· By Thesisroom
Updated
Can You Use ChatGPT for Your Thesis? What Is Safe and What Is Not
Can I use ChatGPT for my thesis is the question every doctoral candidate asks. Here is the honest line between uses that stay safe under current university policies and uses that risk your degree.
The one line to look for in your university’s policy
Can I use ChatGPT for my thesis is a question with a moving answer. What is stable across current university policies is one distinction: whether the tool does the intellectual work or whether it supports work you have already done. Cambridge treats unacknowledged AI use in assessments as academic misconduct (Cambridge University, 2026). Harvard, MIT, Stanford, and Oxford leave specifics to individual departments and instructors, but every institution now expects some form of disclosure when AI shapes submitted work (Thesify AI policy summary, 2026).
The rest of this piece walks the safe uses, the disclosure-required uses, and the prohibited uses under most current policies. None of the categories is a substitute for reading your own institution’s specific rules, which change more often than they used to.
The check-versus-do line in practice
One way to read any of these policies is to sort each use into check or do. Tools that check work you have already produced sit on the safer side. Tools that produce content intended for submission sit on the other. The line is not about how sophisticated the tool is: a simple spell-checker is a check tool, and a powerful model whose text you paste into your thesis is a do tool. What matters is whether the intellectual work stayed with you.
The check side includes:
- Grammar and spelling checkers running on your own writing
- Citation verification against academic databases and DOI resolvers
- Structural review of chapters you have written
- Question generators for rehearsing your defense
- Retraction monitoring for sources you have saved
- Plagiarism cross-checks against published sources
The do side includes:
- Generating chapters or paragraphs for direct inclusion
- Producing bibliography entries that then enter your reference list
- Writing the interpretation of your findings or your discussion section
- Producing analytical arguments you could not defend in detail
If your university asks you to disclose AI use, be specific: name the tool, the purpose and the stage of the work. “I used DeepL to translate three passages from German to English in chapter four, and Grammarly for grammar checking throughout” tells an examiner far more than “I used AI”.
Safe uses: ChatGPT for support, not for output
Three uses sit clearly on the safe side of the line at most institutions.
Explaining a concept you are trying to understand. Asking ChatGPT to explain a methodological choice, a theoretical framework, or a specific technical term is comparable to asking a librarian or a colleague. The output is background reading, not thesis content.
Editing your own writing. Pasting a paragraph you wrote and asking for suggestions on clarity or grammar is close to what Grammarly does. The intellectual work is yours; the tool sharpens the prose. Grammarly, ProWritingAid, and Paperpal do this in more targeted ways than general-purpose ChatGPT.
Brainstorming research questions. Using ChatGPT to explore possible angles on a topic, then choosing one to develop into a formal research question, is an ideation aid. The final research question, the justification for it, and the literature review that supports it are your intellectual work.
For all three, disclosure requirements vary. Some institutions treat these as low-risk and not requiring specific declaration. Others (Cambridge, Princeton) prefer disclosure of any AI use in assessed work.
Disclosure-required uses: safe with declaration
Three uses fall in a middle band. Most institutions allow them but require the candidate to declare the use.
Translation. Using DeepL or ChatGPT to render your writing between languages, or to read sources in a language you do not fluently read, is widely accepted. Disclose in the methodology or acknowledgements when translation shaped substantive passages.
Summarising papers you have read. Using an AI tool to summarise a long paper you are considering citing can save time. The catch: the summary is not a substitute for reading the paper. If you cite the paper based on the summary alone, you may misquote or miscite. Read the paper before citing.
Paraphrasing your own writing. Using ChatGPT to rewrite a passage you have already drafted is close to editing. Some institutions treat this as unproblematic; others require declaration. If your paraphrase changes the argument, not just the wording, the intellectual work has partially shifted to the tool.
Prohibited uses: what puts your degree at risk
Three uses sit clearly on the prohibited side of that line; confirm them against your own university’s policy.
Generating chapters or sections for direct inclusion. Asking ChatGPT to write a literature review, a methodology chapter, or an analysis section and submitting the output as your work is academic misconduct at essentially every institution. Detection tools exist (Turnitin AI detection, GPTZero, Copyleaks) with imperfect accuracy but non-zero detection rates. Turnitin claims sub-1% document-level false positive rate for detections above 20% AI content (Turnitin, 2026).
