Equipping workers with insights about compensation
OpenAI News工资信息影响人们的重要决策：申请哪类工作、是否谈薪，以及某条职业路径是否值得走。但与大多数商品的标价不同，劳动的“价格”往往难以获得且难以解读——尤其对职业初期、转行或搬迁的劳动者而言更是如此。
人工智能成为了一种新的劳动市场信息工具。与其让求职者在多个网站间搜寻、拼凑零散的薪资页或冒着社交风险去问人，不如由模型在几秒钟内综合工资信息并给出一个参考基准。美国的劳动者已经在以这种方式使用 ChatGPT：平均每天在 美国 发出近 300 万条关于工资、补偿或收入的询问。
我们最新的研究报告检视了美国人如何借助 ChatGPT 缩小薪资信息差距。人们使用它主要有两类需求：把零散的薪酬信息换算成可用的基准，以及判断某个职位、公司、职业路径或商业想法在现实中可能拿到多少报酬。在被标注为“工资基准”类的对话中，工资计算占 26%，具体职位占 19%，创业相关占 18%，公司内具体职位占 11%，职业或生涯问题占 11%。这些结论来自一种保护隐私的分析方法，使用自动分类器，并且没有任何人查看单条消息内容。
问题的分布很有意思。职业相关的薪资查询集中在艺术与设计、娱乐、体育和媒体；管理；医疗；交通；销售；以及商业与金融运营等领域。与总体就业比重相比，薪资搜索在需较高技能且薪酬透明度较低的职业中占比偏高，例如创意类、管理、医疗和计算机与数学类岗位，这表明人们最需要信息的地方，往往是工资最难以对标、谈判空间更大或对职业流动性更关键的岗位。创业类问题也呈现类似特征，集中在创意工作和小型服务业——这些地方常常没有公开的工资基准可供参考。
跨行业来看，薪资查询在薪酬更分散、整体更高的领域更为频繁。换句话说，当准确知道报酬更重要且薪酬更难判断时，劳动者更倾向于寻求信息。这不仅仅是查工资这么简单：对潜在收入的误判可能让人滞留在低薪岗位、削弱谈判筹码、推迟职业转变或阻碍对教育和培训的投入。更好的信息不能消除所有不确定性，但能帮助人们形成更合理的预期，从而做出更好的决定。
为更好地评估模型对劳动者的帮助，报告还推出了一个名为 WorkerBench 的评测项目，旨在衡量 ChatGPT 在对劳动者有实际价值的劳动市场任务上的表现。在首轮基准测试中，我们用 2024 年的 OEWS 全国与大都市职业中位工资对比评估了 GPT‑5.4 的表现。样本显示模型准确度很高：覆盖面广、偏差小，几乎所有数值估计都非常接近基准。
工资信息具有重要的经济意义，但往往难以或敏感以至难以获得。劳动者已经在利用 ChatGPT 来应对这一难题，尤其是在不确定性最大、利害最深的劳动市场领域。我们的目标是持续提升这类工具的实用性与可靠性——从国家层面的基准走向更贴近劳动者日常提问的地域、公司、岗位层级与具体薪酬问题。
Wage information shapes important decisions: what jobs people apply for, whether they negotiate, and whether a particular career path is worth pursuing. But unlike the price of most goods, the price of labor is often hard to find and difficult to interpret—especially for workers who are early in their careers, switching fields, or moving locations.
AI is a new type of labor-market resource. Rather than requiring a worker to search across multiple websites, interpret scattered salary pages, or ask a socially risky question, a model can synthesize wage information and return a benchmark in seconds. Workers are already using ChatGPT this way, sending nearly 3 million messages per day, on average in the US, asking about wages, compensation, or earnings.
Our latest research report looks into how Americans are using ChatGPT to close the wage information gap. They most often come to ChatGPT for two kinds of help: translating pay into a usable benchmark, and understanding what a role, company, career path, or business idea might realistically pay. Among labeled wage-benchmarking messages, pay calculation accounts for 26% of questions, followed by specific role (19%), entrepreneurship (18%), specific role at a company (11%), and occupation or career questions (11%). We determined this through a privacy-preserving analysis that uses automated classifiers and never involves a human viewing individual messages.
The pattern of those questions matters. Occupation-related wage searches are concentrated in fields like arts, design, entertainment, sports, and media; management; healthcare; transportation; sales; and business and financial operations. Relative to employment, wage search over-indexes in higher-skill and less transparent occupations such as creative fields, management, healthcare, and computer and mathematical roles, suggesting demand is strongest where pay is harder to benchmark, more negotiable, or more important to career mobility. We see a similar pattern in entrepreneurship-related questions, which are concentrated in creative work and small service businesses—areas where there often is no posted wage benchmark.
Across industries, wage search rises where pay is more dispersed and where wages are higher. In other words, workers seem to seek pay information most when getting the answer right matters more and when pay is harder to read. That is why this matters beyond wage lookup alone. Misunderstanding potential earnings can keep workers in lower-paying jobs, undercut negotiating power, delay career moves, or discourage investment in education and training. Better information can’t eliminate uncertainty, but it can make it easier to form a reasonable view of what work pays and therefore help people make better decisions.
To better understand how our models serve workers, the report also introduces WorkerBench, a new effort to evaluate ChatGPT on labor market tasks that are valuable to workers. In this first benchmark, we evaluated GPT‑5.4 against 2024 OEWS median wages at the national occupation and metro levels. In the observed sample, the model is highly accurate: coverage is high, bias is small, and almost all numeric estimates fall very close to the benchmark.
Pay information is economically important, but often difficult or sensitive to obtain. Workers are already using ChatGPT to navigate that problem, especially in the parts of the labor market where uncertainty is highest and the stakes are most meaningful. Our goal is to keep improving how useful and reliable that help can be—moving beyond national benchmarks toward the geography, firm, level, and compensation questions workers actually ask every day.
Generated by RSStT. The copyright belongs to the original author.