Ready-to-use AI prompts for performance review writing — written for HR Manager and easy to paste into ChatGPT, Claude or Gemini.
Turn evidence into a performance review
You are an HR partner helping a manager write a fair performance review. Employee: [employee]. Review period: [period]. Goals set: [goals set]. Evidence I have: [evidence]. Overall assessment: [overall assessment].
Write the review using Situation-Behavior-Impact for every point. Ground each statement in a specific example from my evidence — never a personality trait. Cover: results against goals, how they worked with others, growth since last period, the two highest-impact development areas, and goals for next period that are specific and measurable.
Keep the tone direct and respectful, consistent with the rating (no 'exceeds expectations' language attached to a 'meets' rating). Flag any claim my evidence does not actually support.
Prepare for a difficult review conversation
Act as an experienced people leader coaching me. I need to tell [employee] that [difficult message]. Evidence: [evidence]. Their likely reaction: [likely reaction]. Their strengths: [strengths].
Give me: an opening that delivers the message clearly in the first 30 seconds, the evidence framed as impact rather than judgment, a path forward with specific and observable criteria, and the boundary between explaining and negotiating.
Then script responses to the three hardest things they might say ('this is the first I'm hearing of it', 'so-and-so does the same thing', an emotional reaction). Note what I must document and what I should not say. Keep it under one page.
Calibrate reviews across a team
You are facilitating review calibration. Here are draft ratings and justifications for a team:
<reviews>
[reviews]
</reviews>
Our rating definitions: [rating definitions].
Check for: ratings whose justification does not match the definition, inconsistent standards across people at the same level, recency bias (evidence clustered in the last month), justifications built on personality rather than outcomes, and language patterns that differ systematically between groups.
Output a table: person label, issue found, what to fix. Then list the questions I should put to managers in the calibration session, ordered by how much the answer could change a rating. Note where I would need more evidence to judge.