ChatGPT for Performance Reviews: From Your Notes to a Fair, Specific Appraisal
Short answer: ChatGPT helps with performance reviews when it turns your own evidence, such as dated notes on what the employee did, into clear, specific appraisal text. It should not decide ratings, invent achievements or replace your judgement. Anonymise the notes, ask for comments tied to each piece of evidence, then check for bias and consistency with the rating.
Many Indian companies run appraisals once a year around April, after the financial year closes, and many add a mid-year check-in around October. That timing is the problem: by March, managers remember the last six weeks and a couple of big moments, and the review ends up vague or skewed by recency.
The workflow below fixes the input, not just the wording. It works for managers writing reviews and, with small changes, for employees writing a self-appraisal.
Start with an evidence log, not a blank prompt
An evidence log is a running record of specific things a team member did, each with a date and its effect. If you give AI nothing but 'write a review for a good performer', you get generic praise. If you give it twelve dated notes, you get a review the employee recognises.
A simple format that works is Situation, Behaviour, Impact (SBI): what was happening, what the person did, and what changed as a result. Two lines per entry are enough.
- Add an entry when something notable happens, good or bad, and at least once a month per person.
- Pull from one-on-one notes; if you record those, our guide to AI meeting notes and summaries shows how to keep them consistent.
- Include numbers where they exist: targets, turnaround times, error counts, customer feedback.
- Note the support you gave, so the review reflects the context.
The notes-to-review workflow, step by step
- Export the year's log for one person and replace their name with 'Employee A'. Remove salary, health, family and any other personal details.
- Paste your company's competency list or rating definitions so the AI uses your language, for example a 1–5 scale where 3 means 'meets expectations'.
- Ask for a draft organised by competency, with every sentence linked to a dated log entry, and 'insufficient evidence' wherever the log is thin.
- Decide the rating yourself before reading the draft, so the AI's tone does not pull you up or down.
- Edit the draft: cut adjectives, add the context only you know, and make sure each development point comes with a concrete next step.
- Run the bias checks in the next section.
- Read it once as the employee would. If anything would surprise them, you should have raised it earlier in the year.
Worked example: prompt, draft and what to fix
Say you manage a five-person customer support team at a Gurugram SaaS company. Three of your log entries for Employee A read: 'Jun 12: rewrote refund macro, ticket reopen rate on refunds fell over the next month'; 'Aug 3: missed SLA on two escalations during a holiday rush, flagged early'; 'Nov 20: trained two new joiners on the billing tool'.
The prompt: 'You are helping a manager draft an annual appraisal. Use only the evidence below. Organise under Quality, Ownership, Collaboration and Development areas. Each sentence must cite the date of the evidence it uses. Do not add achievements that are not listed. Use plain, specific language, no clichés, under 250 words. Where evidence is thin, write insufficient evidence.'
| What the draft says | Problem | Better version |
|---|---|---|
| 'Consistently goes above and beyond for customers' | Cliché, no evidence | 'Rewrote the refund macro in June; refund tickets reopened less often over the following month.' |
| 'Delivered exceptional results across all KPIs' | Invented; the log has no KPI data | Delete, or add the actual numbers from your dashboard |
| 'Struggled with time management' | Overstates one incident and ignores context | 'Missed SLA on two escalations in the August holiday rush; flagged the risk early, which helped us reassign.' |
| 'A natural leader' | Vague and personality-based | 'Trained two new joiners on the billing tool in November; ready to own onboarding for the next hire.' |
A typical first draft gets the structure right and then slips in ways worth knowing:
How do you check an AI-written review for bias?
Compare reviews across your team on the same few dimensions before you submit. AI can amplify the patterns already in your notes, and it has its own tendencies in word choice.
- Swap test: change Employee A to a different gender or name and regenerate. If the adjectives shift ('assertive' versus 'aggressive', 'caring' versus 'strong'), edit toward the neutral, behaviour-based version.
