Which Chinese Jobs Are Most Vulnerable to AI? A Data‑Driven Ranking

OpenAI researchers once claimed that about 80% of U.S. jobs will be affected by AI. To see how this translates to China, we analyzed billions of recruitment records from the past eight years, mapping Chinese occupations to the O*net taxonomy and decomposing them into 19,265 tasks and 23,534 work items.

Each task was evaluated with GPT‑4 (or GPT‑3.5‑turbo) using a prompt that asked the model to score, from 0 to 5, how much human labor could be reduced if a large language model assisted the task. Scoring 0 means no reduction, 5 means complete automation. The model processed 100 tasks per request, costing only $0.12 per batch with GPT‑4 (or $0.03 with GPT‑3.5), so labeling 40,000 items cost under $5, compared with at least ¥10,000 and a week of work for human annotators.

Aggregating the task‑level scores back to the occupation level yields a ranking of AI replacement likelihood for all Chinese jobs. The top 25 most replaceable occupations include translators, insurance underwriters, playwrights, visual designers, decorative artists, art editors, advertising designers, video editors, text editors, web editors, literary writers, journalists, call‑center agents, front‑desk staff, hosts, and secretaries. Surprisingly, even computer programmers appear at rank 25, with an estimated 75% of their work at risk.

The least vulnerable jobs are mainly blue‑collar manufacturing roles, as well as gardeners, cleaners, laundry workers, masseurs, nail technicians, and traditional pastry chefs—positions that typically require lower formal education and wages.

What characteristics make a job easy to replace?

In the original OpenAI study, higher‑paid U.S. occupations showed a positive correlation with AI replaceability, except for salaries above $100k where the trend reverses. In our Chinese data, however, there is no significant correlation between average salary and AI replacement probability.

Instead, we measured a "growth" factor: the annual wage increase associated with gaining one more year of experience, calculated by weighting year‑over‑year wage changes across eight recruitment cycles. This growth rate shows a strong positive correlation (p < 0.001) with AI replacement likelihood. Occupations with less than 8% annual growth are the least likely to be automated, while those with growth exceeding 20% per year have, on average, more than a 60% chance of being replaced by AI.

This pattern suggests that the skills and expertise accumulated through on‑the‑job experience—precisely the knowledge that drives wage growth—are the very aspects that large language models are beginning to supplant.

What exactly is being replaced?

High‑growth jobs often involve tasks that can be learned quickly and scaled, such as drafting documents, creating visual content, or providing customer service. While one might expect that higher‑paid, experience‑rich roles attract more AI investment, the evidence shows that AI can already outperform human practitioners in many of these experience‑based tasks.

Conversely, innate human abilities—those acquired early in life or through biological evolution—remain the hardest for AI to mimic.

In summary, AI is rapidly encroaching on occupations whose productivity improves sharply with experience, indicating that the knowledge and techniques honed over years are increasingly vulnerable to automation.