1. There is no single “correct” AI workflow for every subject. A history student reading archival material, a business student analyzing data, and an engineering researcher writing code will need different tools and different safeguards.

2. The most useful approach is to begin with the task rather than the tool: decide what you are trying to learn or produce, identify the parts that require your own judgment, and then test whether AI improves the process.

3. Below is an AI-assisted writing process from Monash University. The diagram separates the writer's responsibilities from tasks for which AI may provide limited support at different stages.

• Before asking AI for help, write down the assignment or research question, intended audience, required evidence, deadline, and applicable policy. This takes only a few minutes and makes it much easier to recognize when the AI has strayed from your original topic. If assignment instructions are complex or in a second language, you can use AI to break down the requirements or clarify terminology. However, avoid letting it dictate your entire outline without checking against the syllabus.

• AI can help turn a long assignment into a schedule or suggest ways to divide work among a group. It can also play the role of a tutor by asking questions rather than providing the answer immediately. Study modes, custom GPTs, and source-based tools such as Gemini Notebook can be useful here.

• The plan still needs a human check. AI may underestimate the time needed to obtain data, complete ethics review, learn a method, or read difficult material.

• AI is often good at taking an early, vague idea and showing several ways it might be narrowed. You can ask how the topic changes when you specify a population, place, period, method, or theoretical perspective. You can also ask which assumptions are hidden in the way the question is phrased.

• AI cannot reliably determine whether a genuine research gap exists. Its coverage may be incomplete, and a polished statement about “limited prior research” is not evidence. A gap has to emerge from a careful literature review.

• Research topics often emerge from a distinctive life experience, a social observation, a connection between fields, a new method for an old problem, or an unusual dataset. AI may help turn an observation into possible questions, but it cannot certify that the result is original, feasible, or unexplored.

• A more useful conversation might be:

I am interested in university libraries and AI literacy. Show me three ways to narrow the topic by population, institutional setting, and type of learning outcome. (Or: Show me how approaches might differ between Taiwan and [Home Country]). For each option, identify the kinds of evidence I would need. Do not claim that a research gap exists.

• When self-doubt makes it difficult to begin writing, AI can also provide encouragement. That is a legitimate use, as long as the reassurance does not replace feedback from a supervisor or the evidence needed to make a research decision.

• The more knowledge you bring to the conversation, the more specific your prompt and evaluation can become. When the answer sounds flattering—“This is highly original and valuable”—pause and check the literature.

• AI can suggest keywords, translations, related concepts, and possible databases. However, for Taiwan-centric research, AI might generate translated terms that differ from local academic usage. These suggestions are a starting point. Check them against local subject headings in NCCU Primo, review articles.

• Use several channels:

  • Primo and library catalogs for books, e-books, theses, and library holdings;

  • Subject databases for precise disciplinary searching;

  • Scopus and Web of Science for citation relationships;

  • Google Scholar for broad discovery and open versions;

  • Official agencies for laws, policies, and statistics;

  • Archives and physical collections for material that is not online;

  • Academic AI tools for additional exploration.

• Every AI-suggested publication should be verified. Confirm the title, author, year, journal or publisher, identifier, abstract, and full text. 

• AI-assisted screening tools such as ASReview or Rayyan may reduce repetitive work, but they do not decide the review protocol or the inclusion criteria for you.

• Keep a search record. A simple table with the database, date, exact query, filters, result count, and screening decisions is more useful than relying on a chat history that may later disappear.

• AI can explain terminology, but specialized concepts should be checked in academic reference works or field literature. 

• Most academic AI tools generate answers based solely on abstracts. Even a source-based system can omit details. If you depend on the summary, you may miss the exact research design, a qualification in the discussion, or evidence hidden in a table.

• Before close reading, scan the title, abstract, headings, figures, tables, method, and conclusion. Then decide where you need to slow down. AI can be a useful warm-up. Give it the abstract and ask:

  • What should I check when I read the method and results?

  • What alternative explanations might the authors need to address?

  • What background knowledge would help me understand this paper?

  • Which claims should I verify in the full text?

• Once you are reading, treat the article as a conversation. Ask what problem the author is trying to solve, what evidence supports the argument, and where the author's interpretation differs from your own.

• Targeted questions work better than “summarize this paper”:

  • Which section of this article addresses my question about [topic]?