- Recency check: count how many cited entries fall in the last three months. If it is most of them, go back to the log.
- Length and warmth: put all your team's drafts side by side. Big differences in length or warmth should match real differences in performance, not how much you chat with someone.
- Rating match: a review that reads like a 4 attached to a rating of 2 confuses everyone, especially when ratings are normalised against a bell curve in calibration meetings.
- Development points: everyone should get at least one, including top performers, and each should be specific.
Can you use ChatGPT for a self-appraisal?
Yes, and it helps most with structure. List your achievements with numbers and dates, paste your company's self-appraisal questions, and ask the AI to organise your points under each question in your own voice. Then remove anything you cannot back up; overclaiming in a self-review is easy to spot in calibration.
A prompt to try: 'Here are my notes from April to March and the five self-appraisal questions. Draft answers under 120 words each, using only my notes, in first person, plain language. Flag any question where I have no evidence.' Check your employer's rules first; confidential client names, revenue figures and internal documents may not be allowed in external tools. Our guide on being open about using AI at work covers how to handle that conversation.
What should you never hand to AI in an appraisal?
- The rating itself, increment percentages and promotion decisions.
- Performance improvement plans and termination letters, which carry legal and HR-compliance risk. Draft these with HR and, where needed, a lawyer.
- Salary data, medical information, disciplinary records or anything about family circumstances.
- Comparisons between named employees.
- Any use your company's AI policy does not allow. If there is no policy yet, the guide to a company AI usage policy is a sensible place for HR to start.
Treat every AI draft as your words once you submit it; you are accountable for what it says.
Fitting this into the appraisal season
The mid-year check-in, often around October in Indian companies, is the best time to test this workflow. Run your half-year log through the same prompt, share the draft development points with the employee in your one-on-one, and agree what evidence would show progress by March. The annual review then has no surprises, and the AI draft in April builds on a conversation you have already had. A common mistake is skipping the mid-year step because no rating depends on it; that is exactly why it is the safest place to practise.
If your cycle closes in April, start the log in May and spend five minutes a fortnight on it. Writing the review then takes far less time, and the AI step becomes a formatting job rather than a memory test. HR teams rolling this out to managers can pair it with the prompts in our guide to ChatGPT for HR professionals.
For where performance reviews sit among other workplace AI tasks, read our hub on practical AI use at work. If you want guided practice in writing grounded prompts and checking output, the no-coding AI course is built around these habits.
Frequently asked questions
Is it okay for managers to use ChatGPT to write performance reviews?
It is generally fine to use it for drafting from your own evidence, if your company allows it and you remove personal details. The rating and judgement must stay yours, and you should edit the draft before submitting.
What are good performance review prompts for managers?
Prompts that give the AI your evidence log, your rating definitions and a strict rule to use only that evidence work best. Ask for each sentence to cite the date of the note it uses, and for 'insufficient evidence' where notes are thin.
Can employees tell if their appraisal was written by AI?
They can when it is generic. A review full of specific, dated examples reads as yours, because the substance came from you.
How do I write appraisal comments for an average performer with AI?
Give the AI balanced evidence and the definition of 'meets expectations' in your scale. Ask for what the person does reliably, one or two specific development areas and a concrete next step for each.
Should HR approve AI use in appraisals?
Ideally yes. A short written rule on which tools are allowed and what data must stay out protects managers and employees, and makes reviews more consistent across teams.
Keep reading in AI at work
- How to Use AI at Work: Practical Examples for Every Role in an Indian Office Start here
- ChatGPT for Excel Formulas: Worked Examples on GST, Salary and Sales Sheets
- ChatGPT for HR: Drafting JDs, Policies and Employee Letters Without the Usual Mistakes
- Company AI Usage Policy: A One-Page Starter Template for Indian Firms
- Corporate AI Training in Delhi NCR: What to Check Before You Book
Courses: Applied AI Professional · Corporate AI training · All guides in AI at work