How to Prompt AI to Write a Performance Review

AI can turn scattered notes into a clean performance review draft, but only if you feed it specifics first. Asking AI to write a review from just a job title and a rating produces the generic, hedge-everything language employees can't act on: “demonstrates strong communication skills,” “shows room for growth in prioritization.” The fix is one extra step: write down three to five concrete examples from the review period yourself, then let AI organize and phrase them. That is the short answer to how to prompt AI to write a performance review that actually says something. Below: the template, a before and after, and two things AI should never do in a review.

Why AI-drafted reviews usually say nothing

A model has no memory of what your employee actually did this quarter. Left to fill that gap on its own, it defaults to trait language and safe hedging, phrasing that is technically about the person but describes no actual event. It reads professionally. It also gives the employee nothing to point to and nothing to change.

This is the same generic-output problem covered in how to prompt AI without getting generic answers, and reviews are a bad place for it, because vague feedback is worse than no feedback. “Needs to improve stakeholder management” gives an employee nothing to change on Monday. “In the March vendor renewal, you presented pricing to finance without looping in procurement first, which delayed the signature a week” tells them exactly what to change.

The rule that fixes it: examples before adjectives

AI is genuinely good at organizing scattered notes into clear prose. It is bad at judgment, and has no access to your employee's actual work unless you put it in the prompt. So the order of operations has to be: you supply the raw material, AI handles structure and phrasing.

Before you open a chat window, write down three to five specific incidents from the review period. For each one, note:

• What happened, with a rough date or sprint if you have it

• What the person specifically did (their action, not a trait)

• What the outcome was, ideally something measurable or observable

This takes ten minutes and is the actual work of writing a review. Everything after this is AI doing formatting and phrasing, which is the part worth delegating.

An AI performance review template prompt that forces specificity

Copy this, fill in the bracketed fields with your own notes, and do not let AI draft anything before those fields are filled:

You are helping me draft a performance review. Use only the information
I give you below. Do not invent examples, metrics, dates, or quotes that
I have not provided.

EMPLOYEE: [name, role, team]
REVIEW PERIOD: [start date] to [end date]
OVERALL RATING OR LEVEL (if applicable): [rating]

STRENGTHS - specific examples from this period:
1. [what happened, what they did, what the outcome was]
2. [what happened, what they did, what the outcome was]
3. [what happened, what they did, what the outcome was]

GROWTH AREAS - specific examples from this period:
1. [what happened, what they did, what the outcome or impact was]
2. [what happened, what they did, what the outcome or impact was]

GOALS FOR NEXT PERIOD: [1-3 goals, as concrete as possible]

MY TONE: [e.g. direct but supportive, formal, matches our team's usual
review style]

Using ONLY the examples above, write a performance review with these
sections: Summary, Strengths, Areas for Growth, Goals for Next Period.
Reference the specific examples I gave you rather than restating them
as generic traits. Keep language plain and concrete. Do not soften
the growth areas into vague language, state them clearly and pair each
one with a specific next step.

Before and after: same employee, two inputs

Here is what the growth-areas section looks like with a vague input versus a specific one, same employee, same underlying issue.

The first version is not wrong exactly, but it is not useful either. It could describe almost anyone. The second came from the same prompt template, with the only difference being that the manager spent ten minutes writing down what actually happened first.

What to check before you send the draft

• Read every sentence against your own memory. If AI overstates or understates what happened, fix it before it becomes the official record.

• Check the tone matches how you'd actually say this to their face. AI defaults to a corporate register that can sound colder than your normal style.

• Confirm dates, numbers, and project names are exactly right. A wrong date on a real incident undermines the whole review.

Writing reviews for a whole team in one sitting, drafts can drift in tone from one employee to the next as the session runs long. How to get consistent AI output every time covers ways to keep phrasing stable across a batch, worth doing before your fourth or fifth review of the day.

Two things AI should never do in a performance review

Never let it invent an example or a metric that did not happen. Models fill gaps convincingly, and a fabricated example, however plausible, can end up in a formal HR record tied to compensation or a promotion case. Every specific claim in the draft has to trace back to something you actually wrote in the prompt, not something the model added to round out a paragraph.

Never use it to soften or avoid a hard conversation that needs to happen directly. AI writing tends to smooth edges by default, and it is tempting to let a difficult message get gentler with each redraft until it stops saying what it needs to say. If someone needs to hear that a pattern is serious, that message has to survive editing, and it still has to be delivered in a real conversation, not handed over as a document.

This same specificity-first approach works for other business documents too, the same principle behind a strong cold outreach email prompt: generic prompts produce generic output regardless of the document. For the broader framework behind why, see Prompt Engineering: A Framework for Prompts That Actually Work.

If your team is still figuring out where AI fits into their own workflow, how to get your team to actually use AI has practical starting points beyond reviews.

FAQ

Can AI write an entire performance review by itself?

Not a good one. AI can draft structure and phrasing, but it has no memory of what your employee did, so a review written without specific input defaults to generic trait language. Treat AI as a drafting layer on top of your own notes, not a replacement for them.

What information should I give AI before asking for a review draft?

Three to five concrete examples from the review period: what happened, what the person specifically did, and what the outcome was. Include rough dates or project names where you can. Vague input produces vague output.

Is it okay to use AI for negative performance feedback?

It can help you phrase a difficult point clearly, but it should not soften a message that needs to stay direct, and the conversation still has to happen face to face. A draft is not a substitute for delivering hard feedback yourself.

How do I stop AI performance reviews from sounding generic?

Feed it specific incidents instead of adjectives before you ask for a draft, and instruct it not to invent examples or restate your input as generic traits. The prompt template above is built around forcing that step.

Should I tell my employee I used AI to help write their review?

There is no universal rule, but most teams are fine with AI as a drafting tool as long as the manager verified every fact and the review reflects real observation, not fabricated detail. Check your company's AI usage policy if one exists.