A university student asks an AI assistant for sources to support a research essay. It returns three impressive-looking references. One cannot be found in a library search. Another has a real journal title but the article summary exaggerates the evidence. The student now has an important decision: copy the reference list, or follow the trail back to the original work.
Parents and students searching for NUS citation guidelines, using AI in university assignments, APA citations, academic integrity Singapore or AI-generated fake references need a simple distinction. A fluent answer is not necessarily a real source, and a genuine source does not automatically support the sentence next to its citation.
Start with the current instructions. NUS Libraries’ Citing AI Tools guide, updated 28 August 2026, points students to university guidance and reminds them to check their own assignment rules. NUS CELC’s plagiarism guidance discusses crediting borrowed ideas and data. Other universities and individual modules may set different AI-use and referencing requirements.
Find your next learning step
Know whether you have a source, a lead or an unsupported claim · Audit a suggested citation before placing it in your essay · Understand the difference between citation and AI disclosure · Build a source-to-claim trail with a real, checkable example · Complete a safe, honest final submission check
An academic source is something a reader can identify, locate and evaluate. An AI tool’s summary can be a lead towards a source, but it is not automatically a substitute for reading. Start with the research question and ask what evidence would genuinely help answer it. A question about study-space usage, for example, may call for observed patterns or suitable research, not a broad claim that all students behave alike.
Imagine an AI response that confidently says, “A 2023 study in Singapore proves every sheltered campus walkway is cooler and quieter.” This sentence is a fictional demonstration of the problem, not a genuine study or a citable finding. If the assistant does not provide a verifiable work, do not invent an author, journal or DOI to make the reference look complete.
Even when an article exists, check whether the assertion is narrower than the chatbot’s paraphrase. Did the authors measure temperature rather than noise? Was the sample drawn from one site rather than every campus? Did the study show an association instead of a causal relationship? Those details determine what a responsible writer may conclude.
An official university guidance page may support a statement about university academic rules, but it cannot establish a claim about the weather, learning outcomes or economic behaviour. Match a source to the role it actually plays. A policy, a commentary, a primary empirical study and a literary text can all be valuable, but they answer different questions.
The first research habit is separating what I want to know, what a source actually reports, and what I infer. If those three parts become indistinguishable, even a perfectly formatted reference list will be unreliable.
The related University Reading Groups guide recommends locating a statement in the assigned text before explaining or challenging it. Source-checked research writing develops the same habit.
A reference can contain author initials, year, journal, volume, page range and a DOI and still be false. Formatting is not verification. Search the full article title in a university library catalogue, recognised scholarly database, publisher site or legitimate repository. Compare the author list and title with the actual record.
If an identifier is given, check what it resolves to. A DOI that points to a different paper is not a harmless typo when it changes the identity of the cited work. A source that exists but has the wrong publication year or journal information still needs correction.
Read the abstract and relevant original sections. If the claim depends on a table, figure or result, inspect that part of the work. Record the relevant page, figure or passage in your private notes. Do not rely only on a search engine preview that may omit conditions or qualifications.
Ask whether the sample, place, time period and methodology support your exact use. A study that surveys forty volunteers does not represent the entire university population without further justification. A laboratory result does not automatically establish a classroom intervention’s effect. A paper that reports uncertainty should not be rewritten as certainty.
If the source cannot be located, do not include it. If it is real but unrelated, find a better source or revise the sentence. If it offers qualified evidence, preserve that qualification. The goal is not to collect the maximum number of citations; it is to use sources the student can defend.
A compact audit can use six fields: claim, source identity, located passage, what it shows, limitations, citation and disclosure requirement. This lets the student review each reference without needing to reconstruct the search process from memory.
A citation manager can make a list consistent, but an imported PDF can carry incorrect metadata. Confirm that the file title, paper title, authors and journal actually match. The user of the software remains responsible for the finished bibliography.
Three questions should be answered separately. Was the AI tool allowed for this assignment? Did the student use it in a way that must be acknowledged? Are the outside works that support the essay correctly cited? A student might get one of those questions right while failing the others.
NUS Libraries’ AI citation guidance explains that styles such as APA, MLA, Chicago and IEEE differ in their treatment of generated material. The page also instructs students to check the specific assignment requirements. It does not establish one rule that gives every university student permission to use a tool for every assessed task.
NUS CELC’s plagiarism resources describe the broader obligation to credit external ideas, language and data. Naming a real research paper in a reference list does not excuse copying the source without appropriate quotation or attribution. Nor does writing an AI-use disclosure convert a fabricated article into genuine evidence.
An assignment may prohibit AI-generated prose, permit certain brainstorming steps or require an explanation of the tool’s contribution. The correct instruction comes from that assignment and its institution, not from the most permissive website a student can find. Where the rules are unclear, ask the instructor before relying on an output.
