Research Integrity and AI Ethics
This final session of Module 4 covers authorship boundaries, COPE guidelines, CMOS 18 AI disclosure statements, plagiarism prevention, and human verification protocols for cited evidence.
By the end of this session, you will be able to:
- Explain the ethical definition of authorship according to COPE guidelines, and why generative AI cannot be listed as an author.
- Draft clear, transparent AI disclosure statements following CMOS 18 standards.
- Identify ethical risks of AI-assisted drafting, including hallucinated citations, biased synthesis, and unverified claims.
- Conduct rigorous human verification protocols to fact-check AI-processed literature and citations.
- Distinguish among verbatim plagiarism, patchwriting, self-plagiarism, and improper attribution.
Authorship Integrity and COPE Guidelines
Defining authorship
According to the Committee on Publication Ethics (COPE) and major academic publishers, an “author” must be capable of taking legal, ethical, and intellectual responsibility for the work.
The non-human rule
Generative AI tools (e.g., ChatGPT, Claude) cannot be credited as authors or co-authors on an academic paper. COPE’s position is that they cannot take responsibility for the submitted work. As non-legal entities, they also cannot assert the presence or absence of conflicts of interest, manage copyright and license agreements, or accept legal liability for errors or misconduct.
Human accountability
Ultimate responsibility for the accuracy, integrity, and originality of a manuscript rests solely with the human author or authors, regardless of what tools assisted the drafting process. This includes any content that an AI tool produced: the human author is liable for any breach of publication ethics in it.
What a human author must be able to do
Widely used authorship criteria, such as those of the ICMJE, expect an author to:
- make a substantial contribution to the conception, design, data or interpretation of the work;
- draft the work or revise it critically for important intellectual content;
- approve the final version;
- agree to be accountable for all aspects of the work, including questions about its accuracy and integrity.
An AI tool cannot meet the last requirement, and so it cannot meet the criteria as a whole.
Policies differ, so always check
COPE sets a principle, but each journal, publisher, university and course applies its own rules about what AI use is allowed and how it must be disclosed. Read the relevant policy before you write, and ask your instructor or editor if it is unclear. “I did not know” is not a defence.
Extra practice (optional)
Write your explanation for a first-year student.
CMOS 18 Disclosure Protocols and Transparency
When and where to disclose
Two layers of guidance apply, and they work together:
- COPE and most journals: if AI tools were used in writing, producing images or graphical elements, or collecting and analysing data, say how and which tool was used, usually in the Materials and Methods or a similar section.
- CMOS 18: credit AI-generated text or images where they appear, in the text or in a note, and say whether and to what extent AI was used. Minor mechanical changes, such as applying smart quotes or adjusting fonts, do not need to be reported.
In practice, substantive uses such as brainstorming, structural editing, code generation, translation or analysis should be disclosed in a methodology note, acknowledgment section, or preface, unless your journal or institution says otherwise.
Citing AI-generated content (CMOS 18)
CMOS 18 treats chatbot conversations much like personal communications: they are cited in the text or a note where the content appears, and are not usually listed in a bibliography. A note typically identifies:
- the prompt, if it is not already clear from the text;
- the AI tool and its developer;
- the date the content was generated;
- a URL, if a public copy of the conversation exists.
Illustrative note: Text generated by ChatGPT, OpenAI, in response to the prompt “Summarize the main causes of urban heat islands,” January 15, 2026. If you edited the output, add a phrase such as “edited for style and content.” Always check the current edition and your institution’s guidance for exact formats.
Drafting standard disclosures
“The author utilized [AI Tool Name/Version] to assist with initial grammar polish and code optimization. All literature synthesis, empirical interpretation, and final drafting were performed exclusively by the author, who verifies the accuracy of all content.”
Usually acceptable with disclosure vs. not acceptable
| Usually acceptable, if policy allows and use is disclosed | Not acceptable |
|---|---|
| Grammar checking and proofreading | Generating fake empirical data |
| Brainstorming search keywords | Inventing non-existent citations (“hallucinations”) |
| Refining or drafting code, then testing it | Presenting AI-generated text as original human thought without disclosure |
| Translating your own draft, then checking it | Listing an AI tool as an author |
| Summarizing sources for orientation, then reading the originals | Uploading confidential or unpublished material, or participant data, to a public tool without permission |
| Creating an image, credited and permitted by the venue | Using AI to write a peer review of a manuscript you were sent in confidence |
Proofreading formatting and other mechanical edits normally need no special disclosure, but anything that changes wording, ideas, data or images does. When in doubt, disclose.
Phrase bank for disclosure statements
- “The author used [tool, version, developer] to [task] in [section or stage].”
- “Generative AI was used only for [language editing / translation / code drafting].”
