it. A genuine paper made to say what it never said is the failure a reference list cannot show you.

When should I check the formatting, and how much is it worth?

Last, and for the least time, because in published marking criteria formatting separates neighbouring bands, not classes.

The University of Edinburgh's School of Social and Political Science describes work at 60 to 69% as showing "a generally high standard of referencing" with "some errors", 50 to 59% as having "minor errors and omissions", and 40 to 49% as having "significant errors and omissions" (University of Edinburgh, n.d.). Warwick's History department allows "some minor errors in formatting" at 62 to 68 and "significant errors in formatting" at 52 to 58, and keeps "little or no attempt to follow a citation style" for a fail (University of Warwick, n.d.). Formatting alone moves a mark inside a class. Referencing that is missing moves it out of one.

The guidance does not all agree on where the threshold sits: De Montfort lists "adherence to the referencing conventions" in its 40 to 49% band, a condition of passing rather than a distinction between passes. Where your department's criteria differ from those quoted here, your department's win.

Sixty seconds covers what matters: one style throughout, titles in italics, and a DOI or link wherever one exists.

What rubric lines should I use for referencing?

Use lines that say what a strong list shows and what a weak one shows, in the order of the routine.

Put the weight on the first three rows: an essay that fails the first cannot pass the others. And say in the criteria that references are looked up: a student who knows the list will be checked makes a different list.

How do I mark reference lists faster for a cohort of eighty?

Checking student references for a whole cohort means the same routine on every paper, the lookups done before marking starts, and a hand-checked sample to calibrate on. Five minutes a paper is six hours forty for eighty papers, a real share of the marking workload, so the first two minutes should not be done by hand.

1. Look every list up first. Paste each list, or better, upload each paper, into EdCitation's Verify references and keep the results with the scripts. Before reading a word, you know which papers carry a source that could not be found or a paper that was retracted.
2. Sort the pile. Papers where every reference verified take three minutes: match citations, judge the fit, glance at the format. An entry marked not found or "check this" gets the hand check, with a note of where you searched and when; the tool's verdict tells you where to spend the minute, and your own search is what you act on. An entry that could not be checked gets a look at what kind of source it is (a report, a chapter, a thesis) before any conclusion.
3. Hand-check five papers in full. Choose them at random before you start; if the sample and the lookup disagree, find out why before relying on either.
4. Same routine for every student. A check run on the papers that looked suspicious and not on the rest is hard to defend at an appeal.
5. Keep a log. The reference, where it was searched, what came back and the date.

Why the same routine for every paper matters

Marking is less consistent than markers believe, and the reference list is the one criterion where that is cheap to fix. Bloxham et al. (2016) asked 24 experienced UK assessors in four disciplines what they noticed when separating strong work from weak, found variation in which criteria they chose, ranked and scored, and concluded that at this level of judgement "variability is inevitable". Hasan and Jones (2024) had nine examiners at a UK dental school re-mark essays they had marked nine weeks earlier: only one landed in the same grade category, and the size of the change was unrelated to experience.

Whether a source exists is not a judgement, and neither is whether a citation has an entry. Those checks give every marker the same answer, provided every marker runs them.

What should I write in feedback about referencing?

Write one specific sentence per problem, in the order of the routine, with the fix and where to learn it. A bare "ref?" in the margin is a question, not feedback.

Derham et al. (2022) analysed 2,101 in-text comments on 60 marked essays and found most aimed at the task in front of the student rather than at how the student worked, with lower-graded work mostly receiving corrections. A referencing comment can do better, because the fix is a method.

On the first row, never write the accusation. Ask for the source, follow your institution's procedure from the first step, and keep whether the source exists apart from how it got into the essay; how to spot fabricated references in student work sets out a fair process.

Two things could not be confirmed: no peer-reviewed measurement was found of how much marking time goes to reference lists, or of what share of feedback comments concern referencing. The five-minute budget is a working rule, not a measurement.

Which tool does the first two minutes for me?

