You can use AI tools like ChatGPT in your PhD for limited support tasks, but not for writing your thesis or papers. By support we mean small jobs like grammar checking in your own chapter, explaining a concept or finding the bug in your code. Even for these, your university must allow it, and you must disclose it where disclosure is required. Don’t use AI tools to write, to generate references or to analyse confidential data. Big publishers like Elsevier, Springer Nature and IEEE agree on the basics, and so does the Committee on Publication Ethics (COPE). For all of them an AI tool cannot be an author, and the human author answers fully for whatever is submitted.
Scholars mostly want to know where the line is and what publishers ask for. We cover that, and what we found (and did not find) from UGC. After that we come to AI detectors and what to do if your own writing gets flagged. These AI rules are changing every few months, and we checked all of it in September 2026. Still, your university’s current rule and your target journal’s current policy come first.
Last reviewed September 2026.
Is using ChatGPT allowed in a PhD?
That depends on what you use it for, and on your university. India has no single national rule saying yes or no. Asking a tool to explain a regression output is very different from asking it to write your literature review.
All the publishers and most universities do agree on one basic rule. Every word, number and reference in your work is your responsibility, and you must be able to explain and defend all of it. Examiners read a thesis as evidence that you can do research. If somebody else or some tool wrote the text, it shows nothing about you, however nicely it reads. Our own ethics policy draws the line at the same place for the support we give.
| Use | Generally acceptable? | Conditions and risks |
|---|---|---|
| Checking grammar and spelling of your own text | Yes, at most universities and journals | Elsevier does not ask you to declare basic grammar, spelling and punctuation checks. IEEE leaves grammar editing out of its policy. Even so, their advice is to disclose it. Check your university’s rule also |
| Explaining a concept or a statistical output you are trying to understand | Usually | Cross-check it with a textbook or with your supervisor. These tools say wrong things with full confidence |
| Suggesting search terms or synonyms for a database search | Usually | Do the actual search yourself in Scopus, Web of Science or PubMed |
| Debugging or explaining code (Python, R, MATLAB) | Usually, with disclosure where required | Any line that goes into your analysis, you must understand it and be able to explain it |
| Summarising a paper you have already read | With care | Compare the summary with the paper itself. Never cite a paper you know only through a summary |
| Restructuring or substantially rewriting your paragraphs | Grey area; disclose | Most publishers see this as more than copy editing. Your university may also restrict it |
| Generating text, arguments or literature review sections for the thesis | No, unless your university explicitly allows and you disclose | The thesis has to be your own work. Examiners will test that, and the viva will test it again |
| Generating references | No | The tools give references that look genuine but do not exist |
| Uploading interview transcripts, patient data or unpublished results to a public tool | No, without ethics approval and consent | It breaks the confidentiality you promised to participants. It may break data protection law also |
| Using “humaniser” or AI-bypasser tools to disguise AI text | No | This counts as concealment. Turnitin says it now flags text that is likely modified by bypassers |
This table is our reading of current practice, not a rule book. Some universities are stricter and ask you to declare even a grammar tool. A few allow more. If your university has given a circular on AI, that circular overrides anything here.
What do journals and publishers require?
If you plan to publish from your thesis, the journal’s policy is as important as your university’s rule. We checked the main positions on the publishers’ own pages in September 2026.
| Body | Can AI be an author? | What must be disclosed, and where | Notable limits |
|---|---|---|---|
| COPE (position statement, 2023) | No | Which AI tool was used and how it was used, in the Methods or a similar section | Authors carry full responsibility for any content produced with AI |
| Elsevier (journals policy) | No | A separate AI declaration placed before the references. It names the tool, the purpose and how the authors oversaw the work. Basic grammar, spelling and punctuation checks are exempt | No creating or altering research images with AI. No fabricated data or references |
| Springer Nature (risk framework) | No | For low-risk assistive use, disclosure is encouraged. For interpretive use it is required, along with human verification | Not permitted are fabricating data or citations, AI-generated conclusions presented as human work, and uploading manuscripts to public AI tools |
| IEEE | No | AI-generated content (text, figures, images, code) goes in the acknowledgments, naming the system and the sections where it was used | Use for editing and grammar is generally outside the policy, though disclosure is recommended |
On a few points all of them agree. None of them accept AI as an author. The human authors answer for accuracy, and that includes mistakes the tool brought in. Fabricated data, results or references count as misconduct, whether a person or a tool made them.
They differ in the details. One is where the disclosure goes, as a separate statement, in the Methods section or in the acknowledgments. What counts as minor editing also changes from publisher to publisher. Many journals also keep extra rules over the publisher’s policy, so read that journal’s author guidelines once before you submit.
What a disclosure looks like
A disclosure is short and says exactly what was done, like this sample. “During the preparation of this work the author used ChatGPT (OpenAI, [model and version]) to improve the readability of the introduction and discussion. The author reviewed and edited all output and takes full responsibility for the content.” Put the actual tool and version you used. Say the tool corrected language in the introduction only. Then mention that much only.
What do UGC and Indian universities say about AI?
The national rules on plagiarism in theses are the UGC (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations, 2018. These were framed before generative AI came into common use, so they say nothing about it. In this review we could not find any UGC regulation or notice on ugc.gov.in dealing with generative AI in PhD theses. A few websites are putting up “new UGC AI guidelines” with detailed rules. We would not trust such page much till it links to an official UGC document.
So universities actually follow the 2018 regulations plus their own academic integrity policy. Presenting text you did not write as your own work is an integrity problem, whether a person wrote it or a tool. Several Indian universities and institutes have already issued their own circulars or policies on AI use, and more are expected. Check with your research cell or your guide whether your university has one, and keep a copy.
