Is referencing in university assignments only about format?

Direct answer: Formatting slips are mistakes inside the entry, such as a misspelled name or a wrong page. Content errors are claims the cited source does not support. Baethge and Jergas (2025) pooled 32,000 quotations from 46 studies and found 16.9% incorrect, with no improvement over four decades.

A common assumption is that referencing in university assignments is a layout skill: get the commas, italics and hanging indents right and the job is finished. The research on accuracy points elsewhere. Baethge and Jergas (2025) separate two problems. Citation inaccuracy covers technical slips such as misspelled names or wrong pages, while quotation inaccuracy "signifies errors of content" (Introduction), meaning the source does not back the sentence that cites it.

Across 32,000 quotations, the pooled rate of incorrect ones was 16.9% (95% CI 14.1% to 20.0%), and 8.0% were major errors. A meta-regression slope of -0.002 (p = 0.85) showed no improvement over time, which the authors summarise as "one in every 6 quotations is incorrect, half of them in a major way and without improvement during the last 40 years" (Baethge & Jergas, 2025, Conclusion).

The figures come from published medical papers, not student essays. A student who copies an entry from a search result without opening the source is in the same position: the entry looks right and the claim is unchecked.

Source: definitions from Baethge and Jergas (2025); examples and last column are a MAAS illustration, read 2026-10-27.

Direct answer: Knowing a rule and applying it are different skills. In a survey of 60 international students, 86.7% rejected the idea that an unfindable source needs no citation, yet that survey tested attitudes, not paraphrasing. When 66 students rated paraphrases in a second study, only 35% rejected a patchwritten one.

Sanni-Anibire et al. (2021) surveyed 60 international students, 97% of them studying in Manitoba, so the sample is Canadian rather than Australian or British. Of 60 respondents, 52 (86.7%) disagreed that a source the instructor is unlikely to find can go uncited, and 51 of 54 (94.4%) agreed that proper citation acknowledges others' contributions. Confidence was also high: 39 of 49 (79.6%) felt sure they knew how to avoid misconduct. Even so, 14 of 48 (29.1%) felt fearful on reading the integrity policy, and 55.0% had received instruction on citation management software.

The authors are open about the limit: "We did not test the skills related to paraphrasing, citing, referencing, or other ways to avoid plagiarism" (Sanni-Anibire et al., 2021, Discussion). The survey measures knowledge and feelings.

Schwabl et al. (2014) tested skills directly with 9 instructors and 66 second-year students at a western Canadian university, 61 of them Canadian-born and 5 international, so the paraphrase result mostly reflects domestic students. Only 33% of students answered all six paraphrase definitions correctly, against 38% of instructors. In referencing in university assignments, the gap between a sound attitude and a sound paraphrase is the part a reference list cannot show, and the next section picks it up.

Direct answer: Patchwriting keeps the source's sentence frame and swaps some words, while a paraphrase rebuilds the idea in new structure and still names the source. Schwabl et al. (2014) found 35% of students judged a patchwritten paraphrase not acceptable, 23% acceptable and 42% somewhat acceptable.

Schwabl et al. (2014) summarise the result bluntly: "Only 38% and 35% of instructors and students, respectively, correctly recognized patchwriting as not acceptable" (Discussion). Asked to rate one patchwritten paraphrase, 23% of students called it acceptable and 42% somewhat acceptable. The same students were better at the opposite task, since 73% picked out the acceptable paraphrase, and their confidence ran ahead of their accuracy: 28% felt very confident, 61% somewhat and 11% not very confident.

The worked example below uses a real sentence from Baethge and Jergas (2025, Introduction) as the source.

• Source sentence: "Citation inaccuracy usually denotes technical errors, e.g., misspelled author names or wrong page references."
• Patchwritten version: Citation inaccuracy generally means technical errors such as misspelled author names or wrong page references.
• Paraphrase: Baethge and Jergas (2025) reserve the label citation error for slips inside the entry, and use quotation error for a sentence that its source does not support.

The patchwritten line keeps every clause in the original order, so a citation after it would still sit on borrowed structure. The paraphrase changes the grammar, adds the author's distinction and carries the name in the sentence. Asked what help they wanted, 61% of students chose an online course with exercises, and so did all of the instructors.

