Tuesday, 22 September 2026 South Africa About us
Independent South African journalism
The South African Enquirer Because enquiring minds want to know.
Send us a story Corrections
Business & Technology

AI tools flatten umgidi into a stock photo, Gibs lecturer warns marketers

Silas Matlala says campaigns built on global training data lose the Xhosa and Zulu texture that makes South African advertising land, and the waste shows up on the balance sheet.

AI tools flatten umgidi into a stock photo, Gibs lecturer warns marketers
The South African Enquirer · Illustration

A brand that hands its next campaign to an artificial intelligence tool can get back work that is grammatically flawless and culturally wrong, and a marketing lecturer at the Gordon Institute of Business Science says the cost of that mistake is measured in wasted budget and damaged reputation.

Silas Matlala, who teaches in the marketing faculty at the Johannesburg business school, made the case in a discussion about how AI is reshaping the way companies take their products to market. His argument is narrow and practical. The technology is fast, it is cheap and it is useful. It is not a substitute for knowing the audience.

Umgidi is not a coming of age party

Matlala’s clearest example is umgidi, the Xhosa rite that marks a young man’s passage into adulthood. Ask a model for imagery around South African black culture and coming of age and the result, he says, will carry none of the meaning that umgidi actually holds. The Xhosa practice has its own weight and its own rules. Umabo and umwabo, the Zulu equivalents, are different again. A bar mitzvah is a different thing again.

All of them involve a young person crossing a threshold. None of them can be swapped for another, and a model trained mostly on material from outside the continent has no reason to know the difference.

“The nuances that go with umgidi are lost,” Matlala said.

The same problem shows up in imagery. A brief asking for a picture of a burial in a South African township will produce something that misses the occasion entirely, he said, because the way a funeral is conducted here is layered in ways the training data does not carry. He added that the way black South Africans in this country mark a death differs from the way black communities in Nigeria mark one, and that the gap between the two is a subject on its own.

Stereotypes arrive at scale

Matlala was blunt about what the underlying data contains. Material about Africa that sits online leans heavily on poverty and wildlife, and a system that learns from that material will reproduce it. The result is stereotyping, prejudgment and misrepresentation, three failures he named in sequence.

“The creative just loses impact. It just becomes wallpaper,” he said.

That is the commercial consequence, and it is the part a marketing director can take to a board. A campaign that nobody remembers is money spent for nothing. Matlala put the risk in those terms: the budget, the resources and the time that went into the work are simply gone, and the reputational exposure that follows is harder to price. Information now moves in what he called a nanosecond, and a brand that has offended its own customers does not get to choose how long the story runs.

There is a softer failure too. A message that is mispronounced or misread will not offend anybody. It will simply be ignored, and the audience will move on.

The insight travels, the execution does not

Matlala’s answer is not to abandon the tools. It is to be clear about what they are for. Routine business writing, internal emails and straightforward translation are fair game. Representing a community, a township, a suburb or the mood of the country is not.

He used a Coca-Cola campaign about the correct pronunciation of people’s names to make the point. The insight underneath it is universal, he said, because taking the trouble to say somebody’s name properly is a mark of respect wherever you are. The execution was not universal at all. The British version leaned on humour. The South American versions did something else. The South African version was unmistakably South African.

That is the distinction he keeps returning to. The human truth travels. The way it is expressed does not, and no model can be asked to reproduce a register it has never lived in.

He offered two phrases as proof. The first is gees, the word South Africans use for the spirit of a place, a crowd or a moment. It carries a layer of meaning that a literal rendering into English flattens out. The second is the line that something is not DNA, it is South African, an inside remark that lands here and would land nowhere else.

His final example was a Vaseline Petroleum Jelly campaign from Unilever, Umtanaka Gogo. Put that name into a language model and it will offer back a literal English rendering. The literal rendering is not the campaign. What the name carries does not survive the trip.

What the brands should do

Matlala’s prescription is not complicated. Keep the tools, keep the speed, and keep a human being with a South African ear between the output and the audience. The efficiency is real. The judgment is still the marketer’s job.

The warning is worth taking seriously in a market where advertising budgets are tight and the pressure to automate is rising. The brands that get this right will be the ones that treat cultural knowledge as a skill they pay for rather than a step they can skip. The ones that get it wrong will find out on social media, which is the most expensive place to learn.

Source: Moneyweb, AI marketing campaigns risk missing cultural nuances and authenticity

Topics artificial intelligencemarketinggibsadvertisingculturesouth african business
Corrections. If something here is wrong, tell us and we will fix it and say so. Write to .