There is no single rule that applies to every use of generative AI. The appropriate boundary depends on the purpose of the assignment, the research method, the discipline, the institution, and the target journal.

Start with the most specific rule that applies to your work:

• the instructor's syllabus or assignment instructions;

• NCCU, department, or research ethics requirements;

• the target journal or publisher's current policy;

• the rules of the relevant professional or scholarly community.

The person submitting the work remains responsible for its accuracy, originality, confidentiality, and proper attribution. AI cannot be an author because it cannot approve the final version, accept responsibility for errors, disclose conflicts of interest, or respond to questions about the research.

NCCU's basic principles

For coursework, reports, and theses, follow the National Chengchi University Generative Artificial Intelligence Application Basic Principles.

• Planning and structure: AI may help create an initial structure for an assignment, report, or thesis. The user must review and revise that structure for appropriateness and accuracy.

• Rewriting a draft: AI may help modify or expand an initial draft. The writer must rewrite and adjust the content, ensure that the tone and phrasing are appropriate, and take responsibility for the quality of the report.

• Risks and intellectual property: Users should understand the strengths, limitations, and risks of generative AI, including intellectual property concerns. Generated content may contain serious errors, bias, or material derived from copyrighted works. Users must critically review, verify, correct, and properly attribute what they use.

• Attribution: When generated material is used in an assignment or report, its sources and the role of the tool should be documented in accordance with academic integrity requirements.

• Privacy, bias, and discrimination: Do not enter personal or confidential information. Avoid biased or discriminatory wording and content.

Ask the instructor early

The syllabus may permit AI for one part of an assignment and prohibit it for another. Ask at the beginning of the course rather than after submission: May I use generative AI for [specific task], and what records or disclosure do you require?

Examples from other universities

• The University of Hong Kong: A journal article coauthored by HKU-affiliated scholars was reported to contain references to nonexistent articles, books, Legislative Council documents, and institutional reports. The case contributed to the development of clearer university principles for AI in research.

• National Pingtung University: Fabricated AI-generated references contributed to the withdrawal of a thesis and revocation of a degree. 

• University of Washington: The university's guidance on the effective and responsible use of AI in research provides a broader framework for graduate researchers.

• University of Toronto: Institutional guidance for graduate academic milestones emphasizes three points:

  • a doctoral dissertation must present original research and make an original contribution to knowledge;

  • all use of generative AI in the research process must be carefully cited and described, including what was used, how, and why;

  • when a supervisory committee permits AI use, the student's contribution and the tool's contribution must be clearly distinguished, and the student must be able to provide evidence that the doctoral standards have been met.

Academic publishers are responding to the risk that fabricated studies, irrelevant references, manipulated images, or unchecked AI-generated content may enter the scholarly record. Springer Nature has developed internal tools to help identify irrelevant references, signs of AI-generated content, and problematic images, which are then subject to human review.

Most policies share two basic principles:

• AI cannot be listed as an author.

• Human authors are fully responsible for the work.

American Historical Association

The AHA argues that generative AI makes traditional historical skills more important, not less. Students must learn that AI can fabricate sources and that every AI-generated claim should be treated as potentially false until verified.

Examples of potentially acceptable use

• identifying key points for later checking;

• brainstorming;

• generating an initial bibliography that is independently verified;

• generating a historical image with clear disclosure;

• correcting citation format;

• revising grammar or syntax;

• identifying additional perspectives that are fact-checked and rewritten in the student's own words;

• summarizing material in preparation for critical reading.

Examples of unacceptable use

• submitting an AI-generated paper as one's own work;

• directly using an unverified AI summary;

• criticizing another scholar based on an AI summary;

• using AI-generated references.

APA journals

Authors should:

• provide enough information for readers to understand how generative AI affected the research and manuscript;

• disclose the use in the most relevant section, often the Introduction or Method;

• avoid entering confidential research data or client records into AI systems;

• verify AI-generated information by reading the original sources;

• acknowledge AI assistance when a comparable contribution by a human would normally be credited.