Direct answer: Citation habits shift with the language of the essay. In a study of 25 Hong Kong first-year students, 23.14% of English citations put the author in the sentence, against 6.78% in Chinese, and paraphrase rose from 28.27% to 40.57%. The language pair differs from Vietnamese and English.

Mu et al. (2025) compared one Chinese and one English essay from each of 25 first-year students, 778 citations across 110,212 words, and interviewed 12 of them. They report that "Citation density is 6.41 in the L1 Chinese essays and 8.06 in the L2 English essays" (Mu et al., 2025, p. 782). The share of author-prominent citations was 23.14% in English and 6.78% in Chinese, and paraphrase made up 28.27% of Chinese citations against 40.57% of English ones. Direct quotation fell from 27.57% of citations in Chinese to 15.43% in English. The English essays also contained 23 generalisations (6.57%), against none in Chinese.

The authors add that students used more reporting verbs in English but struggled with their meanings. For referencing in university assignments written in English, the practical point for a Vietnamese student is to check the verb, since "claims", "suggests" and "demonstrates" make different promises about how strong the evidence is. The study covers first-year undergraduates in Hong Kong, so the percentages should not be read as a Vietnamese baseline; no Vietnamese study appears in the sources behind this guide.

A second-language writer has a particular reason to compare drafts with sources line by line. The verb you choose is part of the claim you are attributing to the author.

Direct answer: No. Bhattacharyya et al. (2023) checked 115 ChatGPT references and found 47% fabricated, 46% real but inaccurate and 7% accurate. Gravel et al. (2023) found 69% of 59 references fabricated. Open every AI-suggested reference in a library database before it reaches your list.

Bhattacharyya et al. (2023) had ChatGPT-3.5 write 30 short medical papers, with a mean of 3.8 references each, and checked all 115 references against Medline, Google Scholar and DOAJ. The wrong detail most often was the PubMed identifier (93% of papers), followed by volume and page numbers (64% each) and year (60%). Fabrication varied by prompt, from 66% on healthcare disparity to 36% on prevention and 34% on recent advances (p = 0.007).

Of the references checked by Bhattacharyya et al. (2023), 47% were fabricated and 46% were real but inaccurate. Gravel et al. (2023) put 20 medical questions to ChatGPT and evaluated 59 references. Of these, 41 (69%) were fabricated, and 29 of those 41 (71%) were attributed to a known journal, website or preprint repository in correct citation format. The authors found that "Most fabricated citations used names of authors with previous relevant publications, a title that seemed pertinent and a credible journal format" (Gravel et al., 2023, Abstract). A reference that looks right is therefore weak evidence that it exists.

Both studies used 2023 versions of ChatGPT on medical topics, so the exact rates will differ for other tools and subjects. The failure type is what carries over: plausible details that nobody checked.

Direct answer: Open the source, find the passage behind each sentence, and confirm the DOI, volume and pages in a library database. Picazo-Sanchez and Ortiz-Martin (2026) still found eight fabricated references, identified by manual verification, among the 127,063 extracted from 3,541 papers published in 2023, so even peer review does not remove the need to check.

Picazo-Sanchez and Ortiz-Martin (2026) generated 74,196 BibTeX references from nine language models, then audited 127,063 references in 3,541 published papers. Their manual check identified 8 fabricated instances, and they call the mere existence of fabricated content in the peer-reviewed literature "a critical integrity issue" (Abstract). Universities now expect honesty about tool use as well: the University of Queensland Library (n.d.) states, "If you use AI in your assessment, you must appropriately acknowledge your use" (section Acknowledging and referencing AI use).

A five-step check fits into the last hour before submission:

1. Search each DOI or title in the library database and confirm the volume, pages and year match your entry.
2. Open the source and find the sentence or table that supports your claim.
3. Compare your wording with the source side by side for copied sentence frames, the patchwriting pattern shown above.
4. Match every in-text citation to one list entry and every list entry to one citation.
5. Note any AI tool you used, in the form your course asks for.

Step 2 is the one that catches content errors in referencing in university assignments, and it is the one a formatting tool cannot do for you.