NUS announced in August 2026 that first-year undergraduates from academic year 2026/27 would take a compulsory applied generative AI module. That university announcement reflects an emphasis on AI literacy, not blanket permission to disregard academic integrity rules in other subjects.
Do not confuse a university’s student policy with the AI policy of a journal, press or publisher. Scholarly publishing can impose additional or different disclosure and authorship rules. A student preparing work for publication should check the relevant venue as well as any university obligations.
Take this modest research-writing claim: “NUS Libraries advises students to check individual assignment instructions because the rules for using and citing AI vary.” Unlike the fictional research article above, the source for this statement is real and openly accessible: NUS Libraries, Citation Styles: Citing AI Tools.
The student can open the page, locate its guidance about AI citation styles, and verify the statement. They can identify the source as a university library guide rather than pretending it is a peer-reviewed research study. They can then prepare the reference according to the required style, using the correct institutional author or other bibliographic details in the actual page.
Notice what the guide does not say. It does not show that AI-generated sources are always authentic, that the student may use AI on every module, or that a particular citation style is compulsory throughout Singapore. The claim should stop where the evidence stops.
A useful paraphrase should express the source’s meaning without copying lengthy wording. An accurate quotation, where needed, should retain the original words and quotation marks. Both require appropriate attribution following the assignment’s rules. A paraphrase is not automatically original merely because its grammar differs.
Now suppose a student writes, “NUS has removed all plagiarism rules because students are learning about AI.” Attaching the genuine library-guide URL would not rescue that sentence. The original source does not support the conclusion, and the student should correct or remove it rather than searching for a convenient-looking citation.
For every major claim, maintain a note with the original link, specific section or passage, a summary of what was actually found and any unresolved uncertainty. A reader should be able to follow that trail. If the cited page is likely to change, record the access date where the relevant style requires it.
Compare this with eduKate’s History and Separating a Source from a Reconstruction. The subject changes, but the basic intellectual responsibility remains: distinguish a primary record from the interpretation built on it.
Before submission, inspect each factual assertion that relies on outside work. Can you locate the original source? Are author names and the date correct? Does the source support the exact claim? Does the reference appear in the bibliography in the requested style? If a quoted passage has a location requirement, have you supplied it?
Then check the assignment’s AI rules. Did you use a tool to suggest search terms, create an outline, polish text, interpret data or generate passages? Is that use permitted, and is a declaration needed? A student should describe their actual process, not invent or conceal steps to make the form look better.
The privacy check matters too. Do not upload participant data, confidential workplace records, classmates’ private submissions, unpublished research or restricted materials to an external AI service without the necessary authorisation. An assignment’s requirement to analyse information is not permission to put that information into an unapproved tool.
Practise a small weekly skill before attempting a long essay. Choose a real article or institutional page, write one claim that it supports, locate the passage and check its bibliographic details. In later weeks, compare two sources and explain why their results differ. This routine teaches evidence rather than the mere appearance of scholarly writing.
If a student’s reference list repeatedly contains wrong titles, review imported metadata and the source-discovery process. If the sources are genuine but the essay does not answer its question, revisit the argument. If the writing is clear but unsupported, strengthen the reading and research rather than adding citations at random.
A university tutor can be helpful when the student needs to distinguish an unclear research question from a source-evaluation problem or a writing problem. Good feedback identifies the obstacle and gives the student a task they can perform independently. No tutor should substitute invented citations for actual research.
For broader study habits, eduKate’s Secondary 1 Mathematics Tutor Clementi guide illustrates why a learner should be able to explain and check a method rather than display only a final answer. At university level, that principle extends to evidence, writing and honest authorship.
Quick answers: Can AI invent a journal reference? Yes. Does a functioning DOI prove the summary is accurate? No. Does declaring AI use make fabricated evidence acceptable? No. Must every institution use the same citation style? No. Is a university library guide a research paper? No, it is a source for institutional guidance. Is a shorter list of well-checked references worthwhile? Yes, when those references actually support the work.
Research integrity becomes a positive skill when students can point to the origin of a claim, describe what it means, admit what remains uncertain and explain which parts of the analysis are genuinely their own.
Official sources and connected eduKate reading
- NUS Libraries — Citing AI Tools — citation styles, source checking and assignment-specific guidance
- NUS CELC — Plagiarism Resources — attribution of ideas, quotations and data
- NUS Libraries — AI Guidelines — institutional resources for responsible AI learning
- NUS News — Applied Generative AI Learning 2026/27 — context for AI literacy without claiming general permission
- University Reading Groups — eduKate — grounded discussion and textual evidence
- History: Source and Reconstruction — eduKate — transferable source-checking discipline
- Is NUS a Good University? — eduKate — institutional context and further reading
Sources checked 17 October 2026. Independent eduKate educational guidance, not universal permission to use AI. The example of a purported research study is expressly fictional. Follow the responsible institution’s current course, assignment, research and privacy requirements.