The first AI tool many South Africans used seriously did not arrive as a boardroom strategy. It arrived as a forwarded link in a family WhatsApp group, followed by a voice note explaining that it can write an email if you ask nicely. This is not the adoption curve imagined in keynote presentations, but it may be the more honest one.
AI is spreading through the same channels that carry school notices, church announcements, funeral details and load-shedding jokes. A matric learner asks for help structuring an essay. A mechanic drafts a quote. A small online seller rewrites product descriptions. A grandmother asks whether the photo of a politician is fake. The use cases are modest and immediate, which is exactly why they stick.
The useful prompt is usually plain: tell it what role to play, give the context, ask for a format, then check the answer. 'Help me write a polite complaint about a delayed municipal account' is better than 'write letter'. 'Explain this school notice in simple English and list what I must do by Friday' is better than pasting and hoping.
AI will be judged less by novelty than by whether it saves data, time, embarrassment or a trip across town.
The risk is that confidence arrives faster than literacy. A tool that can summarise a legal letter can also invent a clause. A generated image can travel faster than a correction. A fluent answer can feel authoritative even when it is wrong. In a country already navigating scams, poor service and institutional mistrust, AI literacy needs to be practical rather than fashionable.
A family rule helps: use AI for drafts, summaries and ideas; verify anything involving money, medicine, law, politics, school policy or official documents. Check the original source, ask who benefits from the message, and be suspicious of urgency. If the answer cannot show where it came from, treat it as a confident stranger, not a final authority.
The question is not whether AI is coming. It is already in the group. The better question is who teaches the group how to use it well: when to trust it, when to verify it, when to refuse it, and when to send the old-fashioned voice note instead.
