The Value-Added Tax Act 89 of 1991 (VAT Act) is legally complex due to its readability, which makes it challenging to teach, apply in practice, and administer. My PhD empirical investigation establishes that the lengthiest section in the VAT Act is section 17(2) (denied input, entertainment). The study further cites several sections in the VAT Act where passive voice is used and suggests that the Tax Administration Act (TAA) is a good measure of an Act that uses active voice. The VAT Act’s complexity compromises the VAT system’s objective. Even though only some researchers emphasise the impact of tax law complexity on compliance, it is well documented that it also affects the level of legal certainty. Thus, taxpayer compliance may be enhanced by tax simplification.
It is well established in the literature that tax legislation and other related materials are difficult to read and understand. Various forms of assessing these challenges (including readability measures) have been employed in many jurisdictions, but very little attention has been focused on South African tax material.
The South African VAT Act has been shown to be very complex and difficult to read and understand. This study, building on earlier research, provides evidence from applying traditional readability formulae to the lengthiest provision to comprehend section 17(2) (denied input, entertainment) to support this position.
This study takes this analysis an important step further, providing evidence of how the provision could be simplified. It does this by utilising Generative AI (Microsoft’s Copilot) to develop an easier to comprehend provision. Copilot’s suggestion for rewriting the provision was tested through seeking feedback from experienced academic teachers of VAT law in South Africa. An important observation from this feedback is that while using Generative AI is helpful, there remains a need for human overview to ensure the rewritten provision does not change the interpretation and application of the law. TaxGPT®, a specialised LLM, was contemplated; however, it was not accessible in South Africa at the time of publication of this article. TaxGPT® is considered an AI tax assistant that was created for accounting and tax firms or businesses to help them with complex tax queries, research, writing, and document analysis related to tax practice. TaxGPT® may be implemented in future research.
Background
This viewpoint is based on the full article that I, Professor Marina Bornman, and Professor Adrian Sawyer co-authored. Professor Bornman is an Associate Professor: Research Methodology for MCom and Programme Coordinator (MCom SA & International Taxation) at the University of Johannesburg. Professor Sawyer is a Professor at the University of Canterbury, Department of Accounting and Information Systems, UC Business School. The full article was published in the Journal of Australian Taxation and can be accessed at Journal of Australian Taxation 2025 Volume 27 No 1.
AUTHOR
Dr Muneer Hassan, CA(SA)
Chartered Tax Advisor (CTA); Deputy HOD Accountancy and Senior Lecturer Taxation at UJ; Lecturer on the Gauteng Board Course and Academic Member
of SAICA’s National Tax Committee