For example, a literature review already explains which databases and search terms were used; if AI materially affected the search or synthesis, that role should also be described.

Wiley

Wiley distinguishes among AI use in:

• drafting and editing;

• research design and data collection;

• literature review;

• code development;

• data analysis;

• figures and visual material.

The disclosure location depends on what the AI did:

• Acknowledgments: substantial drafting, editing, translation, or formatting;

• Methods: research design, data collection, analysis, code, or literature-review procedures;

• Figure caption or note: generation or editing of images and visualizations.

Example:

The authors acknowledge the use of Microsoft Copilot (accessed August 2024) to generate a first draft of the Acknowledgments section and translate the Methods section from Spanish to English. Gemini (accessed July 2025) was used to reorganize the Discussion section for clarity and flow. All AI-assisted text was reviewed and revised by the authors to ensure accuracy and clarity of meaning.

Step 1: Decide whether the use is substantial enough to disclose

Disclosure is generally appropriate when AI:

• generates a substantial amount of text;

• reorganizes an argument;

• contributes to the method;

• performs data analysis or coding;

• affects a literature review;

• influences research design;

• creates or materially edits figures, illustrations, or visualizations.

Minor spelling, formatting, or wording suggestions may not require disclosure under some policies. The specific rules of the course, institution, publisher, or journal take precedence.

Step 2: Document the necessary information

Include:

• the name of the AI tool and model or version;

• the date or period of use;

• the role it played, such as translation, analysis, code development, or literature searching;

• the material supplied to the tool;

• the author's role in directing, reviewing, and verifying the output;

• privacy, sensitive-data, copyright, and licensing safeguards;

• major prompts or settings when they affect reproducibility.

Step 3: Place the disclosure appropriately

• Methods: Use this section when AI affected the method, data, code, literature-review procedure, or analysis.

• Acknowledgments: Use this section when AI provided substantial drafting, editing, translation, or restructuring, subject to the publisher's rule.

• Figure caption or note: Use this location when AI generated or edited an illustration, chart, or visualization.

• Do not use AI to alter a photograph presented as factual evidence or a documentary record.

APA, MLA, and Chicago provide ways to cite AI-generated text and images. But an AI response often draws on training data, search results, or a source supplied by the user. In that situation, the AI is not the original source.

The principle is the same as following a citation in a journal article: read and cite the original work rather than citing the intermediary.

Keep the prompt and the response. If you share the conversation, check that it does not contain personal or confidential information.

MLA 

MLA is widely used in language and humanities disciplines. MLA does not treat AI as an author. It recommends describing the generated content and identifying the tool used.

A Works Cited entry may include

• the prompt or a descriptive title;

• the AI tool;

• the model or version;

• the developer;

• the date;

• a stable conversation link when available.

Generated text

• “Describe the theme of nature in Jane Austen's Mansfield Park” prompt. ChatGPT, model GPT-4o, OpenAI, 23 Sept. 2024, [shared conversation URL].

• If the generated work is a poem or other creative work, the writer may begin the entry with a title supplied for the work or with its first line.

• If a conversation has a stable public URL, use it. If it cannot be accessed without signing in, use the general tool URL and preserve the interaction separately.

Generated image

• Place the credit in the figure caption:

• Fig. 1. “Create an expressionist-style image of two people standing on a beach looking at the ocean” prompt, DALL-E, version 3, OpenAI, 23 Sept. 2024, [shared conversation URL].

Five practical points

• • A prompt or descriptive title can serve as the first element when citing generated content.

  • The entry should identify the tool, model or version, developer, date, and retrievable link.

  • An AI-generated image should be credited in its figure caption.

  • A generated creative work may be identified by a supplied title or its opening words.

  • When an AI answer points to another source, follow the link, verify the material, and cite the original source rather than the AI summary.

Explore further 

• • • https://style.mla.org/citing-generative-ai-updated-revised/, https://style.mla.org/citing-generative-ai/, https://style.mla.org/describing-ai-use/ 

APA 

APA is widely used in the social sciences. APA does not treat AI as an author. Key guidelines include:

• authors must confirm what use is permitted;

• the user remains responsible for checking accuracy;

• sensitive information, bias, privacy, and copyright require attention;

• prompts and records may belong in the Methods section, an author note, an appendix, or supplementary material;

• sources suggested by an AI must be opened and read before they are used.

