Problem
Millions of people around the world earn a living writing articles, building websites, designing graphics, writing code. These jobs are posted on platforms where employers hire freelancers for specific tasks. By 2020, an estimated 8.5 million freelancers worldwide had found work this way. But when generative artificial intelligence (AI) tools arrived, nobody knew which jobs would be hit hardest or how quickly the damage would show up.
Science
We collected 1,388,711 job posts from a leading global freelancing platform, covering July 2021 to July 2023. We sorted those posts into clusters of similar jobs, grouping writing, engineering and software development together as automation-prone jobs, and data entry, audio and video work together as manual-intensive jobs. Used a method that compares how two similar groups change over time we measured what happened to each group after ChatGPT launched in November 2022 and after image-generating tools launched in mid-2022. We also tracked Google search volumes for terms such as "ChatGPT writing" to measure how aware the public was of AI's ability to replace each type of job.
Outcome
Within eight months of ChatGPT's launch, job posts for writing, software development and engineering fell by 21% more than job posts for manual-intensive work. Image-generating AI produced a separate 17% drop in graphic design and 3D modelling posts. The jobs that fell most were the ones people searched for most in connection with ChatGPT. Competition for jobs in automation-prone categories rose by 9% and employers asked for more complex work at slightly higher budgets.
Change
We now have the first direct evidence, measured from the employer side, that generative AI is reducing demand for certain freelance jobs in real time. Writing and coding jobs are most exposed. Our research maps which tasks could be delegated to AI and which still need people. For policymakers, it signals where retraining and support are most urgently needed. In the long-term, AI may also raise productivity and create new kinds of work. But the short-term displacement is real, it is measurable and it is already here.
Ozge Demirci, Harvard Business School; Jonas Hannane, DIW Berlin and Technische Universität Berlin; Xinrong Zhu, ³Ô¹ÏºÚÁÏ Business School. Working paper, 2024.