Workplace disputes account for 37% of all queries put to My AI Lawyer, the South African AI legal chatbot, according to platform data covering the first six months of 2026 — by a wide margin the largest category, and a window into which legal problems South Africans cannot afford to take to a lawyer.
Criminal law came second at 15%, with users asking about arrests, police procedures and assault. Financial and debt matters accounted for 12%, while family law and property and rental disputes each made up roughly 10%.
Within the workplace category, queries centred on salaries and wages, employment contracts, unfair dismissals and CCMA processes.
The gap it sits in
The platform, developed by Legal Interact with Legal&Tax and available on the web and via WhatsApp, targets a specific and well-documented constituency.
“The sharp growth in the queries we process on our platform reflects a growing need in the South African market for solutions that make legal advice accessible to people who can’t afford a lawyer yet don’t qualify for legal aid,” said Michael Visser, head of services at My AI Lawyer.
That middle band — above the legal aid means test, below the cost of an attorney — is where most South African legal problems go unresolved. Justice Minister Mmamoloko Kubayi identified the same barriers at the University of Johannesburg’s AI and the Law conference in July: prohibitive legal costs, geographic remoteness from courts, lengthy delays and a shortage of practitioners willing or able to serve lower-income communities.
Visser’s reading of the workplace data is that the imbalance is informational rather than legal. “The questions we see show that employers hold a lot of the power in the relationship with their employees, often because they have access to legal resources,” he said. “Even when the law is on the employee’s side, many people do not fully understand their rights in terms of job security, leave or disciplinary processes.”
He added the converse: “It is not unusual for an employee to feel mistreated, even when the employer’s actions are completely legal.”
Finances underlie much of the rest, directly or indirectly. Visser cited someone recently retrenched who then queries how to negotiate rental terms with a landlord or handle debt counselling — one job loss generating queries across three categories.
The argument this is testing
My AI Lawyer is the practical instance of a debate iAfrica has covered in the abstract all year.
Kubayi’s central question in July was whether the pursuit of access to justice through AI risks the opposite. “Are we not, in our quest to increase access to justice through AI, at the risk of increasing access to injustice?” she asked, warning that systems trained largely on foreign datasets carry assumptions that do not fit South African realities.
UN Under-Secretary-General Professor Tshilidzi Marwala put the governance side at the same conference: “If we let the companies write the rules for the very tools meant to close the justice gaps, we risk the same digital divide simply reappearing inside the solution.”
The platform has answers to some of that. Its materials state it is trained on South African law by South African lawyers, operates as a closed system so user questions are not used to train external models, processes data within Microsoft Azure under POPIA, and escalates complex matters to qualified human advisors. It was the first African project awarded Microsoft’s Global Inclusion Changemaker award.
The question left open
What the data release does not address is language.
UJ vice-chancellor Professor Letlhokwa Mpedi framed the test precisely at the same conference: “An AI-driven legal aid tool is not neutral if it cannot reason competently in isiZulu or Sesotho.”
Nine of South Africa’s eleven official languages qualify as low-resource for language modelling, on UCT research. A person navigating a CCMA process or an arrest is unlikely to be doing so in their second language by choice. The company’s own materials reference multilingual capability; the usage data says nothing about which languages queries arrive in, or how performance varies across them. That is the number worth asking for.
The accuracy question sits alongside it. South African courts have already dealt with AI-generated fabrications in legal filings — an advocate was referred to the Legal Practice Council in 2025 after an AI tool cited case law that did not exist. Insurer Santam argued last month that AI hallucination is a foreseeable engineering risk rather than an unforeseeable failure, with consequences for professional liability.
What Visser claims for it
He is explicit that the platform does not replace a qualified representative. Its function is to supply local legal context to questions people would otherwise not ask anyone.
“Most legal questions on our platform never become formal legal cases, but a conversation about what the law says and what the options are can make all the difference,” he said. “The right advice helps people act earlier, make better decisions and approach difficult situations with greater certainty.”
On the query distribution, that framing holds up. Someone who understands their position before a disciplinary hearing is in a different situation from someone who does not — regardless of whether a lawyer is ever involved.