Generating citations for your reference list. ChatGPT and similar tools produce plausible-looking citations that often do not resolve. Walters and Wilder (2023) found 55% of GPT-3.5 references and 18% of GPT-4 references were fabricated across 636 citations in 84 literature reviews. Bhattacharyya and colleagues (2023) found 47% of ChatGPT-generated medical references were completely fabricated. A fabricated citation surfacing in the viva is a live integrity issue.
Generating your findings or interpretations. Asking a language model to interpret your data, to draw conclusions from your analysis, or to write the discussion section shifts the doctoral judgement to the tool. The doctoral standard rests on the candidate’s own reasoning; delegating this to AI defeats the point of the degree.
The viva is the real detector
Universities use AI detection tools, but the more reliable check is the viva. An examiner asking specific questions about your methodology, findings, and citations catches AI-generated content faster than any detector. AI-generated arguments do not hold up under specific questioning from someone who knows the field.
Mullins and Kiley (2002), interviewing 30 experienced examiners, note that panels form early impressions from structure, references, and the opening chapter. A candidate whose chapters cannot be defended in specific detail is exposed regardless of what detection tools flagged in advance.
The safest position is straightforward: use AI for support, not for output. The Thesisroom method sits on this side of the line. Chapter Audit reviews chapters you have written; it does not draft them. Citation Check verifies citations you already have; it does not generate them. Defense Practice generates questions from your thesis for you to answer; the answers are yours.
What Thesisroom does with this
Thesisroom is built to sit clearly on the safe side of the AI-in-academia line. Every module reads what you have written and reports back; none generates content for submission. The integrity policy sets this position formally: the writing stays yours, the thinking stays yours, and the tool never produces submittable content.
For candidates who want to use ChatGPT alongside Thesisroom, the safer pattern is to use ChatGPT for concept explanation and editing, then use Thesisroom to check the writing and rehearse the defense.
Frequently asked questions
Is using ChatGPT to help me edit my thesis considered cheating?
At most institutions, no, if the intellectual work is yours and you disclose the use. Editing your own paragraphs is different from generating new ones. Some universities require declaration of any AI use in assessed work; check your specific policy.
Can I use ChatGPT to generate my bibliography?
No. AI-generated citations are unreliable at scale. Walters and Wilder (2023) found 18% to 55% of AI-generated citations were fabricated. Every citation in your reference list needs to be verified against Crossref, OpenAlex, or an equivalent database before it enters the thesis.
Will my supervisor know if I used ChatGPT?
Sometimes. Stylistic markers of generated text are detectable by tools and by experienced readers. More reliably, the viva exposes AI-generated content when the candidate cannot defend specific chapters in detail. The safer position is honest disclosure of what tools were used and how.
Is Grammarly the same as ChatGPT for these rules?
Similar in principle, different in scope. Grammarly checks existing prose. ChatGPT can generate new content. Most institutions treat Grammarly as low-risk (comparable to a spell-checker). ChatGPT for editing is close to Grammarly; ChatGPT for drafting is closer to ghostwriting.
Can I use ChatGPT to translate my thesis into English?
Generally yes, with disclosure. Translation is a widely accepted use of AI tools in academic writing. Substantive translated passages should be noted in the methodology or acknowledgements. The intellectual work of the thesis remains yours; translation is a technical step.
How does Thesisroom differ from ChatGPT for a thesis?
ChatGPT is a general-purpose language model that can generate text. Thesisroom is built to review, verify, and question text you have already produced. It does not generate citations, chapters, or arguments. The distinction is between a tool that does the work and a tool that checks the work.
One takeaway before you close this tab
Can you use ChatGPT for your thesis? Yes for support work (concept explanation, editing, brainstorming) with appropriate disclosure. No for output work (chapter drafts, citations, findings interpretation). When you want the check side of the AI line covered without disclosure worries, Thesisroom’s method is built for that seat.
References
Bhattacharyya, M., Miller, V. M., Bhattacharyya, D., & Miller, L. E. (2023). High rates of fabricated and inaccurate references in ChatGPT-generated medical content. Cureus, 15(5).
Cambridge University. (2026). Academic misconduct and AI policy.
Mullins, G., & Kiley, M. (2002). ‘It’s a PhD, not a Nobel Prize’. Studies in Higher Education, 27(4), 369-386.
Thesify. (2026). Generative AI policies at top universities.
Turnitin. (2026). Understanding false positives within AI writing detection.
Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13.