OpenAI Research: How Workers Use ChatGPT for Compensation Insights

OpenAI has released a research report detailing how workers in the United States use ChatGPT to bridge the wage information gap. US workers send an average of nearly 3 million messages per day to ChatGPT asking about wages, compensation, or earnings to synthesize scattered salary data and establish benchmarks for career decisions.

Labor Market Information Gap and AI Synthesis

Wage information is often difficult to find and interpret, particularly for those early in their careers, switching fields, or relocating. OpenAI reports that AI serves as a labor-market resource by synthesizing wage information and providing benchmarks in seconds, replacing the need for workers to search multiple websites or ask socially risky questions.

Patterns of User Inquiry

Analysis of wage-benchmarking messages—conducted via privacy-preserving automated classifiers—shows that users primarily seek help in two areas: translating pay into usable benchmarks and understanding potential earnings for specific roles, companies, or business ideas.

Among labeled wage-benchmarking messages, the distribution of queries is as follows:

  • Pay calculation: 26%
  • Specific role: 19%
  • Labeled role at a company: 11%
  • Occupation or career questions: 11%
  • Entrepreneurship: 18%

High-Demand Sectors for Wage Insights

Demand for AI-driven wage insights is strongest in fields where pay is less transparent, more negotiable, or critical to career mobility. Wage searches over-index in higher-skill occupations, including:

  • Creative fields (arts, design, entertainment, sports, and media)
  • Management
  • Healthcare
  • Computer and mathematical roles
  • Transportation, sales, business, and financial operations

Entrepreneurship-related questions are specifically concentrated in creative work and small service businesses, sectors where posted wage benchmarks are typically unavailable.

Economic Implications of Information Access

Workers seek pay information most frequently when wages are higher and more dispersed, indicating that the answer "matters more" and the stakes for accuracy are higher. OpenAI notes that misunderstanding potential earnings can lead to several negative outcomes:

  • Retention in lower-paying jobs
  • Reduced negotiating power
  • Delayed career transitions
  • Discouragement of investment in education and training

Technical Evaluation via WorkerBench

To measure the model's utility in labor market tasks, OpenAI introduced WorkerBench, a new evaluation framework. In the initial benchmark, GPT-5.4 was evaluated against 2024 Occupational Employment and Wage Statistics (OEWS) median wages at both national occupation and metro levels.

According to the report, GPT-5.4 demonstrated high accuracy, high coverage, and small bias, with nearly all numeric estimates falling very close to the OEWS benchmark.

Future Directions

OpenAI aims to continue improving the reliability of compensation insights by moving beyond national benchmarks toward more granular data, including specific geography, firm-level data, and the complex compensation questions workers encounter daily.

Sources