Generative AI tools can summarise an article in seconds, suggest search terms and tidy a paragraph’s grammar. They can also invent references that do not exist, flatten your argument into generic prose and put your dissertation in breach of your faculty’s rules. For a South African postgraduate in 2026, the question is no longer whether generative AI exists but exactly what your university allows, how to disclose it, and where it quietly fails. This guide explains how generative AI works in plain terms, the failure modes that matter for a dissertation, what two South African universities’ published rules actually say, and a safe way to work. If you want help with the shape of the work rather than the words, Tesify helps you structure your dissertation chapter by chapter.
What generative AI is, in one paragraph
Generative AI refers to models that produce new text, images, code or audio in response to a prompt. Large language models, the kind behind chat assistants, are trained on very large amounts of text and generate a response by predicting what text is likely to come next, given your prompt and what they learned. That design explains both their strengths and their weaknesses: they are fluent and fast at producing plausible text, but plausibility is not the same as accuracy, and the model has no built-in way of knowing whether a sentence it produces is true.
Where generative AI fails in a dissertation
These are the failure modes supervisors and examiners in South Africa now see regularly:
- Invented or distorted references. A chat assistant asked for sources can return authors, titles and journals that look real but do not exist, or real papers that say something different. Every reference must be found and read by you.
- Confident errors about local facts. Models are weak on recent South African policy, statistics and institutional rules, which change frequently. A model may describe last year’s bursary criteria or an outdated faculty rule as current.
- Generic argument. Generated text tends towards the average of what has been written on a topic. A literature review that sounds like everyone else’s is exactly what an examiner marks down for lack of synthesis.
- Loss of your voice and reasoning. Marks for a dissertation reward your demonstrated understanding. Text you did not reason through is text you cannot defend in a supervision meeting or an oral.
- Data protection. Pasting interview transcripts or participant data into an online tool may breach your ethics approval and your obligations to participants.
What South African universities actually say: two examples
Rules differ between universities, faculties and even departments, so the only rules that apply to you are your own. Two published documents show the direction of travel.
UCT Faculty of Commerce (December 2025)
UCT’s Faculty of Commerce framework for AI in research dissertations applies to all honours, master’s and PhD dissertations in the faculty. Its key standards include:
- Dissertations must reflect the student’s own intellectual work; AI use is only permitted where explicitly authorised and must be acknowledged.
- Spelling and grammar software is permissible. Any broader use must be specified in departmental guidelines, and if no such guidelines exist, students should assume broader AI use is not permitted.
- Supervisors are not authorised to independently permit AI use beyond the framework; the departmental guidelines are authoritative.
- No marks are awarded for AI-generated content as such; marks are based on the student’s own demonstrated understanding, reasoning and argumentation.
- Every dissertation must include a front-matter disclosure of AI use stating the tool, version and purpose, and students must keep a record of prompts and outputs.
- AI detection tools are described as unreliable and may not be used as the sole evidence of misconduct; concerns are to be investigated through draft reviews, AI records, supervision records or oral explanations.
For the literature review stage, the framework lists as permitted with disclosure, if authorised: idea-generation prompts, search-term brainstorming, drafting reading plans and generating outlines, titles or summary labels as starting points. It lists as not permitted for credit inserting AI-generated summaries or citations directly into the dissertation.

Stellenbosch University (postgraduate declaration)
Stellenbosch’s internal student declaration for the use of AI tools serves as a record and discussion tool between students and supervisors. It states that any use of generative AI tools in research submissions must be explicitly declared. Students tick whether they used generative AI only for basic language assistance or for listed purposes, including idea generation, literature search, methods design, data analysis, theoretical development, code development, visualisation, editing, writing assistance and citation formatting, and then state the tool, purpose and extent and justification of use. The supervisor confirms that a discussion of the declaration took place.
The common thread is clear: declare, keep records, and treat your department’s rule as the ceiling. For a wider overview of institutional positions, see our guide on whether AI is allowed for dissertations at South African universities.
A safe way to work with generative AI on your dissertation
- Find your rule first. Read your department’s guidelines and course outline. If nothing is written, ask your supervisor in writing and, under a framework like UCT Commerce’s, assume only spelling and grammar help is allowed.
- Keep an AI log from day one. Date, tool and version, prompt, output and what you did with it. It takes two minutes per session and is exactly the record that UCT Commerce requires and that protects you if questions arise.
- Use AI for process, not product. Where permitted, brainstorming search terms or a reading plan is far lower risk than generating sentences you submit.
- Verify everything. Every fact, figure and source suggested by a tool must be checked against the original. Never cite a source you have not opened.
- Write in your own words. Even where some assistance is allowed, the argument, analysis and conclusions must be visibly yours.
- Protect participant data. Do not upload transcripts, survey data or identifying information unless your ethics approval explicitly covers the tool.
- Disclose accurately. Complete your faculty’s declaration or front-matter statement honestly. Our guide to citing ChatGPT and other AI tools in Harvard and APA covers the referencing side.