Direct answer: The style is set by your course profile or lecturer, not by personal preference. The University of Queensland Library (n.d.) lists APA 7th, Chicago 18th, MLA 9th, IEEE, AGLC, Vancouver, ACS and CSE, and tells students to check the course profile. The University of Leeds publishes its own Harvard variant.

The format of referencing in university assignments changes from course to course, and the accuracy checks above do not. The University of Queensland Library (n.d.) defines a referencing style as "a set of rules on how to acknowledge the thoughts, ideas and works of others", and adds that "Referencing is a crucial part of successful academic writing, avoiding plagiarism and maintaining academic integrity in your assignments and research" (Referencing guide). QUT's cite|write site lists APA, Harvard, AGLC and Vancouver and has a format page for generative AI. The University of Leeds Library publishes its own Leeds Harvard guide, alongside Leeds Numeric and MHRA, plus a referencing help service.

For a side-by-side comparison of the two most common styles, read our guide to APA 7th vs Harvard referencing. The APA 7th referencing guide covers the entry formats, and how to write an annotated bibliography shows how to keep source notes that make step 2 quick. Start the notes when you first read a source, because adding references at the end is where the wrong sentence gets the wrong source.

Once you have run the five checks yourself, a second reader can catch what is left. Bring your own draft to the MAAS assignment audit and review service, where an expert reads referencing in university assignments the way a marker does. The comments flag a quotation that outruns its source, a paraphrase that keeps the original sentence frame, an entry whose volume or pages do not match the database, and an in-text citation with no entry in the list or the reverse. Each comment explains the rule behind it, so you learn the pattern while you fix it.

The service does not promise a grade and does not write any part of your work. You wrote the draft, and you make every correction yourself, which is also the safest position if your course asks you to explain your sources in a viva or an integrity meeting.

• APA 7th vs Harvard referencing: how the two most common styles differ in the text and the list
• How to write an annotated bibliography: source notes that make verification faster

• Baethge, C., & Jergas, H. (2025). Systematic review and meta-analysis of quotation inaccuracy in medicine. Research Integrity and Peer Review, 10(1), Article 13. https://doi.org/10.1186/s41073-025-00173-z
• Bhattacharyya, M., Miller, V. M., Bhattacharyya, D., & Miller, L. E. (2023). High rates of fabricated and inaccurate references in ChatGPT-generated medical content. Cureus, 15(5), Article e39238. https://doi.org/10.7759/cureus.39238
• Gravel, J., D'Amours-Gravel, M., & Osmanlliu, E. (2023). Learning to fake it: Limited responses and fabricated references provided by ChatGPT for medical questions. Mayo Clinic Proceedings: Digital Health, 1(3), 226-234. https://doi.org/10.1016/j.mcpdig.2023.05.004
• Mu, R., Zhu, X., Zhang, X., & Cheong, C. M. (2025). Source text use in first-year university students' academic writing: A within-subject analysis of L1 Chinese and L2 English citation practice. Open Journal of Modern Linguistics, 15(5), 773-793. https://doi.org/10.4236/ojml.2025.155045
• Picazo-Sanchez, P., & Ortiz-Martin, L. (2026). Evaluating the integrity of LLM-generated citations: Prevalence and risks of fabricated references in scientific literature. Data, 11(5), Article 122. https://doi.org/10.3390/data11050122
• Sanni-Anibire, H., Stoesz, B. M., Gervais, L., & Vogt, L. (2021). International students' knowledge and emotions related to academic integrity at Canadian postsecondary institutions. International Journal for Educational Integrity, 17(1), Article 21. https://doi.org/10.1007/s40979-021-00088-4
• Schwabl, K., Rossiter, M. J., & Abbott, M. L. (2014). University students' and instructors' paraphrasing and citation knowledge and practices. Alberta Journal of Educational Research, 59(3), 401-419. https://doi.org/10.55016/ojs/ajer.v59i3.55730
• University of Queensland Library. (n.d.). Referencing guide. Retrieved October 27, 2026, from https://guides.library.uq.edu.au/referencing

This article is part of the MAAS Journal series for students. MAAS is an academic support partner: experts guide students through each stage of their own work, and we do not complete assessed work on a student's behalf.