Johannesburg – Artificial intelligence is quietly reshaping how journalists across southern Africa work, with editors in South Africa and Zimbabwe increasingly using it for transcription, headline writing, translation and content preparation, even as they insist that editorial judgement must remain in human hands.
That is the central finding of a study by University of Johannesburg professor Mandla J. Radebe and post-doctoral fellow Mbongeni J. Msimanga, who interviewed senior editors across the region to examine how AI is influencing newsroom production processes, ethical standards and job security.
The researchers found that while AI is improving efficiency and, in some cases, quality, editors are not prepared to hand over editorial control to machines. 
Senior editors described faster turnaround times, quicker transcription, easier summarisation and speedier headline generation as among the clearest benefits. In Zimbabwe, some newsrooms have gone further, deploying AI-powered presenters to read weather bulletins and assist with news delivery. South African newsrooms have been more cautious, using AI mainly in editing, reporting support and headline optimisation, with full article generation remaining limited because editors insist on rigorous human verification.
For now, AI is functioning as a newsroom assistant rather than a replacement.
The push toward AI adoption is being driven in part by structural pressure. Print circulation across the region has been falling, advertising revenue remains fragile and newsroom staffing has shrunk. South Africa saw newspaper circulation drop by 17.3% in 2024, with several major titles cutting back operations or moving to digital-first models. Journalists are being asked to produce more content, across more platforms, at greater speed.
But AI brings its own complications. Editors flagged risks including factual inaccuracies, bias embedded in training data and poor contextual understanding of African languages, local idioms, satire and politically nuanced situations. One case cited in the research involved South Africa’s government-developed AI strategy, which was found to contain several fictitious academic references, likely the result of AI hallucinations.
Zimbabwean editors noted that AI tools frequently draw from online sources without distinguishing between verified reporting and misinformation. South African editors raised concerns about plagiarism, weak attribution and unverifiable sourcing. This creates a practical paradox: AI speeds up writing but also creates additional work, as journalists must verify machine-generated content before it can be published.
The linguistic and cultural gap is a further concern. Most AI systems are built in the global north and trained on western datasets, leaving African languages severely underrepresented. Editors reported problems with the pronunciation of indigenous names and poor handling of local nuance, pointing to a need for greater investment in African-centred AI research and local language datasets.
On job security, editors offered a more measured view than the public debate typically allows. Most do not expect journalists to disappear but anticipate pressure on technical roles such as sub-editing and layout. In Zimbabwe, high subscription costs and infrastructure challenges are limiting how widely AI tools can even be adopted. In South Africa, current adoption levels are seen as too limited to drive major labour restructuring. Some editors noted that AI tools have actually freed up financial resources that were redirected toward hiring freelance journalists.
Governance remains a weak point. Zimbabwe’s Zimpapers group has introduced internal AI policies covering disclosure, verification and training. Many South African newsrooms have yet to put similar frameworks in place. The researchers argue that existing press codes provide a foundation but need to be adapted to address the new ethical, operational and transparency risks that AI introduces.
The study’s overall conclusion is direct: AI may assist the newsroom, but journalism must remain under human editorial control. Machines can generate text and process information at speed, but they cannot fully grasp political sensitivity, moral consequence or historical meaning.
For newsrooms in Eswatini and across the region navigating the same pressures of shrinking revenues and growing digital demands, the message from editors in neighbouring countries is clear. AI is a tool, not a replacement.