Generative AI by dissertation stage
| Stage | Lower-risk uses (only if your rules allow, with disclosure) | High-risk uses to avoid |
|---|---|---|
| Topic and proposal | Brainstorming angles to discuss with your supervisor | Submitting a generated problem statement or research questions |
| Literature review | Search-term ideas; a reading plan | Generated summaries or citations inserted into the chapter |
| Methodology | Checking your understanding of a method, then confirming in a textbook | Generated justifications of your design |
| Analysis | Help understanding software error messages, if permitted | Uploading participant data; generated interpretation of findings |
| Writing up | Spelling and grammar tools | Generating or rewriting paragraphs you submit as your own |
Faculty rules decide which column each use falls into for you. Departments may be stricter than this table, never looser than their own published guidelines.

Why detection is not the point
Many students focus on whether an AI detector will flag their work. UCT Commerce’s framework describes such tools as unreliable and says they may not be the sole evidence of misconduct; it also records that UCT discontinued the Turnitin AI score function from 1 October 2025. The investigation route it describes, through draft reviews, AI records, supervision records and oral explanations, means the real test is whether you can show and explain how your work developed. Keeping your own drafts, notes and AI log is therefore more useful than worrying about a score.
Writing an AI disclosure statement (illustrative)
If your faculty requires a front-matter disclosure, write it plainly and specifically. A vague statement such as “AI was used in places” invites questions; a specific one answers them. An illustrative example for a student whose department permits limited use:
Declaration of AI use. I used a generative AI chat assistant (name and version) during the literature search stage to brainstorm alternative search terms for “teacher attrition” and “rural schools”, and spelling and grammar software throughout. No AI-generated text, summaries or citations appear in this dissertation. All sources were located through the library catalogue and databases and read in full. A log of prompts and outputs is available on request.
Adapt the wording to your faculty’s template, which may ask for specific headings such as tool, version, purpose and extent. If you used no generative AI beyond spelling and grammar help, say that too: a clear “none” is also a disclosure.
Questions to ask your supervisor about generative AI
Because rules are often set at department level and supervisors may not be allowed to go beyond them, have this conversation early and keep a written record of the answers:
- Which departmental guideline applies to my dissertation, and where is it published?
- Which specific uses are permitted for my project, and at which stages?
- What form of disclosure does the faculty require, and where in the dissertation does it go?
- What process evidence should I keep (drafts, prompts, outputs, verification notes)?
- Does my ethics approval allow any tool to process participant data, and if not, what tools may I use for transcription and analysis?
Stellenbosch’s declaration builds this discussion into a form both of you sign, which is a useful model even if your university does not require one. A shared understanding at the start prevents disputes at examination.
How students are actually using AI
Use of generative AI among South African students has grown quickly. Our data overview of how many South African students use AI summarises what surveys show, and our AI-safe workflow for an MBA research report applies the principles above to a specific degree. The pattern in both is the same: students who use AI for process and keep records have far fewer problems than those who use it to produce text.
Prompts that keep you in control
If your department permits limited use for process tasks, the way you phrase a request matters. Prompts that ask the tool to do your thinking produce text you cannot defend; prompts that support your own thinking keep the work yours. Compare:
- Instead of “Write my literature review on teacher attrition”, try “List alternative terms researchers use for teacher attrition so I can search databases more completely.”
- Instead of “Summarise these five articles”, read them yourself and ask “What questions should I ask myself when comparing the methods of several survey studies?”
- Instead of “Fix my discussion chapter”, use spelling and grammar software and ask your supervisor or writing centre for feedback on the argument.
Record each prompt in your AI log, whatever you ask.
FAQ
Can I use generative AI for my dissertation in South Africa?
It depends on your faculty and department. UCT Commerce, for example, permits spelling and grammar software and only broader use that departmental guidelines authorise, always with disclosure.
Do I have to declare generative AI use?
Under the documents reviewed here, yes. Stellenbosch requires explicit declaration of generative AI use in research submissions, and UCT Commerce requires a front-matter disclosure stating tool, version and purpose.
Can I use AI to find references?
Only as a starting point for searching, if permitted. Never cite a source an AI tool suggested without finding and reading it yourself, because tools can invent or distort references.
Will an AI detector flag my dissertation?
UCT Commerce describes detection tools as unreliable and not sufficient evidence on their own. Keep drafts and an AI log so you can show how your work developed.
Can I paste interview transcripts into a chatbot?
Not unless your ethics approval explicitly covers it. Participant data is protected, and online tools may store what you upload.
Is Tesify a generative AI writing tool?
Tesify helps you structure your dissertation chapter by chapter; it does not write your dissertation, and the text stays 100% written by you.
Structure your dissertation with Tesify
The safest dissertation is one where every chapter’s purpose is clear and every sentence is yours. 50,000+ students have written 230,000+ chapters with Tesify, which helps you structure your dissertation chapter by chapter, and every word stays 100% written by you. Structure your dissertation with Tesify.
Sources cited
- Stellenbosch University (2025) Internal Student Declaration for the Use of Artificial Intelligence (AI) Tools. Available at: files.su.ac.za.
- University of Cape Town, Faculty of Commerce (2025) Framework for the Use of Artificial Intelligence in Research Dissertations. Available at: commerce.uct.ac.za.
