Can AI Guess What You Know? Performance Comparison of Large Language Models for Human Domain Knowledge Estimation From Communication Logs

In three linesComparative study of 7 LLMs (Gemini, Claude, GPT) to estimate professional expertise from Slack logs. On 27,188 messages from 43 users, Gemini 2.5 Flash achieves lowest error (MAE 21.13%). Accuracy depends only weakly on message volume.

Summary generated by Claude — human-verified