How to write a fair performance review with AI

Give the AI the manager's notes and the role expectations, and require an example for every point. You get a structured review that separates opinions with no evidence, so the manager must support them or drop them.

Reviewed by AICourseHubPro · Last updated

Who this is for

  • Managers who write reviews at the last minute
  • HR partners who help managers prepare reviews
  • Small teams that want every employee judged by the same standard

What you need before you start

  • The manager's notes for the review period, with names removed
  • The role expectations or goals
  • An AI tool your organization has approved for HR material

Worked example: Draft a six-month review from notes that mix facts with opinions

The input

Act as an HR partner helping a manager write a fair review.
Context: Notes about the employee's last six months (fictional example, no names): Resolved an average of 52 tickets a week against a goal of 45. Wrote 6 help-centre articles in March after a spike in password-reset tickets; those tickets fell from 38 to 21 a month. Customer satisfaction score 91% against a goal of 90%. Missed 3 of 6 weekly stand-ups in April without notice. Helped train two new hires in May. Has an attitude. Not a team player. Role expectations: resolve 45 tickets a week, keep satisfaction at or above 90%, attend team stand-ups, document recurring issues.
Task: Draft a review with four sections: Strengths, Results against goals, Areas to develop, and Goals for the next six months.
Format: For every point, include a specific example from the notes. If a note is an opinion without an example, list it separately under "Needs evidence before use".
Limits: Describe behavior and results, not personality. Use the same standard you would apply to anyone in the role.

The output

Strengths
- Documents recurring issues: wrote 6 help-centre articles in March after a spike in password-reset tickets; those tickets fell from 38 to 21 a month.
- Supports colleagues: helped train two new hires in May.

Results against goals
- Tickets: averaged 52 a week against a goal of 45.
- Customer satisfaction: 91% against a goal of 90%.
- Team stand-ups: missed 3 of 6 weekly stand-ups in April without notice, against the expectation to attend.

Areas to develop
- Attendance at weekly stand-ups, or giving notice when unable to attend.

Goals for the next six months
- Keep ticket volume and satisfaction at or above current levels.
- Attend stand-ups, or give notice in advance.
- Continue documenting recurring issues.

Needs evidence before use
- "Has an attitude": no example given.
- "Not a team player": no example given. (The notes show the employee helped train two new hires, so this needs specific examples before it can be used.)

What to check or change by hand

  • Check every example and figure with the manager and the records
  • Make sure the opinions listed under "Needs evidence" are either supported with specific behavior or dropped
  • Check the wording describes actions and results, not character
  • Ask yourself whether you would be comfortable if the employee read it

Free template

Performance review prompt and checklist (Word) (DOCX), free with your email address.

Cautions

  • Never paste real names or sensitive details into an AI tool unless your organization has approved that use.
  • The manager owns the review. Read every line and change anything that is not accurate.
  • Check your local employment rules and your own policy before a review is shared with the employee.

Frequently asked questions

Why does the review separate opinions from evidence?

An opinion such as "has an attitude" gives the employee nothing to act on and can be unfair. The prompt flags it so the manager adds a specific example or drops it.

Can AI write the whole review?

It drafts the structure from the manager's notes. The manager remains responsible for the content and must check every line.

How does this help fairness?

The same structure and the same evidence rule apply to every employee, so reviews are judged by one standard.

All guides · All courses