We have written about the similarity levels and the penalties in the 2018 regulations in our plagiarism limits guide.
Can Turnitin or DrillBit detect AI writing?
Turnitin and DrillBit both show AI-writing indicators now, along with the similarity report. (DrillBit is the checker that INFLIBNET gives to universities through ShodhShuddhi.) The number they give is only a probability estimate. It cannot prove anything.
Turnitin’s own guidance is careful here. As per Turnitin, institutions should not penalise a student on the AI writing score alone. Somebody has to judge that score along with the institution’s policy. Turnitin saw more false positives at low scores. For that reason, its report gives an asterisk in place of a percentage for AI scores between 1% and 19%. In August 2025 Turnitin announced that its detection now also covers text likely modified by “AI bypasser” or humaniser tools. This feature is trained and tested for English.
For a scholar this means a few things.
- False positives happen. Formulaic academic prose can get flagged. So can heavily edited text, and writing by non-native speakers of English, even when the scholar wrote it.
- Running a paraphrasing tool can increase your AI score. Suppose you run a chapter through a rewriter only to bring similarity down. That chapter can then show up as AI-modified. More on this is there in our paraphrasing guide.
- A low AI score is not permission. Detectors miss things. And examiners and viva panel test your understanding directly.
The similarity side is covered in the Turnitin report guide.
Why do AI tools invent references?
A chatbot just writes the words that usually come after your kind of prompt. A reference list normally has author names, a likely title, a journal and a year. So the tool gives exactly that, whether the paper exists or not. You may get real author names and a real journal also. But the DOI goes nowhere, or it opens some other paper.
Examiners do check references, and even one fabricated reference in a thesis is serious. When they find a citation to a paper that doesn’t exist, they understand you never read what you cited. Elsevier and Springer Nature policies both name fabricated citations as unacceptable.
- Never copy a reference from a chatbot into your reference manager. Find the paper in Google Scholar, Scopus, Web of Science or the publisher’s site, and import the reference from there.
- Some tools search real databases and link to their sources. Such tools reduce the risk to some extent only. Open the source and check it really gives the informations the tool is claiming.
- Keep your references in Zotero or Mendeley, imported from the database or the DOI, so you can trace each one back.
How do you protect yourself if your work is questioned?
What protects you most is proof of how you did the work. Start collecting that proof today itself. Old drafts have a habit of vanishing from laptops, and after a report flags you it is too late.
- Maintain a simple AI-use log with the date, the tool and its version, what you asked, what you used and how you checked it.
- Keep your own drafts and their version history, through Word version history, Google Docs history or files saved with dates.
- Ask your supervisor in writing which uses they are fine with, and follow any circular your university gives out.
- Any factual statement, number or reference a tool gives you, check it against the original source.
- Participant data, unpublished results and a manuscript under review should never be pasted into a public AI tool.
- When you disclose AI use, follow the format your university or the target journal has given, not a format of your own.
Confidentiality and data protection
Interview transcripts, survey responses with identifiers, patient records and unpublished results should not go into a public AI tool. At consent time participants were told how their data would be handled, and the ethics approval says the same. A public AI tool was almost certainly not part of that promise. For personal data, India’s Digital Personal Data Protection Act, 2023 also applies. If you want to run an AI tool on research data, first discuss about it with your guide and the ethics committee. If your institution has approved some tool, use that one.
If you are accused
Don’t panic. Ask for the specific evidence. Share your drafts, version history, notes and AI-use log. Offer to explain any part of the work in person, because that is the strongest evidence you can give. If they are going by the detector score only, politely mention that Turnitin itself wants a person to judge that score. A scholar who can show how the work developed is in a strong position.
Want a specialist to help with this in your own work? Research Topic and Gap Selection: a free first consultation, then a written scope and fixed quote.
Sources
- COPE position statement: Authorship and AI tools
- Elsevier: Generative AI policies for journals
- Springer Nature / Nature Portfolio: Artificial Intelligence (AI) editorial policy and risk-assessment framework
- IEEE: Author guidelines for artificial intelligence (AI)-generated text
- Turnitin Guides: Using the AI writing report
- Turnitin press release (August 2025): AI bypasser detection
- UGC (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations, 2018 (PDF, ugc.gov.in)
- APA Style blog: How to cite ChatGPT
FAQ
Questions scholars ask
Can I use Grammarly or QuillBot on my thesis?
Correcting grammar in your own text is mostly accepted, though some universities want you to declare it. The paraphrasing and rewriting features are another matter. Those features change full sentences on their own, and a chapter changed like that can easily get an AI flag. Read your university rule for thesis chapters before trying them.
Should I cite ChatGPT as a source?
Not as a source of facts. Chatbot output is not reliable, and nobody else can get back the same output. If you quote or analyse AI output in your research, APA and other styles have formats for citing it. When the help was only on the language side, you just give a disclosure statement.
My own writing was flagged as AI. What should I do?
First gather your drafts, notes and version history. Then ask for a chance to explain the work in person. Turnitin has written in its guidance that a human must look at the score first, before any decision. Do not put the text through a humaniser tool. That only makes things worse for you.
Can AI tools help with my statistical analysis?
Yes. Up to a point. You can ask it to explain a method simply, and it can also check your code when it keeps failing. But if a tool recommends which test to run or explains your outputs, Springer Nature wants human verification and disclosure for that use. Choosing the analysis and justifying it is your own job, and at the viva you must be able to explain it.
Can I use AI to translate my thesis from Tamil or Hindi into English?
Ask your university first. If you wrote the text and the tool only translates it, that is nearer to language support than generating content. Still, a thesis translated by a tool needs careful checking by you, and ideally by an editor who knows your field. Most publishers also expect you to disclose it.
