Blog

· By Thesisroom

Updated

Is Using AI for a Thesis Academic Misconduct? The Honest Answer

Is using AI for a thesis academic misconduct? Depends on the use and the policy. Here is the honest breakdown: what is clearly allowed, what needs disclosure, and what triggers formal proceedings.

A ruled cream notebook page with a faint ring stain and a ribbon bookmark, on a dark wooden table.

The answer nobody wants to hear: it depends

Is using AI for a thesis academic misconduct? The honest answer is that it depends on the use and on the institution’s specific policy. There is no international standard. Cambridge treats unacknowledged AI use in assessments as misconduct. Harvard, MIT, Stanford, and Oxford leave the details to individual departments while requiring some form of disclosure in most cases (Thesify AI policy summary, 2026). Some fields prohibit AI use in specific assessments entirely; others allow it with clear boundaries.

What every current policy has in common: the intellectual work must be the candidate’s, and any AI use that shapes submitted work needs to be disclosed. Failing to disclose, or delegating the intellectual work to a tool, is where misconduct starts.

Three tiers, roughly

University policies tend to sort AI use into three tiers. Some uses are clearly allowed, such as editing your own prose or asking for a concept to be explained. Some are allowed with disclosure, such as translating or paraphrasing substantive passages. Others are prohibited, such as generating chapters, citations or findings for submission. Our guide on what is safe and what is not when using ChatGPT for your thesis walks through each tier with examples. The rest of this piece is about what happens when a use crosses the line.

What triggers formal proceedings

Three specific patterns trigger academic misconduct investigations at most institutions.

Fabricated citations in the bibliography. An examiner or integrity officer resolves a DOI, finds it does not exist, and the citation is documented as fabricated. This is one of the fastest routes to a formal investigation. Walters and Wilder (2023) found 18% to 55% of AI-generated citations were fabricated across their audit; a single fabrication surfaced in a reference list is a live issue.

Text-similarity flags to AI content. Turnitin’s AI detection flags text with 20% or more AI-generated content at above 1% false positive rate (Turnitin, 2026). GPTZero and other detectors add signal. High flags trigger review; combined with other evidence, they trigger proceedings.

Viva exposure of chapters the candidate cannot defend. An examiner asking specific questions on a chapter and receiving vague or wrong answers signals that the candidate did not do the intellectual work. Mullins and Kiley (2002) note that panels form early impressions from structure, references, and the opening chapter; a chapter that cannot be defended in specific detail is a live signal.

Any one of these triggers a review. Combinations trigger faster and heavier proceedings.

What formal proceedings look like

The specific procedure varies by institution, but the pattern is consistent.

Initial flag. An examiner, integrity officer, or automated system raises a concern. The candidate is notified in writing.

Response window. The candidate is given time (usually days to weeks) to respond, present evidence of their own intellectual work, and explain any AI use.

Panel review. A panel (usually including academic staff outside the candidate’s department) reviews the evidence and decides on findings.

Outcome. Ranges from no action, through required corrections and resubmission, to termination of registration and revocation of a previously awarded degree in severe cases.

Candidates who can produce a paper trail (dated drafts, reference-management records with save dates, supervisor emails showing the intellectual work developing under human effort) resolve investigations faster than those who cannot.

How to stay clearly on the safe side

Three habits keep AI use clearly within current policy.

Use AI for support, not for output. Concept explanation, editing, brainstorming, and translation are Tier 1 or Tier 2 uses at most institutions. Drafting, citations, and findings interpretation are Tier 3.

Verify every citation. AI-generated citations are the fastest route to a fabrication finding. Every reference in your bibliography needs to resolve against Crossref, OpenAlex, or an equivalent database. Thesisroom’s citation check runs the verification in one pass, drawing on nine registries.

Disclose what you used. When in doubt, declare. Cambridge and Princeton require disclosure of AI use in assessed work; other institutions are moving the same direction. Declaring Tier 1 or Tier 2 use is not misconduct; failing to declare Tier 3 use is.

The Thesisroom method sits clearly on the safe side. Every module reads what you have written and reports back; none generates content for submission. The integrity policy sets this position formally.

What Thesisroom does with this

Thesisroom is built to be a Tier 1 tool: it checks the work you have already produced. Citation Check verifies existing citations. Chapter Audit reviews existing chapters. Defense Practice generates questions from your existing text. None of the modules generates content for inclusion in the thesis.

The paper trail from these sessions supports the honest-use position. Dated verification runs, reviewed chapters with timestamps, and rehearsal sessions with logs all serve as evidence of the candidate’s own intellectual work.

Frequently asked questions

Is using Grammarly considered academic misconduct?

At almost every institution, no. Grammarly corrects existing prose rather than generating new content. It sits in Tier 1 (allowed, often without disclosure). Some universities have moved toward requiring disclosure of any AI use in assessed work; check your institution’s specific policy.

Is using ChatGPT for editing considered misconduct?

Depends on how it is used. Editing your own writing with ChatGPT is closer to Tier 1 or Tier 2 at most institutions. Generating new writing for direct inclusion is Tier 3. The line is between correcting text and producing text.

What if I use AI and do not disclose it?

At institutions that require disclosure (Cambridge, Princeton, and a growing number of others), undisclosed use is treated as misconduct regardless of what the use was. At institutions without a disclosure requirement, the misconduct threshold is at the intellectual-work line, not the disclosure line.

Can I appeal an academic misconduct finding?

Yes. Every institution has an appeals process, usually on procedural grounds (fair process, adequate response opportunity, evidence review). Successful appeals typically hinge on evidence the panel did not consider, not on disagreement with the finding itself.

What is the safest way to use AI for a thesis?

Use it for concept explanation, editing your own writing, and brainstorming, with disclosure appropriate to your institution’s policy. Verify every citation against Crossref or an equivalent database. Do not have AI generate chapter drafts, citations, or findings interpretations. Keep a paper trail of your own intellectual work.

Does Thesisroom stay within academic integrity rules?

Yes. Thesisroom does not generate submittable content. Every module reads what you have written and reports back. Citation Check verifies existing citations; Chapter Audit reviews existing chapters; Defense Practice generates rehearsal questions from your text. The tool is designed to sit clearly on the safe side of current AI-in-academia policies.

One takeaway before you close this tab

Is using AI for a thesis academic misconduct? Not universally, but the line is at intellectual work: if the tool did the thinking, it is misconduct at most institutions. Use AI for support (editing, translation, brainstorming), verify every citation, and disclose per your policy. Thesisroom’s method sits on the safe side, so nothing it produces enters your thesis.

References

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 capabilities.

Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13.

misconduct · work · thesis · tier · citations