Equipping workers with insights about compensation
Key Points
- ~3M wage-related messages/day (US)
- WorkerBench: GPT‑5.4 closely matches 2024 OEWS medians
- Queries cluster in creative, management, healthcare, and tech
Summary
OpenAI research (published Mar 17, 2026) finds workers are using ChatGPT to close the wage-information gap: U.S. users send nearly 3 million wage-related messages per day to get pay benchmarks and realistic earnings estimates. Two primary user needs are identified: translating pay into usable benchmarks and estimating what a role, company, or business idea might pay. A privacy-preserving analysis using automated classifiers labeled message types and found concentrated demand in creative fields, management, healthcare, transportation, sales, and technical roles. The report introduces WorkerBench, which evaluated GPT‑5.4 against 2024 OEWS median wages at national and metro levels; the model shows high coverage, low bias, and numeric estimates close to benchmarks.
Key Points
- Volume: ~3 million U.S. wage-related ChatGPT messages per day on average.
- Labeled wage-benchmarking breakdown: pay calculation 26%, specific role 19%, entrepreneurship 18%, specific role at a company 11%, occupation/career 11%.
- Concentration: queries over-index in higher-skill or less-transparent occupations (creative, management, healthcare, computer/mathematical), and entrepreneurship questions cluster in creative and small service businesses.
- Method: privacy-preserving automated classifiers were used; no human reviewed individual messages.
- WorkerBench result: GPT‑5.4 vs 2024 OEWS medians — high coverage, small bias, and most numeric estimates closely match official benchmarks.
- Engineering implications: instrument and classify wage queries, prioritize geographic/firm-level and role-specific benchmarks, maintain privacy-preserving telemetry, monitor bias/coverage metrics, and extend evaluation datasets for firm- and job-level granularity.
Practical next steps for engineers
- Add telemetry to detect and route wage-related requests to specialized pipelines or models.
- Integrate additional geographic and firm-level salary datasets to improve relevance.
- Build evaluation suites (like WorkerBench) to track coverage, bias, and numeric error against public benchmarks (OEWS and other sources).
- Ensure analytics remain privacy-preserving (automated classifiers, aggregate metrics) when measuring usage and model behavior.