  • How do these papers define [concept] differently?

  • What similarities, disagreements, or limitations do these studies identify?

  • Here is my understanding of the papers. Challenge my interpretation and identify missing perspectives.

  • Does my paraphrase accurately preserve the author's meaning?

  • Based on my previous reading notes, which ideas do I understand well, and what should I read next?

• Keep quotations, paraphrases, and your own interpretation in separate notes. This makes it less likely that a generated paraphrase will be mistaken for your own analysis.

• Do not upload subscription articles to another platform (e.g., for full-text translation or summarization) unless the article is open access or the license permits it. Doing so violates publisher agreements and data privacy. Do not use automated tools to bulk-download NCCU database content.

• AI can classify readings or create structured comparison tables, but its categories may not match the logic of your research question.

• It can also explain citation formats, yet automatically generated references must be checked against the style guide and the source record.

• AI can explain a method, suggest diagnostic checks, or draft code. It should not choose a design simply because that design is easy for the tool to generate.

• For research design, ask about alignment between the question and method, possible alternative explanations, feasibility, validity, and ethical risks. Discuss consequential choices with a supervisor, methods specialist, or statistician.

• Survey and interview questions generated by AI still need to be checked for leading language, double-barreled wording, cultural assumptions, accessibility, and missing constructs. Pilot testing remains essential.

• When writing code, start with synthetic or non-sensitive data. Specify the software and version, read every transformation, test edge cases, and compare at least part of the output with a manual calculation.

• For qualitative analysis, define the research question and codebook yourself. AI may help test or organize a coding scheme, but it should not be credited as an independent coder. Record what data were supplied, which prompts were used, and how human decisions changed the result.

• Never upload identifiable participant data, student records, confidential manuscripts, or restricted datasets to a public AI service without clear authorization.

• AI is often most useful after you already have an argument or a substantial set of notes. It can identify weak transitions, suggest a clearer order, generate alternative titles, or point out places where a reader may need more explanation.

• A responsible workflow is straightforward:

  • Write the argument or detailed notes.

  • Tell the AI who the reader is and what the text needs to do.

  • Ask for feedback.

  • Decide which suggestions to accept.

  • Revise in your own voice.

  • Check every factual claim and citation.

• For translation, pay particular attention to technical terms, names, numbers, negation, and degree of certainty. Browser translation extensions can make bilingual reading faster because the original and translation remain visible together, but specialist terminology still needs to be checked.

• Authors remain responsible for the final work. Expert review is particularly important for specialist translation.

• A reviewer should not upload another person's confidential manuscript to an AI system. An author may use a permitted service to obtain preliminary feedback on their own manuscript, but should protect sensitive sections and revise the document as a whole after working on separate passages.

The comments below are translated from selected responses to the NCCU Library activity “Dual Reading Personalities: Brain Mode vs. Heart Mode”. Minor editing has been applied for readability while preserving the speakers' meaning. 

Contextual vocabulary and academic expression

• The two main difficulties for me are reading comprehension and written expression. Academic articles contain unfamiliar vocabulary and sentence structures, and a dictionary definition often does not explain how the word works in context. I sometimes have to reread the same paragraph or the whole article. When writing, I may know what I want to say but cannot express it naturally and precisely.

• I use AI to explain difficult words within the passage and summarize the main point. For writing, I first put down my ideas in a raw form and add notes about the tone I want—for example, more formal or more persuasive—then ask AI to revise it. I want help developing my own academic-reading strategies and academic language, not just translation or rewriting.

Multilingual and cross-cultural research

• As a student of Asia-Pacific studies, the hardest part is understanding diverse historical, political, economic, and social contexts and the cultural differences among countries. I need support in analyzing and comparing multilingual sources. I would use AI as an early research tool—for example, to summarize major perspectives in five recent articles on China's influence in the South China Sea—then use those summaries as a basis for deeper research. I would also like help identifying bias and considering the issue from several cultural perspectives.

Large amounts of literature

• • The difficult part is handling a large volume of information while keeping the argument rigorous. Reading requires quickly identifying each paper's contribution and relating it to the research question. Writing requires turning complex thinking into a clear outline and expressing an original argument without plagiarism. I would use AI as a literature assistant for initial summaries, comparison, outline logic, and language polishing.

