AI doesn't understand ‘Hlala Kahle’ - and that's a problem
Yet, marketing isn’t a set-it-and-forget-it function.
AI has genuine value when used correctly. It excels at processing datasets, generating initial concepts, setting up A/B tests, and handling repetitive tasks. It’s brilliant at removing grunt work and accelerating output – freeing skilled professionals from manual data entry.
The keyword there is “assistant.” AI speeds things up. It handles the heavy lifting so humans can focus on strategy, nuance, and decision-making.
But it’s still an emerging technology. No board would sign off on a business strategy developed entirely by ChatGPT, Deepseek, Grok, or Claude. So why do we let it happen with marketing?
The people-pleaser problem
One of the intrinsic problems is that large language models like Claude are trained to give you exactly what you think you want. If you ask Claude whether your campaign makes sense in a way that signals you want validation, Claude will tell you it’s brilliant. It’s the equivalent of asking a question that begs for a “yes” instead of digging for the logical gaps.
This is the Eliza effect in action – named after an early natural language processing program from the 1960s. It describes how humans instinctively attribute intelligence, understanding, and empathy to computers, even when we know we’re talking to a machine.
We trust software with human names for purely psychological reasons, not based on empirical output.
That’s fine when you need a pep talk. It’s disastrous when you need constructive feedback on a campaign.
The feedback loop problem
This creates a dangerous cycle. Wrong input leads to wrong creative, measured against wrong success metrics. The AI mistakes what is wrong for what it thinks you want to be right.
The result is a campaign that looks good on paper. Click-throughs might be strong. But people are clicking out of curiosity, not intent. They’re not your target audience, and they’re not converting.
Throwing more AI at it isn’t the fix. No amount of prompting or tweaking will fix a campaign built on flawed foundations. You’re optimising for outcomes that were wrong from the start.
This is why human oversight matters from the beginning, not as a rescue mission after the damage is done.
Local is lekker
We’ve seen this all too often.
An advertisement featuring a family with their golden retriever gets shown exclusively to dog lovers. You’ve just excluded every cat owner, every household without pets, and every potential customer who doesn’t fit that narrow profile. The ad performs – just for the wrong audience.
And you eliminate the entire audience you did want now and for the rest of the campaign because AI has learnt, wrongly, that you are targeting people with yellow-coloured dogs. And we’ve seen this happen.
This problem amplifies in South Africa’s complex, multicultural market. Language, cultural nuance, and regional buying behaviour vary dramatically. AI doesn’t understand what it means when you tell someone to “Hlala Kahle” and it can’t grasp the warmth, familiarity, and trust embedded in that phrase.
It can’t understand cultural codes that distinguish different parts of the country.
AI can’t factor in ground-level reality unless someone trains it on how South Africans actually respond to certain formats or messages.
It can’t interpret nuances between Google Analytics, Meta Ads Manager, and your CRM data when they don’t sync. It can’t tell you why a technically successful campaign isn’t moving product.
AI can generate. It can, to some extent, predict. It can support. What it can’t do is understand context, nuance, or South African consumer behaviour – and that’s where performance is won or lost.
AI is strongest when paired with experienced marketers who know how to interrogate, refine, and challenge what the tools produce. It gives you speed, but people give you accuracy. Success comes from a hybrid model – not from replacing people with automation.
It’s not magic
Yes, AI has a place. Its place is not setting strategy, interpreting data, or running unsupervised campaigns. It’s a tool, not a strategist. It’s your assistant, not your replacement.
It can help with information to develop strategies; it can’t make them. And it can’t bring you revenue when it’s missing your target market completely.
The next time ChatGPT asks how you’re doing or what it can help with, feel free to vent or ask for a recipe. Just don’t tell it to create an advertisement targeting 10- to 15-year-old female South Africans to sell them flip flops ahead of summer.
That campaign will, literally, flop.
