South African content creators are using generative AI routinely but disclosing it rarely — and the reason, according to a new UNESCO-backed study, is that they believe admitting to it damages their audience’s trust.
The research, conducted by the Centre for Analytics and Behavioural Change as part of UNESCO’s global Behind the Screens project, surveyed 48 creators and interviewed 11 working across multiple platforms and genres. It is a small, qualitative sample, and its percentages should be read accordingly — each respondent accounts for roughly two percentage points.
Within that limitation, the findings are consistent and pointed.
The disclosure gap
Nearly a quarter of creators use generative AI regularly and another third occasionally — around 58% in total — primarily for writing captions, generating ideas, editing and research.
Only 33% consistently tell audiences when AI has been used.
Interviews explained why. Creators worry that admitting AI use would damage perceptions of authenticity and weaken audience trust. Many argued AI should speed up creative work rather than replace original thinking, and that audiences still want authentic personalities rather than AI-generated content.
That is a commercial disincentive to disclose, and it sits directly against government policy. Minister in The Presidency Khumbudzo Ntshavheni announced in May that South Africa plans mandatory disclosure of AI-generated content on digital platforms, alongside a public fact-checking platform. This study suggests two-thirds of creators are not disclosing voluntarily and have a reason not to.
It is the same tension Spotify encountered this month when it introduced AI Persona labelling for musicians — a voluntary scheme whose success depends on artists volunteering information they may believe reduces their appeal.
Confidence outrunning practice
Creators rated their own media and information literacy highly, and more than half believed they actively encourage critical thinking among followers.
Their reported practices differ. Some 36% do not consistently verify information before sharing it. Thirty-four percent treat likes and views as a signal of credibility. Only 30% prioritise supporting evidence and documentation when judging whether something is true.
Researchers describe the gap between self-assessment and practice as significant — particularly in an environment where South Africans were rated the most propagandised people in the world in the Ipsos Perils of Perception Index 2026, a figure Ntshavheni cited in Parliament.
Algorithms as editors
Interviewees described recommendation systems, trending sounds, hashtags and engagement metrics as powerful editorial forces shaping what they post and how often. Some admitted producing content designed for algorithms rather than their own values, because visibility depends on platform performance.
Creators also reported frustration with opaque moderation, saying content is removed, demonetised or restricted without explanation. Algorithmic recommendations received the lowest satisfaction rating of any platform feature measured.
Widespread abuse
Three-quarters of respondents said they had experienced or witnessed technology-facilitated gender-based violence, including cyber harassment, impersonation, reputational attacks and image-based abuse. Four in ten never reported these incidents, lacking confidence that reporting mechanisms would help.
Forty-three percent reported personally experiencing identity-based harassment, and 72% had experienced or witnessed xenophobia online. More than a third reported real-world violence connected to online attacks.
Researchers identify xenophobia as a distinctive feature of South Africa’s online environment, reflecting broader social tensions around migration and inequality.
No infrastructure behind the influence
The study found almost no professional support structure for creators. Only 10.4% had received media and information literacy training. Half had no journalism background. Fewer than one in five belonged to a relevant professional association — and existing bodies largely serve journalists or marketers. The South African Content Creators Union was established only in late 2025.
“What struck us most wasn’t that creators lack good intentions,” said Jesse Cann, research director at the CABC. “They want to be trusted, they want to do right by their audiences. But we’re asking people to shoulder the same responsibilities as journalists without any of the training, legal grounding or support structures journalists have built up over decades. That gap is where the real risk sits. Not with the creators, but with the ecosystem around them.”
The appetite for support is there. Nearly 88% said ethical guidelines would be valuable and 87% wanted training on ethics and freedom of expression. Almost 30% said they want to monetise their content but do not know how.
Creators are also not primarily commercial. More than a third said their main motivation is sharing knowledge or expressing opinions on issues they care about; about one in five named making money as the main goal.
The report recommends creator-focused media literacy programmes, practical AI ethics guidance, peer learning networks, stronger online safety support and clearer education on legal rights and responsibilities.
Download the full report here.
