How to Write Performance Reviews with AI (Without Sounding Robotic)
A step-by-step guide to writing performance reviews with AI: what to gather, the exact prompts to paste, what to check, and what never to hand to a chatbot.
For Managers, Engineering and Executives
How do you write a performance review?
Write a performance review by collecting evidence first, then drafting from it. Gather the person's goals, specific results, feedback from others, and examples of behavior across the whole review period. Organize them into what went well, what needs to improve, and what comes next. Write each point as a specific observation with an outcome, not a label. AI can speed up the drafting and tighten the language, but you supply the facts and the judgment.
TL;DR
- Start with evidence, not prose. A review is only as good as the notes behind it, so spend your time collecting examples before you write a sentence.
- Use AI to draft and edit, not to judge. It can turn your notes into clean paragraphs, but it cannot know who this person is or what happened.
- Specific beats polished. "Cut onboarding time from three weeks to two by rewriting the checklist" is better than "demonstrates strong ownership."
- Robotic writing comes from thin input. Give the AI real details and your own phrasing and the output sounds like you.
- Protect confidential data. Check your company's AI policy before pasting names, ratings, or health and pay details into any tool.
- Never let AI set the rating. The rating, the promotion case and any performance concern are your decisions, and you have to be able to defend them.
Why do AI-written reviews sound robotic?
They sound robotic because the prompt was thin. If you type "write a performance review for a strong marketing manager," the model fills the gaps with the most common phrases in its training: "consistently exceeds expectations," "a valuable team player," "demonstrates strong communication skills." Every employee gets the same sentences, and the person reading it can tell.
The fix is not a cleverer prompt. It is better raw material. When you give the AI three real projects, one real quote from a colleague, and your honest read on a weakness, it has something specific to work with. It will still reach for stock phrases, so you also need an editing pass. Both are covered below.
How do you write a performance review step by step?
1. Collect evidence for the whole period
Most review bias comes from memory. You remember the last six weeks and the biggest wins or failures. Build a running file instead. Pull from:
- The goals or objectives set at the start of the period, and how each turned out
- Project outcomes, with numbers where they exist (deadlines met, revenue, tickets closed, errors, customer scores)
- Notes from your one-on-ones
- Peer or stakeholder feedback, including anything solicited through your company's process
- Emails, messages or thank-yous that show behavior, not just output
- Misses, with the context: what happened, what the person did next
Thirty minutes of skimming your calendar, project tracker and sent mail will surface things you forgot.
2. Decide your overall view before you write
Know the headline first. Is this person exceeding, meeting or falling short of the role, and what is the one thing you want them to hear? If you cannot say it in two sentences, you are not ready to write. Do this part yourself. Your company's rating scale and calibration process also matter here, so check them.
3. Draft each section from your notes
A workable structure for most companies:
- Summary: two or three sentences on overall performance
- Key results: three to five accomplishments with outcomes
- Strengths: behaviors, with an example for each
- Areas to improve: one to three items, each with an example and a clear expectation
- Goals for next period: specific and measurable where possible
If your company uses its own form, map your notes onto its fields. The content is the same.
4. Use the "situation, action, result" test
For every claim, check that it names what the person did and what changed because of it. A quick way to write it:
- Weak: "Priya is a strong communicator."
- Better: "Priya ran the weekly vendor call and sent written recaps within a day. When the packaging supplier slipped in March, the recaps gave us a paper trail that helped us renegotiate the delivery date."
5. Write the improvement feedback carefully
Feedback on weaknesses should be about observable behavior, not character. "Missed three of five status updates in Q2, which left the design team waiting on decisions" is something a person can act on. "Needs to be more reliable" is not. State what good looks like next period.
6. Edit, check facts, and read it aloud
Confirm every date, number and name. Read it aloud. If a sentence sounds like something you would never say to the person's face, rewrite it.
7. Prepare for the conversation
The written review is half the job. Decide what you will say in the first minute, and leave time for the employee to respond. Nothing in the document should be a surprise. If it is, that is a signal you should have raised it earlier.
How do you use AI to write a performance review?
Use AI in three places: organizing your messy notes, drafting from them, and editing the result. Here is a prompt for each.
Prompt 1: Turn messy notes into a draft
Replace the bracketed parts. Use initials or a first name if your policy asks you to limit identifying information.
You are helping a manager write a performance review. Use ONLY the facts I give you. Do not invent accomplishments, numbers or quotes. If something is missing, write [NEED MORE DETAIL] instead of guessing.
Role: [job title], [level if relevant]
Review period: [dates]
Goals set at the start: [list]
My notes on what happened:
[paste bullets: projects, outcomes, numbers, feedback from others, misses]
My overall view: [meets / exceeds / falls short, plus the one thing I want them to hear]
Write the review with these sections: Summary (3 sentences), Key results, Strengths, Areas to improve, Goals for next period.
Style rules:
- Plain, direct language. No phrases like "consistently exceeds expectations" or "valuable team player."
- Every strength and every improvement area must include a specific example from my notes.
- Write as me, a manager speaking to a colleague I respect. Second person ("you") is fine.
- Keep it under 500 words.
The line "write [NEED MORE DETAIL] instead of guessing" matters. It stops the model from inventing a plausible accomplishment, which is the biggest risk in this task.