- “All AI-assisted output was reviewed, corrected and verified by the author.”
- “The author takes full responsibility for the content of this paper.”
- “No generative AI tools were used in [the analysis / the writing] of this paper.”
- “The figure was generated with [tool, developer] using the prompt [prompt] and edited by the author.”
Extra practice (optional)
A student wrote a draft in Spanish, translated it into English with an AI tool, and then revised the English herself.
A researcher used an AI assistant to draft Python code for cleaning survey data, then tested the code and checked the outputs.
The journal asks authors to state whether AI tools were used.
The Danger of Hallucinated Citations and Human Verification
Understanding AI “hallucinations”
Large Language Models (LLMs) operate on probabilistic text prediction rather than database retrieval, frequently producing plausible-sounding but entirely fabricated journal titles, author names, DOIs, and direct quotes.
The human-in-the-loop protocol
Never accept an AI-generated quote or statistic at face value.
Track down the original primary PDF or database record for every cited source.
Verify page numbers, publication dates, and specific context before including any cited claim in a manuscript draft.
Verification checklist for every AI-suggested source
| Check | Question to ask |
|---|---|
| 1. Existence | Does the journal exist, and can you find this article in a library database, the publisher’s site or Google Scholar? |
| 2. Identifiers | Does the DOI resolve to the same article? Do the authors, year, volume and pages match? |
| 3. Content | Does the article really contain this quote, statistic or claim? Read the passage in the original. |
| 4. Context | What sample, setting and time frame does the finding apply to? Is it overstated (for example, “across all industries”)? |
| 5. Status | Is it peer-reviewed? Has it been corrected or retracted? |
| 6. Corroboration | Do other independent studies support the finding? |
Red flags for a fabricated citation
- The title, authors and journal cannot be found together in any database.
- The DOI does not resolve, or leads to a different article.
- The page numbers or volume do not match the journal’s real issues.
- The claim is unusually precise or sweeping and no other source repeats it.
- The quote does not appear anywhere in the source you find.
Protect confidential and personal data
Many journals, funders and ethics committees expect you not to upload unpublished manuscripts, peer-review material, interview transcripts or other participant data to public AI tools unless privacy and confidentiality are assured and you have permission. Check your ethics approval and data-management plan before using any AI tool with research data.
Extra practice (optional)
“Rahman and Ortiz (2022) found that 90% of students prefer video lectures (Journal of Digital Learning, 14(2), 55-70).”
“Rahman and Ortiz (2022) proved that 90% of students prefer video lectures.”
Plagiarism Spectrum and Self-Plagiarism
Patchwriting
Paraphrasing another author’s work by merely changing a few words or rearranging sentence structure while retaining the original syntax. This is a common error among non-native writers, who may not realize that surface-level rewording still counts as plagiarism.
Self-plagiarism and duplicate publication
Reusing substantial portions of one’s own previously published papers or course assignments without proper attribution or publisher permission. Even reusing your own earlier work can constitute academic dishonesty if undisclosed.
| Form | Description |
|---|---|
| Verbatim plagiarism | Copying another author’s exact words without quotation marks or citation. |
| Patchwriting | Superficially rewording a source while retaining its original structure and syntax. |
| Self-plagiarism | Reusing one’s own previously published or submitted work without disclosure. |
| Improper attribution | Citing a source incorrectly or misrepresenting what it actually says. |
Worked example: patchwriting vs. proper paraphrase
| Version | Text | Verdict |
|---|---|---|
| Original (illustrative) | “Students who receive feedback within two days are more likely to revise their drafts than students who wait a week.” | Source text |
| Patchwriting | “Learners who get comments within 48 hours are more likely to revise their work than learners who wait seven days.” | Not acceptable: same structure, only words swapped. |
| Proper paraphrase | “Lee (2022) reports that a quick turnaround of feedback is associated with more revision than longer delays.” | Acceptable: new structure, own wording, cited. |
Quote, paraphrase or summarize?
- Quote when the exact wording matters. Use quotation marks and give the source (with a page number where required).
- Paraphrase a specific passage by rebuilding the sentence in your own structure and words, keeping the meaning. Always cite.
- Summarize a longer text in a few sentences that capture its main point. Always cite.
Reusing your own work
Rules on reusing earlier assignments, theses or published text depend on the institution and the publisher. In general, disclose the earlier work, cite it, and get permission where needed. Ask your instructor or editor before you reuse substantial text, including from a methods section.
Extra practice (optional)
Original: “Regular short breaks during study sessions help students to maintain concentration over long periods.”
Write two or three practical steps.
AI Ethics and Disclosure Audit Workshop
Evaluate each scenario below and determine the appropriate ethical course of action.
Scenario A: A researcher uses ChatGPT to generate a literature review section, pastes it directly into their manuscript, and adds real-looking citations without checking if the papers exist.
