How to Start Using AI to Write Performance Reviews: A Practical Playbook

Every review cycle, I watch the same thing happen. Managers stare at a blank form, trying to remember what an employee did eight months ago, and HR ends up chasing half-finished reviews for weeks. 

I have been there, refreshing a Google Sheet at 11 pm, hoping the last three managers finally submitted their comments.

AI has changed that equation, but only when I use it the right way. Used carelessly, it produces generic, corporate-speak reviews that employees can spot from a mile away. 

Used well, it turns a manager’s rough check-in notes into a strong first draft in minutes, leaving more time for the conversation that actually matters.

This guide walks you through exactly how to do that, step by step, prompt by prompt.

What Is a Performance Review

Here is why using AI in this process matters right now:

• Most HR teams are still stitching together Google Sheets, Excel, PDFs, and manual forms to pull this evaluation together, and it does not scale as the team grows.
• Managers spend hours searching for “the right words,” especially when dealing with very high performers or difficult conversations.
• AI can summarize feedback from multiple sources (manager, peers, self, 360 feedback) into a single coherent draft rather than five disconnected documents.
• Done properly, it standardizes review quality across managers without making every review sound the same.

How to Use AI to Write Performance Reviews (Step-by-Step)

Here is exactly how to use AI to write a performance review, starting with the one setup step that makes every review after it faster: building a reusable workspace instead of starting from a blank prompt each time. If you are writing reviews for an entire team or company, you do not want to rebuild your instructions from zero for every single person. 

Both Claude and ChatGPT let you build a persistent “Project,” a dedicated workspace that remembers your instructions and reference material across every conversation you have in it. Set it up once, and it works for any employee, on any team, at any level, from then on.

Step 1: Create the Project

In Claude, click “Projects” in the left sidebar, then “Create Project,” and give it a specific name like “Performance Reviews, Q3.”

In ChatGPT, create a Project the same way from the sidebar. Either way, this becomes the one home for all your review-writing conversations.

Step 2: Write the Project Instructions

This is the part that does the real work. Inside the project’s instructions field, define:

• Role: “You help me draft first-person, constructive performance reviews from manager notes.”
• Required structure: strengths, one growth area, two specific next steps.
• The non-negotiables: never invent facts not in the notes provided, flag inconsistent or biased language, keep tone constructive rather than harsh.
• Masking rule: remind yourself, every time, to strip employee names and client names before pasting notes in.

Write this once, and every review you draft in the project follows the same rules automatically, no matter who wrote the underlying notes.

Step 3: Upload Your Knowledge Base

Add reference material to the project’s knowledge section, so Claude or ChatGPT has real context to draw from instead of generic defaults.

• Your competency definitions or review rubric
• A few past reviews you consider strong examples of tone and depth
• Your company’s review template or format

This is what lets the same project handle an engineer’s review and a salesperson’s review without you re-explaining what “good” looks like each time.

Step 4: Test It With One Real Employee

Paste in masked notes for a single employee and see what comes back. Check it against the final review checklist below. If the tone or structure is off, go back and tighten the instructions rather than fixing it draft by draft.

Step 5: Reuse It for Every Review, Every Role

Once the instructions and knowledge base feel right, every new review is just a new conversation inside the same project. 

Paste in that employee’s masked notes, and the project already knows your format, your tone rules, and your non-negotiables, whether the employee is in engineering, sales, support, or HR. This is what actually saves time at scale; you are not rewriting your setup for every single person you review.

7 Best Practices for Turning Notes Into a Finished Review

Once your project is set up, here is the core workflow I walk every manager through inside it, from a blank page to a review that is ready to share.

1. Gather Your Inputs

Pull together everything that actually happened this cycle before you open any AI tool.

• Goal progress and OKR updates
• 1:1 and check-in notes
• Self-assessment
• Peer or 360-degree feedback

The more specific and dated this is, the stronger the draft. A prompt with no real notes behind it always produces generic, forgettable output, no matter how good the tool is.

