Direct answer: There is no national rule. Each South African university sets its own policy on AI in dissertations, and the published policies genuinely disagree. UCT stopped using AI detection tools in 2025 and focuses on process; Wits permits disclosed, augmentative use; Stellenbosch requires a declaration; Unisa pairs a “balanced” stance on AI with zero tolerance for misconduct. Your institution’s own document is the only one that governs you.
Why is there no single South African answer?
Because AI use in postgraduate research is regulated at institutional level, not by a national statute or a single sector rule. The Department of Higher Education and Training does not prescribe how universities must treat generative AI in a master’s dissertation or doctoral thesis. What exists instead is a set of university-level frameworks, policies and guidelines, each endorsed through that university’s own governance structures. That is why a student at UCT and a student at Unisa can read confident, generic advice online — “AI is allowed if you reference it” or “AI is banned” — and both will find that the advice does not match their own institution’s document. The generic answer circulating on content farms is wrong in both directions.
What does UCT’s policy actually say?
The University of Cape Town’s AI in Education Framework was endorsed by Senate’s Teaching and Learning Committee in June 2025, after being drafted by the Online Education Subcommittee from September 2024. It rests on three pillars: promoting AI literacies for staff and students, promoting and ensuring assessment integrity, and exploring AI-enabled opportunities for innovation in teaching and learning.
The most consequential decision for postgraduate writers: UCT discontinued the use of AI detection tools, including Turnitin’s AI Score feature, with effect from 1 October 2025. Deputy Vice-Chancellor Brandon Collier-Reed put the reason plainly: “AI detection tools are widely considered to be unreliable, and can produce both false positives and false negatives.” UCT’s Sukaina Walji added: “We’re focusing instead on assessing the process of learning not just the product.” In place of detection, UCT points to oral examinations, observed work, process-oriented evaluation and assignments that require students to disclose and critically reflect on their AI use. The guidance to students is to follow lecturer and supervisor rules on permitted use, remain critical of AI output, and act with integrity.
What does Wits say?
The University of the Witwatersrand publishes an institutional AI policy built on six principles, including upholding integrity and personal accountability, fostering AI literacy, prioritising human oversight and augmentative use, and equitable AI practices. Three sentences matter most for dissertation writers:
- “AI should be positioned and used as an augmentative and consultative tool that supports human intellect, not as a substitute for it.”
- “The use of AI must be transparently disclosed and appropriately acknowledged in line with university policy.” Wits expects researchers to describe AI use “in dedicated sections in the final write up of the study as a matter of course.”
- “AI cannot be an author on any work.” And the tool never shifts the blame: “The use of a tool does not absolve the user of responsibility for any academic misconduct, including but not limited to plagiarism, falsification, fabrication or improper attribution.”
In short: permitted, disclosed, augmentative — with you carrying full responsibility for every word.
What does Stellenbosch require?
Stellenbosch University’s library guidance frames allowable AI use around four principles: authenticity (AI may assist “where relevant”, but cannot “complete the assignment on your behalf”), accountability (you are responsible for what you create), fairness (use must be ethical and comply with academic integrity standards) and transparency — students “should clearly and honestly declare the use of AI tools and their outputs as well as the extent of the use”. The university’s guidance is blunt about the line that may not be crossed: “Do not copy word-for-word or claim any AI-generated content as your own work. This is plagiarism.” Stellenbosch’s named documents include its interim guidelines on allowable AI use and academic integrity and a position statement on the ethical use of AI in research and teaching-learning-assessment.
What about Unisa?
Unisa’s public position, set out in a media statement of 5 July 2024, combines two things that students often read as a contradiction but are not: a “balanced and reasonable approach to the use of AI tools like ChatGPT by students”, and “a zero-tolerance approach against any form of academic misconduct and dishonesty”. At the time of that statement, Unisa said it was finalising its AI guidelines and policy, introducing a compulsory Academic Integrity course from 2025, and it explicitly promoted “Turnitin, a web-based text matching software that has an AI content detector” among its integrity measures. Note the vintage: that statement predates UCT’s 2025 decision to abandon AI detection, and it describes a direction of travel, so check Unisa’s current policy pages before relying on it.
The one comparison that matters
On the single most anxiety-inducing question — will my work be run through an AI detector? — the two largest names in South African higher education point in opposite directions. UCT formally stopped using AI detection from 1 October 2025 because the tools produce false positives and false negatives. Unisa, in its most recent public statement on the matter, promoted a detector as part of its toolkit. Neither position is irrational; they reflect different institutional contexts. But it means the risk profile of identical behaviour differs by institution, and it is why you must read your own university’s current document rather than any summary — including this one.
