Integrity: citation and AI
In this lesson
By the end, you’ll be able to
- Verify that a citation exists and actually supports the claim attached to it
- Tell quotation, genuine paraphrase, and patchwriting apart, and attribute each correctly
- Disclose AI use to a verifiable standard and never trust an unverified AI-generated citation
Sign in to save your progress
You can keep reading without an account, but completed lessons won’t be saved.
Everything in a paper rests on trust: the reader trusts that your sources are real, that they say what you claim, and that the words are either yours or clearly marked as someone else's. Integrity is not a separate virtue bolted onto research. It is what makes the citations, and therefore the argument, mean anything. This lesson is about three ways that trust breaks, and how to keep it: verifying citations, attributing wording honestly, and being straight about where you used AI.
A citation has to pass two gates before it earns its place. First, the source has to exist. Second, it has to actually support the specific claim you attach to it. Both fail more often than you would think: studies that check reference lists find a meaningful share of citations that do not support, or even contradict, the statement they are cited for. Formatting a reference correctly proves nothing, and neither does the fact that a paper is real and famous. The only real check is to open the source and read it against your sentence.
Exercise
What does it actually mean to verify a citation before you use it?
Two things must be true: it is real, and it says this.
The second kind of trust is about words. A quotation uses the source's exact words in quotation marks with a citation. A genuine paraphrase restates the idea in your own structure and wording, and still cites it. Between them sits the trap: patchwriting, taking the original sentence and swapping a few words for synonyms while keeping its shape, then presenting it as your own. Patchwriting is not paraphrase, and it is not honest, even when a citation is attached, because the wording is still the source's. Copying without any citation is plain plagiarism. The test for a paraphrase is simple: could you have written it that way without the original open in front of you?
Categorize
Sort each writing practice into whether it is acceptable or crosses into misconduct or patchwriting.
Quote or genuinely reword and cite: acceptable. Near-copy, uncited copy, or unchecked AI text: not.
AI tools change the stakes of both problems at once. They write fluent prose, which makes patchwriting and undisclosed copying easier, and they fabricate citations: realistic-looking references, with plausible authors, journals, and page numbers, for papers that do not exist. Studies auditing model output find fabricated and inaccurate references at rates far too high to trust anything unchecked. So two rules hold. Never let an AI-generated citation into your work without opening the source yourself, and disclose your AI use to a verifiable standard, naming what the tool did and stating that a person reviewed and verified the output. AI tools cannot be authors, because they cannot be accountable for the work.
- Walters WH, Wilder EI. Fabrication and errors in the bibliographic citations generated by ChatGPT. Sci Rep. 2023;13:14045.
- Flanagin A, Bibbins-Domingo K, Berkwits M, Christiansen SL. Nonhuman 'authors' and implications for the integrity of scientific publication and medical knowledge. JAMA. 2023;329(8):637-639.
Signature interactive
Citation Verifier
Three claims, each with a citation attached. You have already done the checking; the result is shown. Deliver the verdict for each: does the source support the claim, is it real but off-point, or is it not a real source at all?
Claim. Spacing study sessions over time improves long-term retention compared with massing them.
Cepeda NJ, Pashler H, Vul E, Wixted JT, Rohrer D. Distributed practice in verbal recall tasks. Psychol Bull. 2006;132(3):354-380.
When checked: The paper is found, and its abstract reviews exactly this: distributed practice reliably beats massed practice for retention.
Claim. Students who used our tutoring app scored higher because the app improved their motivation.
Ryan RM, Deci EL. Self-determination theory and the facilitation of intrinsic motivation. Am Psychol. 2000;55(1):68-78.
When checked: The paper is real and widely cited, but it is a theory of motivation in general. It never studied this app, and says nothing about it raising test scores.
Claim. A 2021 meta-analysis found that gamified quizzes raise retention by 30% in secondary students.
Nguyen TP, Alvarez M. Gamification and retention: a meta-analysis. Journal of Learning Sciences. 2021;29(4):511-538.
When checked: No database returns this article. The DOI does not resolve, and that journal has no volume 29, issue 4 matching it. The AI draft that produced it likely invented it.
Constructed example. The point is the reasoning, not these specific references.
Written response
Lab. You used an AI tool to help draft a literature review. It produced a fluent paragraph that includes the citation 'Nguyen et al. (2019), Journal of Learning Sciences,' but when you search for it, no such paper exists. Write two things: a short methods or acknowledgement note that discloses your AI use to a verifiable standard, and a sentence explaining what you must do about the citation before submitting.
Disclose: what the tool did + that you checked it. Citation: it does not exist, so remove it and re-ground the claim in a real source.
0/60 words
Checkpoint · item 1 of 5
Why is it risky to paste an AI tool's citations into a paper without checking them?
What do models sometimes invent that looks completely real?
Reflection