2. Mask Identifying Data

Before pasting anything into a public AI tool, strip out what you would not want to leave the building.

• Employee name
• Client or project names
• Anything proprietary or confidential

Replace each one with a placeholder like “the employee” or “Project A.” This single habit removes most of the data privacy risk people worry about.

3. Write a Specific Prompt

Tell the AI the employee’s role and level, the review period, and paste in the real notes from point 1. Name the competencies or role-specific evidence you want it to weigh, not just “write a review.” A vague prompt gets you filler text, a specific one gets you a usable draft. Here are three templates you can copy, paste, and fill in for your own use case.

For a standard review:

For a high performer up for promotion:

For an underperforming employee:

4. Generate Two Drafts and Blend Them

Ask for one version that leans more direct and one that leans more developmental in tone, then combine the two. This keeps the final review from reading too harsh or too soft in either direction.

5. Fact-Check and Bias-Check

Read the draft against your original notes line by line.

• Confirm every name, metric, and date actually happened
• Watch for language applied inconsistently across genders or roles
• Flag anything that sounds plausible but was never in your notes

AI can confidently invent details that sound completely real. This check is not optional.

6. Edit Until It Sounds Like the Manager

Replace generic lines with real specifics: “communicates well” becomes “walked the finance team through the Q3 model changes in plain language.” Cut corporate phrases like “leverages synergies.” Read the final version aloud; if it does not sound like something the manager would actually say in a 1:1, rewrite it.

7. Run a Final Check Before It Reaches the Employee

The manager, not the AI tool, is accountable for what gets shared, so treat this as your last line of defense before anything goes out.

• Voice: Does it sound like the manager, or like a generic template?
• Bias: Any language, “abrasive,” “emotional,” “not a team player”, applied inconsistently across people?
• Facts: Does every name, metric, and date check out against the original notes?
• Tone: Is the same competency rated generously in one line and criticized in another without explanation?

Review Comments by Competency and Roles

Different competencies and roles require different evidence to support the prompt. Here is what I feed AI for each one, and what a solid output looks like.

Building an AI Usage Policy for Reviews

If more than a handful of managers are using AI for reviews, put it in writing rather than leaving it to individual judgment.

• Define What’s Allowed: Spell out exactly which tools are approved, general LLMs, purpose-built HR tools, or both, and what tasks AI can and cannot be used for in a review cycle.
• Set Data Rules: Require that employee names, client names, and proprietary details be masked before pasting anything into a public AI tool. Make this a hard rule, not a suggestion.
• Require Human Sign-Off: State explicitly that no AI-generated review reaches an employee without a manager review and edit. Put this in writing so it survives a change in leadership.
• Train Your Managers: Run a short session on what a good prompt looks like and what the human-in-the-loop checklist requires. Most bad AI reviews come from managers who were never shown how to prompt well, not from the technology itself.

Choosing the Right AI Tool: A Quick Buyer’s Guide

Not every AI tool is built for this job, and the right pick depends on your size and your existing systems.

Note: An AI-generated review is only as good as the data behind it, garbage in, garbage out. PeopleGoal‘s OKR software keeps goals updated weekly rather than setting them once and forgetting them; its 360 Feedback software pulls in structured input from managers, peers, and stakeholders rather than scattered emails.

On top of that, real-time analytics & reports give you a benchmarked view of performance, rather than a manual export. Feed AI clean, centralized data like this, and the first draft comes out dramatically stronger.

Turn AI Into a Better Performance Review Partner

I hope now you understand how to write a performance review using AI. Remember, AI will not replace the judgment a manager brings to a performance conversation, but it will save you the hours spent staring at a blank page trying to find the right words. 

Feed it real, specific, well-organized data, run every draft through the human-in-the-loop checklist, and you will consistently ship reviews that are faster to write and more useful to read.

If your reviews are only as strong as the data behind them, start there. PeopleGoal keeps goals, check-ins, and 360 feedback centralized and current, so whatever AI workflow you build on top of it has something real to work with. 

Get started free or book a demo to see it with your own team’s data.