POPIA questions sound simple until you have to name the actor. Processing conditions, operators, and responsible parties are easy for a chatbot to blur. The words sit next to each other in training data. They are not interchangeable in the Act.
A generic answer will often describe a privacy duty in the abstract, then attach it to the wrong party. It may import language from another country’s data protection statute because the shape of the question looks familiar. If you cannot open the POPIA text that supposedly supports the sentence, you do not have a source. You have a paraphrase of the internet.
NOMOS Legal AI should show the POPIA text it retrieved. Titles and portions belong next to the analysis, so you can click through and read the provision yourself. If retrieval is thin, it should say so. A short, honest miss is better than a fluent mix of operator and responsible-party language that nobody can check.
Pick South Africa in the country picker. Ask the question in the terms the Act uses, not only in business shorthand. If a condition or a definition is cited, open that portion. Do not let a generated summary stand in for the section.
NOMOS is research assistance, not legal advice. It will not tell you whether a particular processing activity is lawful on your facts. It can get you to the licensed text faster, and it can refuse when that text is not on the page. How sourcing and limits work is on Trust.