Ethical assessment: Severe academic misconduct — uncredited text generation and unverified or fabricated citations.
Corrective action: Rewrite the section independently, locate genuine peer-reviewed primary sources, and verify all claims.
Scenario B: A student uses an AI tool to translate a section of their draft from their native language into English, then uses a grammar checker to refine the prose.
Ethical assessment: Permissible technical assistance, provided it is disclosed.
Corrective action: Include an explicit disclosure note in the paper stating that translation and language-editing tools were utilized.
Extra scenarios (optional)
For each scenario, write your assessment and the action you would take, then compare it with the sample.
A researcher uses an AI assistant to write analysis code, runs it without testing, and does not mention AI in the paper.
A student includes an AI-generated diagram in a thesis without any credit.
A doctoral student pastes full interview transcripts into a public chatbot to speed up coding.
An author copies the methods section of their own published paper word for word into a new paper without a citation.
Literature Synthesis and Ethics Audit
Below is a 300-word draft containing un-synthesized literature, unhedged claims, an unverified AI-generated citation, and no disclosure statement.
Task: Rewrite the draft into a cohesive, thematic literature review. Correct the improper citation (the Whitmore and Chen reference and its “40% across all industries” claim should be treated as an unverified, potentially hallucinated citation and either removed or clearly flagged pending verification). Perform a mock human-verification check by noting, in brackets, what you would need to confirm before trusting this source. Calibrate unhedged claims appropriately. Append a compliant CMOS 18 AI Disclosure Statement at the end to demonstrate the correct format. Label it clearly as a practice statement: in real work, a disclosure must describe only what you actually did.
Model answer: thematic literature review, verification note and disclosure (202 words in the review)
Employers are increasingly adopting artificial intelligence (AI) tools in hiring, although the extent of adoption varies and needs further documentation [citation needed]. Research on these tools can be grouped into three themes: effectiveness, candidate experience and regulation.
In terms of effectiveness, managers report that AI tools may save time in screening [citation needed]. One study reportedly found that AI-assisted hiring reduced bias by 40% (Whitmore & Chen, 2024) [UNVERIFIED: see verification note]. However, a claim of this size that applies “across all industries” is unlikely to be supported without strong evidence, and it should not be relied on until confirmed.
Turning to candidate experience, some candidates may feel disrespected by algorithmic screening [citation needed]. This suggests a possible tension between efficiency gains and perceived fairness.
On the regulatory side, human resources departments appear to be struggling to keep pace with new rules, and some jurisdictions have restricted particular AI hiring practices [citation needed]. Regulators in the EU and the US have taken different approaches [citation needed], and trade unions have begun to resist automated decision-making in employment [citation needed]. Consequently, organizations that overlook these concerns may face legal risk. Overall, the long-term effects remain uncertain, and further research and clearer regulation are needed.
[Verification note: before trusting Whitmore & Chen (2024), I would (1) confirm that the Journal of Applied Organizational Psychology exists and is peer-reviewed; (2) locate the article through its DOI, volume and page numbers; (3) read the full text to see whether the 40% figure appears and what it measures; (4) check the sample, the sectors covered and how “bias” is defined; (5) check for corrections or retractions; and (6) look for independent studies that support the finding.]
Practice AI disclosure statement (model format only): “The author used [AI tool name and version, developer] on [date] to check grammar in the final draft. All source selection, literature synthesis, interpretation and final drafting were carried out by the author, who verified every citation against the original source and takes full responsibility for the content.”
Why this works: The draft is now organized by theme (effectiveness, candidate experience, regulation) rather than by isolated statements, and it uses signposts such as “In terms of,” “However,” “Turning to,” “On the regulatory side” and “Consequently.” Informal claims (“love,” “hate,” “Nobody really knows”) are replaced with hedged, evidence-ready wording (“may,” “appear to,” “suggests”). The Whitmore and Chen claim is flagged instead of trusted, and the verification note lists the checks from Phase 3. Claims that need sources are marked [citation needed], since no real sources were provided. The disclosure statement follows the COPE and CMOS 18 pattern: tool, task, human responsibility. It is labelled as practice only.
Submit your revised literature review and disclosure statement using the form below.
Session 16 Summary Sheet
Download a two-page PDF summarising COPE authorship rules, disclosure and citation of AI use, the verification checklist, data protection, and the plagiarism spectrum from this session.
Session 16: Research Integrity and AI Ethics
COPE authorship · disclosure statements · human verification · plagiarism spectrum
Module 4 complete
You have now completed all four sessions of Module 4: source hierarchy and substantiation, literature synthesis, functional phraseology, and research integrity with AI ethics. Download your takeaway summary any time from the previous step.