EdCitation's Verify references is the best tool for the first two minutes of this routine, since it reads each entry against the publisher's record instead of composing anything, which no chatbot can claim. The reason a lookup beats a glance is in the Linardon study: 21 of the 33 DOIs on its fabricated references were real DOIs pointing at unrelated papers, which a tired eye passes and a record does not. Upload the paper and both minutes are covered at once: you get a verdict per entry (verified, "check this" for a doubtful match, or not found), a flag on any retracted paper, and the citations and the list paired up both ways. An entry that could not be checked is never shown as not found, so a report or a thesis the indexes do not hold is never passed off as missing. Even a true "not found" is where the hand check starts, not a finding about the student. It is free and needs no account.

The routine itself costs nothing. Pro, at $8 a month, is for the writer's side, with References from a file among its tools; Max, at $24, brings Theoretics QA and the Library as well.

For a department, the Institution licence covers every student of a programme, integrates with the LMS and sign-in, and comes on one invoice. Students can then run the same check before they submit, and when the lookups are done before marking starts, the marker's five minutes begin at the third, with the hand check kept for the entries the tool marks.

Quick questions

Should I check every reference in every paper?

By hand, no: sample five per paper, chosen the same way for every student, and check the whole list when one fails. With a tool that looks a whole list up at once, yes: every student gets the same check.

Is a wrong page number evidence that a reference is invented?

No. A scoping review of 105 studies found that 32.7% of references in published journal articles carry at least one error. A wrong detail on a paper that exists is a referencing error; a paper that cannot be found where it should be is a different finding.

How long should marking a reference list take?

About five minutes a paper: existence and retraction, citations matched to the list, fit with the brief, then formatting. Uploading each paper to EdCitation's free Verify references before marking starts does the first two minutes.

What do I write when I cannot find a student's source?

Say where you searched and when, and ask for the PDF or a link, without calling the reference fake. A missing record is strong evidence for a recent journal article and weak for a report, a chapter or a thesis.

References

• Bloxham, S., den Outer, B., Hudson, J., & Price, M. (2016). Let's stop the pretence of consistent marking: Exploring the multiple limitations of assessment criteria. Assessment & Evaluation in Higher Education, 41(3), 466–481. https://doi.org/10.1080/02602938.2015.1024607
• De Montfort University. (n.d.). DMU generic undergraduate mark descriptors. https://www.dmu.ac.uk/documents/about-dmu-documents/quality-management-and-policy/academic-quality/learning-teaching-assessment/ug-mark-descriptors.pdf
• Derham, C., Balloo, K., & Winstone, N. (2022). The focus, function and framing of feedback information: Linguistic and content analysis of in-text feedback comments. Assessment & Evaluation in Higher Education, 47(6), 896–909. https://doi.org/10.1080/02602938.2021.1969335
• Hasan, A., & Jones, B. (2024). Assessing the assessors: Investigating the process of marking essays. Frontiers in Oral Health, 5, Article 1272692. https://doi.org/10.3389/froh.2024.1272692
• Linardon, J., Jarman, H. K., McClure, Z., Anderson, C., Liu, C., & Messer, M. (2025). Influence of topic familiarity and prompt specificity on citation fabrication in mental health research using large language models: Experimental study. JMIR Mental Health, 12, Article e80371. https://doi.org/10.2196/80371
• Logan, S. W., Hussong-Christian, U., Case, L., & Noregaard, S. (2024). Reference accuracy of primary studies published in peer-reviewed scholarly journals: A scoping review. Journal of Librarianship and Information Science, 56(4), 896–949. https://doi.org/10.1177/09610006231177715
• University of Edinburgh, School of Social and Political Science. (n.d.). Marking criteria. https://www.sps.ed.ac.uk/students/undergraduate/your-studies/assessment-regulations/marking-descriptors
• University of Essex, Department of Government. (n.d.). Marking criteria: Undergraduate courses. https://www1.essex.ac.uk/government/documents/current/Marking_Criteria_UG.pdf
• University of Warwick, Department of History. (n.d.). Assessments and marking. https://warwick.ac.uk/fac/arts/history/students/undergraduate/assessments-marking
• Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, Article 14045. https://doi.org/10.1038/s41598-023-41032-5