How to Write a Performance Review With AI
The review form asks for “overall performance,” and the cursor starts blinking. You remember good work, a missed deadline, and several conversations—but none of it is organized well enough to write fairly.
Make the judgment yourself. Give AI verified examples of work and results, then ask it to help organize and phrase the comments. Do not let it choose a rating, infer someone's attitude or potential, or invent evidence to make the review sound complete.
That rule applies whether you are a manager writing about an employee or an employee writing a self-review.
Separate the decision from the writing
AI can help with wording. It does not know the person, the full review period, your workplace policy, or the context behind every result.
Before using it, decide:
- Is this a manager review or a self-review?
- What period does the review cover?
- Which goals and responsibilities actually apply?
- What rating or assessment has the responsible person already decided?
- Which examples can be supported by records or direct observation?
- Which HR or company rules control the final wording?
If the rating is still undecided, stop there. Do not ask a general-purpose chatbot to evaluate the employee for you.
Build a small evidence bank
Write facts before adjectives.
| Weak note | Evidence the review needs |
|---|---|
| “Great communicator” | What was communicated, to whom, and what improved? |
| “Not proactive” | What observable task or follow-up was missed? What was expected? |
| “Exceeded expectations” | Which agreed goal was exceeded, and by how much? |
| “Needs more leadership” | Which situation required a decision, coordination, or follow-through? |
| “Worked hard” | What output, result, customer outcome, or team contribution shows it? |
For each usable example, note:
- the situation or responsibility;
- what the person did;
- the result;
- the date or review-period context;
- the goal, standard, or competency it relates to;
- a next step when development is needed.
Do not include protected personal information, health details, accommodations, complaints, compensation, disciplinary records, demographic information, or private colleague feedback in a consumer AI tool. Follow the employer's policy and use an approved system when required.
First ask AI to challenge your notes
Do not request polished review comments yet. Ask the AI to show where the evidence is thin.
I am preparing a performance review for [manager review / self-review] covering [dates].
Examine the notes below. Do not write the review yet.
For each note:
– identify whether it describes observable behavior and a result or only gives a judgment;
– flag any missing date, context, result, or supporting example;
– flag language about personality, attitude, motivation, or potential;
– identify any number or claim that needs verification;
– write
EVIDENCE NEEDEDinstead of inventing a detail.NOTES:
[paste cleaned, relevant notes]
SHRM's current guidance takes a similar approach: pressure-test the notes first, and do not let AI add examples or make the evaluation.
Then draft the comments
Once the evidence is complete, use the appropriate branch.
If you are writing about an employee
Help me draft performance-review comments from the verified evidence below.
The rating and assessment have already been decided: [state them]. Do not strengthen, weaken, or change them.
Connect each statement to a specific example. Describe behavior and outcomes, not personality. Make development feedback actionable. Do not infer motivation, attitude, potential, health, or personal circumstances.
If a statement is unsupported, flag it rather than completing it.
If you are writing your own review
Help me draft a first-person self-review from the verified evidence below.
Organize it into results, collaboration, problems solved, skills developed, and next-period goals. Keep the tone direct and specific. Do not exaggerate my role, turn team results into personal achievements, or invent metrics.
Flag any claim that needs a date, number, or example.
Paste only the minimum cleaned material needed for the draft.
Remove the phrases that sound like nobody
AI often makes performance reviews smooth and empty at the same time.
Replace lines such as:
- “consistently demonstrates exceptional dedication”;
- “leverages cross-functional synergies”;
- “continues to be a valuable asset”;
- “shows a proactive mindset”;
- “should focus on improving communication.”
With evidence:
Coordinated the revised launch schedule across support and sales, documented the new handoff, and reduced unanswered setup questions from 18 to 6 during the final two weeks.
Or, for development:
Two weekly updates arrived after the agreed Friday deadline. For the next quarter, send the update by 3 p.m. Friday or flag the delay to the project lead that morning.
The second versions are not necessarily kinder or harsher. They are clearer.
Check fairness before you submit
AI-assisted wording can still repeat a biased note or make subjective language sound official. The U.S. Equal Employment Opportunity Commission has specifically identified AI and algorithmic tools used in employment decisions as an area where discrimination risks still apply.
Review the draft for these problems:
- different standards being applied to similar roles;
- personality labels instead of observable work;
- assumptions about caregiving, health, age, disability, culture, or communication style;
- praise that credits one person for a team result;
- criticism based on something outside the review period;
- development goals the person cannot act on;
- a rating that is not supported by the written evidence.
AI is not a fairness check by itself. A human who knows the policy and context remains responsible for the review.
What happened in H5W's performance-review test
H5W tested the method with fictional notes that mixed good evidence with three common problems: a missing completion date, vague praise such as “always positive,” and an unsupported claim that support tickets fell by 30%.
ChatGPT separated the five notes before drafting. It labeled the missing date NEEDS DATE, the personality-style praise TOO VAGUE, and the unsupported percentage UNSUPPORTED. The final paragraph used the supported FAQ and workflow examples and kept a visible placeholder for the onboarding date instead of quietly filling it in.
It also calculated a 34.1% change from two supplied ticket counts. The arithmetic was correct, but a newly calculated percentage is still not the same as an approved performance metric. Check the calculation, measurement period, and review policy before adding a derived number to the final form.
Read it aloud once
Before submitting, read every paragraph aloud.
Ask:
- Could I explain the evidence behind this sentence?
- Would the person understand what went well or what must change?
- Did AI add a metric, date, example, or tone I did not choose?
- Does this sound like a real manager or employee—not a corporate template?
- Is any sentence better handled in a conversation than in a form?
Add required HR or compliance language yourself. Do not ask AI to recreate wording that must be exact.
Keep the conversation human
The review document is preparation for a conversation, not a substitute for one.
Use AI to spend less time fighting the blank page. Keep the assessment, examples, fairness, and final conversation in human hands.
For other workplace writing, see How to Write a Professional Email With AI. Use How to Check Whether an AI Answer Is Correct for factual claims, and review What Information Should You Never Share With AI? before entering workplace information.
Sources and verification
Information last verified: October 13, 2026.
- SHRM: How to Use AI to Draft Performance Review Comments
- SHRM: AI for HR Tool Evaluation Checklist
- EEOC: Artificial Intelligence and Algorithmic Fairness Initiative
- EEOC: Employment Discrimination and AI for Workers
- OpenAI: Data Usage for Consumer Services FAQ
Free printable resource
AI Business Writing Brief Pack
Use the practical version while you work. It is a selectable-text U.S. Letter PDF with no email gate.
Public H5W PDF with selectable text and permanent source links.
Keep going
Choose your next step
Return to Work for the full collection, or continue with a closely related guide.
