I’ve been using ChatGPT for two years now and have seen a dramatic improvement in its capacity to write. It has become one of my supervisors in many ways, as the questions I might ask one of my supervisors I can put to ChatGPT instead, saving everyone time.
However, it is incredibly easy to get ChatGPT, and other generative AI large language models to write just about anything, then make some minor adjustments, check or add references and it’s done. In business, this is great. In preparing lessons for teaching, this is great. In learning and writing a PhD thesis, this is not great. The whole point of education is to learn. If we let ChatGPT do everything for us, where’s the learning?
So, in case you’re using generative AI to check your writing, like Word and Grammarly check your spelling and grammar, here’s a prompt I used to use which leaves most of the writing up to you, if you’re posting draft paragraphs for review:
Please review this section as a PhD supervisor would, suggesting in-line refinements, theoretical frameworks, and methodological improvements, to help me to strengthen my literary analysis. Keep my original writing intact but suggest enhancements to the vocabularly used in the argument where necessary, ensuring it aligns with PhD-level critical engagement.
Update this prompt to match your level and discipline of study. Note that this is only for early drafts to get you started. This will help maintain academic integrity as ChatGPT is not writing it for you, only giving suggestions. Also, these suggestions still need to be researched by you, and even then, after you’ve written your first draft, your supervisors are likely to advise on how to rewrite it again to be better. Remember that supervisors know how examiners will look at your writing and will be advising from that perspective. ChatGPT is suggesting improvements from the point of view of writing best practices, or what editors would do, or similar sentences it has stored that it wants to replace yours with, which may not be appropriate.
Update 2025: I don’t use this prompt now that I’m in my third year. That was just to get me started on my first drafts, to see how I should write rather than what I should write. Now, I use the ‘Customise ChatGPT’ field to guide answers to questions. I also rely on my supervisors more this year who look at my work with many years experience and point to areas that should be composted, or discarded, and point to arguments that don’t have enough support – something ChatGPT, with its ‘wants to please’ nature, isn’t likely to.
I have this embedded in the Customise ChatGPT field:
“You are my PhD supervisor and research assistant that is:
- neutral, unbiased and analytical.
- Gives detailed and explanatory responses.
- Uses academic vocabulary related to the humanities and digital humanities disciplines
- Include full direct URL to any reference found.
- Python scripts must have a # comment explaining each section as I am a Python beginner.
- Consider that my goal is to fulfil the following AQF level 10 framework criteria:
Graduates at this level will have: - A systematic and critical understanding of a complex field of learning and specialised research skills for the advancement of learning and/or for professional practice.
- A systemic and critical understanding of a substantial and complex body of knowledge at the frontier of a discipline or area of professional practice.
- expert, specialised cognitive, technical and research skills in a discipline area to independently and systematically:
• engage in critical reflection, synthesis and evaluation
• develop, adapt and implement research methodologies to extend and redefine existing knowledge or professional practice
• disseminate and promote new insights to peers and the community
• generate original knowledge and understanding to make a substantial contribution to a discipline or area of professional practice”
AQF stands for Australian Qualifications Framework
If you’re in the early stages, getting some feedback from ChatGPT on your first drafts can help you get over some minor hurdles, rather than wait for your busy supervisor to get back to you. However, third year, my feeling is that it should all be by you with more structured help from your supervisors. ChatGPT is not a supervisor employed at your university. What it accesses is an amalgamation of many universities’ reports and many pieces of advice online reflecting multiple countries’ frameworks that, combined, is unlikely to be entirely relevant when you’re getting your passion project ready for submission in your specific university.
From another perspective, ChatGPT is actually incapable of writing a PhD dissertation anyway. It can’t really deal with more than 1000 words when analysing, and it can’t access the earlier parts of a long conversation or multiple conversations, and it can’t consider what you’ve already done in the past few years and what you’re likely to do next. It can’t suggest an appropriate path to follow if it can’t consider all the other paths you’ve already taken. (If you approach your use of it with the idea that it can only act with 1000 words in its ‘head’ at any time, you’ll realise how limited it is. In my experience, more words than that, and it starts to hallucinate.) Also, on the most basic level, it is word matching, so if you ask for an appropriate theoretical framework for your idea, the one it suggests might lexically match what you’re writing, but may not actually be relevant. And, what it says the first time will be different to what it says the second time. Refresh if the first analysis isn’t useful and you’ll see how different the second one is.
ChatGPT is great when you’re stuck on something small and don’t want to bother your supervisor about it. Or scholar.google.com is not helping you find a particular quote. Give a bullet point question and see if ChatGPT can help. When we post a question in Google, we automatically get Gemini’s take on the answer. It’s not against Academic Integrity to read Gemini’s answer or even use it to research further into what scholars have said about a particular subject. However, it is, if you just copy Gemini’s answer into your dissertation and claim it as your own.
By using this prompt, (and ChatGPT’s ‘customise’ field) you can be informed on new paths to focus on or new areas to explore, in the same way that you might from your supervisor, then you can do further research to help elevate your PhD writing and content, saving everyone time and maintaining your academic integrity. Remember, ChatGPT is a tool. Use it as a tool and you should be okay. Never try to get it to write whole paragraphs for you. At the very least, a thesis full of em dashes isn’t going to make your examiner happy!
Hope this helps!
Disclaimer: This is my opinion only and does not represent any university. It is necessary that you look into your university’s policies on this matter so that you are fully informed of what you can and can’t do. For example, at the ANU, the only AI recommended is CoPilot Enterprise. The ANU has a great overview of the dos and don’ts of using generative AI here: Guide for students: best practice when using Generative AI
