Original title: The Pulse Beneath the Job Title: Monthly Readings of Requirements and Tasks from 750 Million Chinese Job Ads
Qin Chen, Ying Fang, Xiangyu Wang, Leo Yang Yang · August 2026
Job titles hide what jobs actually involve. Two people called accountants can have completely different qualifications and daily duties, but official databases lump them together into slow averages. Meanwhile, typical online research mixes up hiring rules with daily work tasks. The researchers solved this by separating the two ideas. They analyzed 752.6 million Chinese job ads to build two catalogs: one tracking what applicants must have, and one tracking what workers actually do. They also scored how easily a language model could take over each task, updating their view every month. Looking beneath titles revealed sharp shifts. Official registries show a single accountant job, but the ads reveal a clear staircase where higher certificates link to higher wages. Furthermore, while counting job titles suggests that jobs exposed to language models are vanishing, counting individual tasks shows that far less of this work is actually disappearing.
Imagine you're looking at a menu that lists 'chef's special' for every dish. You don't know if it's a salad or a steak, or what ingredients you need. In this paper, the job title is like that menu item, and the ads provide the actual recipe—the specific skills and tasks required. By reading millions of recipes, they show that the same menu item can require very different ingredients depending on the restaurant.
For job seekers, this means knowing exactly what certificate or skill is needed for a specific wage level, not just a generic title. For policymakers, it reveals which tasks are disappearing due to automation, month by month. This allows targeted training and policy responses instead of relying on outdated occupational directories.
The abstract does not specify the accuracy of the extraction or validation. It only covers Chinese job ads from 2022-2026, so findings may not generalize to other countries or time periods. The language model absorption score is a single measure that may not capture all aspects of automation risk.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
The Pulse Beneath the Job Title: Monthly Readings of Requirements and Tasks from 750 Million Chinese Job Ads — Qin Chen, Ying Fang, Xiangyu Wang et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.