How ChatGPT Dismantled Kenya’s Ghostwriting Industry in Two Years

Over the past two years, an entire line of work that supported tens of thousands of people in Nairobi has collapsed: writing term papers, essays, and theses for students at universities in the United States and Britain. A New York Times report on September 5 traces how ChatGPT has almost entirely erased the business model. The case is uncomfortable, because the trade was built on cheating. And it is instructive, because it is one of the first named labor markets that generative AI has visibly made collapse.

Key takeaways

• At its peak, at least 40,000 people in Nairobi worked as academic ghostwriters for foreign students.
• After the launch of ChatGPT in 2022, orders and prices fell; surviving writers now earn $500 to $800 a month instead of $900 to $1,200.
• One operator with 100 staff has shut down; adjacent work such as transcription and data annotation also shrank.
• Oxford researcher Mark Graham expects similar breaks wherever work shifts sharply.
• The trade was illegal contract work: some students handed over their university login details so ghostwriters could submit directly in the learning system.

An industry that officially did not exist

Kenya has a young population with strong English skills and, since 2016, a government that has deliberately promoted online gig work. About 80 percent of employment in the country is informal. Out of this mix grew a market in which Kenyan writers produced the coursework of Western students for pay, often for whole semesters at a time and sometimes using the clients’ login credentials. To the universities it was fraud; to thousands of Kenyans it was a solid income. At its peak, at least 40,000 people in Nairobi alone are said to have lived on it, with new cars, current smartphones, and reserved tables at the club.

The Bundi case

The New York Times tells the story through Teresios Bundi, 34, who moved to Nairobi in 2011 to study public health. On the side, he wrote papers in engineering, medicine, and computer science, more than 2,500 of them over twelve years, on some days three. Before ChatGPT, he charged $40 to $70 per paper. All told, that earned him roughly five times what a job in the health sector he had trained for would have paid. His business is now closed. Bundi works for a German development organization and helps young Kenyans gain a foothold in the very digital economy that once supported him.

What ChatGPT changed

The break came fast. After ChatGPT launched in late 2022, clients made a simple calculation, according to the Times: why pay a freelancer in Kenya when a free chatbot delivers the same work in seconds. Orders dropped, prices fell. Those still in the business now earn $500 to $800 a month, down from $900 to $1,200. One operator who employed 100 writers has closed. It is not only essays: transcription, meeting minutes, data annotation, and moderation work, which the same people used to stay afloat, are scarcer and worse paid.

Why the case points beyond Kenya

Oxford internet researcher Mark Graham frames the episode as a warning sign: such disruptions will not stay limited to Kenya but will appear wherever work shifts sharply. That is the real significance of the report. So far, most evidence on AI’s labor-market effect has pointed to a cautious conclusion: that AI on the job shifts tasks rather than cutting whole occupations. Kenya’s ghostwriters are a counterexample with numbers: a clearly bounded, counted trade that all but vanished within two years.

The parallel on the other side of the value chain comes from OpenAI itself. While paid writing collapsed in Nairobi, the company reports internally 3.1 AI-workdays per human workday in its own research. The same technology that speeds up knowledge work removes the livelihood behind it elsewhere.

Conclusion

You do not have to romanticize Kenya’s ghostwriting trade to learn something from its end. The market was morally questionable and legally fraught, but it was real, named, and counted. That is exactly why it works as an early indicator: when a task consists mostly of text, is clearly bounded, and is assigned over the internet, it can be replaced by a language model as soon as the model’s output is good enough and cheaper. For the Kenyans affected, that is already the present. For many office jobs in the West, it is a question that poses the same simple calculation.