How should you use AI safely in your dissertation?
Across all four institutions the published documents converge on the same safe zone: AI as a disclosed assistant for structure, feedback and refinement — never as the author. A tool like Tesify is built for exactly that zone: it helps you plan chapters, tighten your argument and manage references while the writing remains 100% your own, which is the standard every South African policy quoted above demands. If your proposal is still taking shape, start with our guide to writing a research proposal for a South African university, and remember that your methodology and data plans also pass through ethics clearance, where undisclosed AI processing of participant data can sink an application.
How do you actually write an AI-use declaration?
Wits expects AI use to be described “in dedicated sections in the final write up of the study as a matter of course”, and Stellenbosch requires you to declare the tools, their outputs and the extent of use. In practice, a defensible declaration answers four questions, each in a sentence or two: which tool you used (name and version, because capabilities change), for what task (brainstorming chapter structure, language editing, reference formatting, summarising your own notes), on what material (your own drafts only, or published literature — never raw participant data without ethics approval), and what remained yours (the argument, the analysis, the final wording). A declaration written this way does two jobs at once: it satisfies the transparency requirement, and it demonstrates to your examiner that you understood where the intellectual work sat. The students who get into trouble are almost never the ones who declared too much.
Keep a use log as you go — date, tool, task — the same way you keep a fieldwork diary. Reconstructing your AI use honestly at submission time, months after the fact, is nearly impossible, and a vague declaration reads worse than a precise one. If your faculty provides a declaration template, use it verbatim; if it does not, put the four answers above in your preface or methods chapter and tell your supervisor you have done so.
The data question nobody reads until it bites
The sharpest edge in the Wits policy is not about writing at all: sensitive or personal data must not be entered into public AI systems, because doing so may compromise research participants or intellectual property. Think about what that covers in a typical South African master’s project — interview transcripts with identifiable participants, employee survey responses collected under a confidentiality undertaking, unpublished company financials obtained under an NDA. Pasting any of that into a public chatbot to “summarise the themes” is not a writing shortcut; it is a disclosure of confidential data to a third party, and it can breach both your ethics clearance conditions and POPIA obligations regardless of what your university’s AI policy says about writing assistance. The safe pattern is to keep AI tools on your side of the data boundary: your own words, your own drafts, your own structure — not your participants’ words.
FAQ: AI and dissertations at South African universities
Is using AI for a dissertation illegal in South Africa?
No. There is no law against it. The rules that bind you are your university’s policy and your faculty’s assessment rules, plus ordinary academic integrity standards.
Do I have to declare AI use in my dissertation?
At Wits and Stellenbosch, yes — disclosure is an explicit written requirement. At UCT, disclosure and reflection are built into assessment design. Whatever your institution, declaring use is the safe default.
Can my university detect AI-generated text?
Detection tools exist but are unreliable — that is UCT’s own stated reason for discontinuing them, citing false positives and false negatives. Unreliability cuts both ways: honest students get flagged and dishonest ones pass. Do not build your writing process around beating or trusting a detector.
Can AI be listed as a co-author of my thesis?
No. Wits states it directly: “AI cannot be an author on any work.” Authorship implies accountability, which software cannot carry.
What happens if I submit AI-written work as my own?
Under every policy quoted here, presenting AI-generated content as your own writing is misconduct — Stellenbosch’s guidance calls it plagiarism outright. Penalties run through each university’s disciplinary process.
Does my supervisor decide what AI use is allowed?
Partly. Institutional policy sets the frame, but faculties and supervisors set assessment-level rules. UCT’s guidance tells students to follow lecturer rules on permitted use; Unisa tells research students to agree AI use with their supervisor. Ask before you use, and get the answer in writing — and if your supervisor has gone quiet on you, our escalation guide for an unresponsive supervisor covers how to get that answer without burning the relationship.
Is it safe to put my research data into an AI tool?
Be careful. Wits warns that sensitive or personal data should not be entered into public AI systems, as this may compromise research participants or intellectual property. If your data involves human participants, POPIA and your ethics clearance conditions apply.
Can I use AI to fix my references?
Formatting help is squarely inside the “augmentative tool” zone in the published policies, and reference accuracy remains your responsibility. Tesify’s bibliography tools format Harvard and APA entries while you keep control of the sources.
Where do I find my university’s current AI policy?
UCT: the AI in Education Framework via CILT. Wits: the university’s AI policy page. Stellenbosch: the library’s AI and academic integrity guides. Unisa: the policy register and current student communications. Policies are moving fast — check the date on anything you rely on.