Describe AI use in the method

• For example: We used Claude 3.7 Sonnet (Anthropic, 2025) to generate an initial list of grammar concepts that students should learn by the end of Grade 12. The list was reviewed and revised by the authors against the curriculum and relevant language-learning literature.

References for shareable AI conversations

• Anthropic. (2025, May 20). Essential grammar topics for high school graduates [Generative AI chat]. Claude Sonnet 4. [Share URL]

• Google. (2025, May 22). High school grammar concepts overview [Generative AI chat]. Gemini 2.5 Flash. [Share URL]

• OpenAI. (2025, August 21). High school grammar concepts [Generative AI chat]. ChatGPT. [Share URL]

• In-text citations: (Anthropic, 2025; Google, 2025; OpenAI, 2025) or  Anthropic (2025), Google (2025), and OpenAI (2025)

Explore further 

• https://apastyle.apa.org/blog/cite-generative-ai-references, https://apastyle.apa.org/blog/cite-generative-ai-search-software, https://apastyle.apa.org/blog/cite-generative-ai-allowed 

Chicago

Chicago style is widely used in history and the humanities.

When AI-generated text is reproduced, identify the tool and its role in the text or in a note. A publicly accessible conversation may be included in the bibliography; a nonpublic interaction is treated more like personal communication.

• Note example: Text generated by ChatGPT, OpenAI, March 7, 2023, https://chat.openai.com/chat.

• If the prompt is not described in the main text: ChatGPT, response to “Explain how to make pizza dough from common household ingredients,” OpenAI, March 7, 2023.

• In an author-date system, information not stated in the sentence may appear parenthetically: (ChatGPT, March 7, 2023)

• Generated image: “A modern office rendered as a cubist painting,” image generated by OpenAI's DALL-E 2, March 5, 2023.

Explore further 

• https://www.chicagomanualofstyle.org/qanda/data/faq/topics/Documentation/faq0422.html,  https://www.chicagomanualofstyle.org/qanda/data/faq/topics/Documentation/faq0423.html 

Differences among disciplines

Computing and engineering — IEEE

• AI-generated text, tables, figures, images, or code used in an article should be disclosed in the Acknowledgments, identifying the system, the affected section, and the extent of use.

• Routine editing and grammar correction may not require disclosure.

Medicine and life sciences — ICMJE

• AI-assisted writing should be described in the Acknowledgments.

• AI used for data collection, analysis, or figures should be described in the Methods.

• AI cannot be an author.

• Human authors must ensure that every citation is properly supported.

Medicine — JAMA

• Requirements include:

  • • the tool, version, manufacturer, date, purpose, and affected sections when AI is used to create, review, revise, or edit a manuscript;

    • details of research design, application, prompts, modifications, datasets, training, and ethics review when AI is used in the method;

    • source and licensing information for generated visual material;

    • discussion of bias, inaccuracy, sensitivity, and inadequate representation, together with steps taken to reduce these risks.

Law — Harvard Journal of Law & Technology

• Authors may use generative AI at different stages, but must disclose all tools and functions in a title-page footnote, including proofreading, drafting, figure generation, citation formatting, research, and idea generation.

• Authors remain responsible for nonexistent or missing sources, weak engagement with the literature, poor style, and plagiarism.

Accounting and business — The Accounting Review

• AI must remain under human control and supervision.

• Authors are responsible for errors, omissions, bias, and copyright issues.

• An AI disclosure statement appears after the abstract and before the main text.

• Use in data generation, collection, processing, or analysis must be explained in the Methods, including the tool, procedure, date, and possibly prompts.

• Spelling and grammar correction, reference management, and ordinary web searching do not require disclosure.

• AI-generated or AI-modified images, interactive graphics, and video are prohibited, as are attempts to use AI to bypass review.