Prompt 2: Strip out the robot
After you have a draft, paste it back with this:
Here is my draft performance review. Rewrite it so it sounds like a thoughtful person wrote it. Replace generic phrases with specifics from the text. Flag every sentence that makes a claim without an example, and list them at the end so I can add detail. Do not add new facts.
[paste draft]
Prompt 3: Make hard feedback clearer and kinder
I need to give this feedback in a performance review. Rewrite it so it describes observable behavior and its impact, avoids judging character, and ends with what I expect going forward. Keep it honest. Do not soften it into vagueness.
My draft: [paste]
Prompt 4: Plan the conversation
Based on this review, list the three points I should cover in the first five minutes of the conversation, two questions to ask the employee, and the most likely thing they will push back on, with a suggested response.
[paste review]
What should you check in an AI-written review?
AI makes consistent, predictable mistakes in this task. Run this checklist before anything leaves your desk.
| Check | Why it matters |
|---|---|
| Every number, date and project name | Models fill gaps with plausible but invented detail |
| Every adjective and rating word | "Exceptional" or "needs improvement" may not match your rating |
| Tone toward the weaknesses | AI tends to soften or pad criticism until it disappears |
| Tone toward the strengths | AI tends to inflate praise into generic superlatives |
| Same phrases across your team | If three reviews share sentences, people will notice |
| Gendered or biased language | Words like "abrasive," "bossy" or "nurturing" can show up unprompted; read for them |
| Consistency with last cycle | A sudden change in tone or rating needs an explanation |
What should you never hand to AI?
- The rating or the promotion decision. The model does not know your calibration rules, your budget or the person. You do.
- Performance concerns with legal weight. If a review might support a warning, a performance plan or a termination, involve HR and write it with them.
- Sensitive personal information. Medical leave, family situations, immigration status and similar details do not belong in an AI prompt or, usually, in the review.
- Confidential data in tools your company has not approved. Many employers have rules on which AI tools are allowed and what can be pasted into them. Check yours. If you are unsure, remove names and identifying details, or ask.
- Feedback you have not already given. A review should confirm what the person has already heard from you.
What does a good review paragraph look like?
Here is the same point written two ways.
Generic AI output: "Marcus is a strong performer who consistently demonstrates ownership and delivers high-quality work. He is a valuable member of the team and communicates effectively with stakeholders."
With real notes: "Marcus took over the quarterly pricing report in April, after the analyst who built it left. He rebuilt it so it refreshes from the source data instead of being pasted together by hand, which cut prep from two days to half a day. Sales leadership now uses it in their Monday meeting. Next period, I'd like him to start flagging data problems earlier: the July report went out with a duplicated region, and we caught it only after finance asked."
The second version gives Marcus credit he can recognize, names one clear thing to improve, and could not be mistaken for anyone else's review.
How do you handle self-reviews and reviews of your manager?
The same method works. For a self-review, keep a "wins and evidence" list all year and paste it into Prompt 1 with the note "write in first person, confident but not boastful." For upward or peer reviews, give the AI your concrete examples and ask it to make the feedback specific and constructive, then remove anything you would not say in person.
Can you keep your notes in one place all year?
Yes, and it is the habit that improves reviews the most. A simple running document works: one line per event, with the date, the person, what happened and the result. Some managers build a small internal tool for this so feedback can be logged in a minute after a meeting.
Could you build your own review-notes tracker?
If you are tired of reconstructing the year from memory, a small tool that logs feedback, goals and wins per person all year, and assembles a summary when review season arrives, is a reasonable thing to own. Remy builds full-stack business apps from a plain-English description, with a database and sign-in included, so you could describe that tracker, with roles so only you see your team's notes, and have it deployed to a live URL. Check with HR and IT first about where employee data may be stored. Remy also has a 7-day free trial, and the platform price is on its pricing page.
Frequently Asked Questions
How long should a performance review be?
Long enough to cover results, strengths, one to three improvement areas and next goals, usually one to two pages. Follow your company's form if it has one. Length matters less than specificity: a short review full of examples beats a long one full of adjectives.
Is it okay to use ChatGPT or another AI tool to write performance reviews?
Often yes, but it depends on your employer's policy. Many companies restrict which tools can be used and what data can go into them. Use an approved tool, leave out sensitive details, and treat the output as a draft you must verify and own.
How do I write a review for someone I need to give negative feedback to?
Describe specific behaviors and their impact, say what you expect instead, and offer support. Make sure the person has heard it before the review. If the feedback could lead to formal action, bring in HR before you write it.
How do I avoid bias in performance reviews?
Use evidence from the full period, apply the same criteria to everyone on your team, and compare each review against the role's expectations rather than against other people's personalities. Read your draft for vague character words, and check whether similar behavior is described differently for different people.
Can AI write a whole review from a job title?
It can produce text, but not a good review. Without your facts it produces generic praise and may invent details. Give it real notes, tell it not to guess, and check every claim.
Topics
How we source, write, and correct our articles. Editorial standards
Related articles
First Time Manager Tips: What AI Can and Can't Do
How to Negotiate a Raise (and Use AI to Prep for It)
1 on 1 Meeting Template for Managers (Plus an AI Prep Prompt)
