一封好的求职信以前帮助雇主将申请人的简历与职位要求联系起来，并作为申请人付出努力的信号，特别是对于那些就业经历与其职业目标不相关的工人。大型语言模型的兴起使求职者能够大规模地生成完美针对性的求职信，让任何人都能呈现为细心的申请者；Galdin 和 Silbert 使用 Freelancer.com 的数据来评估对劳动力市场的影响。

比较 ChatGPT 出现前后，两个量化变化突出。信件中位长度从 LLM 出现前的 79 字上升到出现后的 104 字，且该平台在 2023 年推出的 AI 工具产生的 AI 撰写申请的中位数为 159 字（比人工基线多于两倍）。研究者使用 AI 在九个类别上给每封信打分（每项 0–2，最高 18 分），中位总分从 LLM 前的 3.9 上升到出现后几乎翻倍。

这些变化抹去了优质提案的工资溢价：在 LLM 前，这样的提案每项任务额外带来约 26 美元，而该平台的任务中位酬劳为 100 美元；AI 出现后这一溢价消失。Galdin 和 Silbert 估计平台工资比没有 AI 求职信的世界低约 5%，招聘量低约 1.5%。雇主无法区分强弱候选人，于是降低薪酬并招入质量较低的工人，导致工人的损失大于企业的收益。

A good cover letter previously helped employers connect a candidate’s CV to job demands and signal applicant effort, particularly for workers whose employment histories were orthogonal to their ambitions. The rise of large language models enables jobseekers to produce perfectly targeted cover letters at scale, letting anyone present as a careful applicant; Galdin and Silbert analyze Freelancer.com data to assess labour-market effects.

Comparing pre- and post-ChatGPT activity, two quantitative changes stand out. Median letter length rose from 79 words pre-LLM to 104 words post-LLM, and the platform’s 2023 AI tool yields AI-written applications with median 159 words (more than twice the human-written baseline). Using AI to score letters across nine categories (0–2 each, maximum 18), the median overall score rose from 3.9 pre-LLM to nearly double post-LLM.

These shifts erased the wage premium for a well-written proposal: pre-LLM, such a proposal added about $26 per task on a platform where the median task paid $100; after AI the bump vanished. Galdin and Silbert estimate platform wages are ~5% lower and hiring ~1.5% lower than in a world without AI cover letters. Employers, unable to separate strong from weak candidates, cut pay and hire lower-quality workers, producing net losses larger for workers than the gains to firms.