Building an argument from several papers

• • I find it difficult to integrate the viewpoints of several sources into an argument of my own. After reading, I often have a vague idea but do not know how to organize it. I would like help comparing the methods, theoretical frameworks, and conclusions of the papers and identifying useful follow-up questions. I would also use AI to map concepts from my notes, check whether my paraphrase reflects the original, and practice writing an abstract or research motivation in English.

Finding the core method and maintaining structure

• • The hardest part is quickly identifying the main point and method, especially when the article contains unfamiliar terminology. In writing, I struggle with organization, logic, citation, and formal English. I would use AI to summarize papers, revise sentences, check grammar and spelling, and suggest academic wording, while still reviewing the final text.

Simulating a reader

• • I would use AI to help identify key points, improve academic English, check citations, draft an outline, translate between Chinese and English, and simulate a reader who tells me whether the argument is clear.

Evaluating source quality

• • The hardest part is finding high-quality sources that are genuinely relevant. Some material only looks convincing on the surface. I would like AI to compare definitions across scholars, help develop research questions and a conceptual framework, and organize the core concepts of several papers.

Understanding argument rather than vocabulary alone

• • Academic reading is difficult when a paper contains many technical terms and background citations. I may understand the words without understanding the argument. Writing is difficult because I need to use sources while maintaining my own position. I use AI to organize literature, clarify sentence structure, check grammar, and try different styles, but I revise the result so that it still expresses my view.

Academic wording and professional communication

• • Academic wording is hard to manage. If I write too conversationally, the text loses professionalism; if I use technical terms awkwardly, it becomes difficult to read. I have used ChatGPT to revise emails to instructors so that the wording is smoother and appropriate for the recipient.

Reliable sources and limited time

• • Finding reliable information is the hardest part. Because time is limited, I often only scan abstracts. I need help comparing and organizing data and managing sources accurately. I would like AI to highlight key concepts and point out gaps in my thinking.

Making a new concept approachable

• • I need AI to help build a bridge from point A to point B when I encounter a new concept. The hardest part of academic reading and writing is sometimes simply not daring to begin. Support that makes entry into a topic faster and friendlier is important.

Paraphrasing and plagiarism

• • I find it difficult to rewrite another person's idea in my own words without creating a risk of plagiarism. I want to learn literature-review and organization skills and use AI to comment on my revised paragraphs.

Translation uncertainty

• I most often use AI for Chinese–English and English–Chinese translation, but I do not know whether the result is correct.

• AI can summarize quickly, but I still read the original. AI-translated technical terms are sometimes so obscure that the passage becomes difficult to comprehend.

A conversation partner

• • When there is no suitable person available to discuss the topic, AI gives me an immediate opportunity to interact and organize my research question.

Manga

• • I enjoy the drawings and exaggerated expressions that are rare in real life but common in manga. AI cannot replace the emotional ups and downs. A work can make us feel that we are in another world, lifting us when it is joyful and pulling us down when it is sad or angry. AI may process a book, but it cannot experience those emotions as a person does.

George Orwell's 1984

• • The setting of Big Brother and thought control makes the loss of freedom feel oppressive and makes me think about the relationship between information and power. I often stop and reflect on the meaning beneath the text. AI can explain the content, but it cannot replace the shock, resonance, and questions about humanity and the future that arise while I read.

The Little Prince

• • Every rereading gives me a different experience. As a child I saw the story; later I understood loneliness and love; as an adult I recognize responsibility and choice. AI can summarize or explain, but reading allows us to build the scene slowly, and foster a deeply personal connection. That immersion and conversation with the author cannot be simulated.

In Order to Live: A North Korean Girl's Journey to Freedom

• • This account by a North Korean defector was the first time I learned about North Korea through the description of someone who had actually lived there rather than only through the internet. AI cannot reproduce the emotion that comes from reading sentence by sentence. If we frequently use AI instead of reading, we may lose both the pleasure and the ability to read.

Legend of Fei (有匪)

• • The novel combines fantasy and martial arts, and I identify with the courage and resilience of the characters. Its friendship and love are moving. The part AI cannot replace is the emotional connection: each reader understands a character through personal experience.

Tsubaki Stationery Store series ツバキ文具店

• • The letters and conversations convey a warmth that makes readers laugh and cry with the characters. An AI summary may omit details, distort the work, and lead readers astray. The emotional experience depends on the text itself.