When you're negotiating salary and want AI-backed data
The Move
Use AI to gather comp data, but strip your identity from the prompt. Always cross-reference and use the higher number as your floor.
The Script
"Based on market data from [source], the 75th-percentile range for this role in [city] is [$X–$Y]. I'd like to discuss how my [specific experience/results] positions me at the upper end of that range."
Your AI Prompt
You are a compensation analyst. Using only public market data (BLS, Levels.fyi, Payscale, Glassdoor) — not assumptions about the candidate — give the 75th-percentile salary range for [role] in [city] with [X] years' experience. Show your sources.
⚠️ Always verify the salary number against the actual sources cited — LLMs sometimes hallucinate data points.
Why This Works
Sorokovikova et al. (GeBNLP 2025) tested four commercial LLMs and found pronounced bias in salary advice, with gaps most severe in law, medicine, and business. Nghiem et al. (PLOS ONE 2024) concluded organizations should be 'deeply suspicious' of integrating AI into HR. Bowles, Babcock & Lai (2007) showed women face social penalties for direct negotiation. Anchoring to public market data bypasses both the AI bias and the social penalty. Always run the prompt twice: once with your name, once with a male name.
Research Behind This Move
Gender Bias in LLM-Generated Salary Recommendations
Sorokovikova, Chizhov, Eremenko & Yamshchikov · GeBNLP 2025 / ACL (arXiv:2506.10491) · 2025
preprintAsking an AI for Salary Negotiation Advice Is a Matter of Concern
Nghiem et al. · PLOS ONE (PMC11805401) · 2024
Social Incentives for Gender Differences in Negotiation
Bowles, Babcock & Lai · Organizational Behavior and Human Decision Processes · 2007