How to Write a Performance Review That Says Something
· 9 min read · Officeagent research
Try it on a real task
Routing slip · Officeagent
Status: Ready
Action requested
Pick a task above and press RUN IT. Officeagent handles it end to end; you approve the send.
Reading page /
☐ Approve · nothing sends without you
In the product you edit the draft right here; the agent learns your correction.
Officeagent drafted, you approved, it executed.
Sample data · Officeagent always waits for your approval before anything is sent
To write a performance review, work from dated notes rather than memory, read the employee's self-evaluation before you draft, and put a specific example under every rating you give. Name two or three strengths and no more than two development areas, each with the incident, its effect on other people, and what better looks like concretely. Then set three to five measurable goals for the next period.
The hard part is not the form. It is writing sentences that carry information. Most performance reviews are a page of statements that could describe anyone, and an employee cannot act on "great team player" or "needs to work on communication" any more than you could.
Before you write: the notes problem
Recency bias is the single biggest reason reviews are inaccurate. Without notes, a manager sits down in December and writes an assessment of October and November while calling it a year. The employee who had a difficult September and a strong spring gets a review that describes neither.
The fix costs ten minutes a month. Keep a running file per person, dated entries, three lines each, whenever something notable happens in either direction: a project that went well, a handoff that did not, a piece of feedback from someone else, a deadline that moved. When review week arrives you are summarizing a document instead of interrogating your memory, and the difference is visible in the first paragraph.
If you are starting this now with no notes at all, do the honest thing: write the review from what you can actually evidence, say less rather than inventing detail, and start the file today so next cycle is better.
Read the self-evaluation first
Send the self-evaluation a week ahead of the meeting, and read the answers before you write your own draft. Four open questions work better than a competency grid the employee fills in about themselves:
- What are you proudest of this period?
- What did not go the way you wanted?
- What do you want to be doing more of?
- What do you need from me that you are not getting?
Two things happen. Employees name their own development area more often than managers expect, which changes the conversation from adversarial to shared. And the second question frequently surfaces context you did not have, like a project that slipped because a dependency you approved arrived three weeks late.
Writing the review itself
Structure each section around evidence. Here is the difference in practice, using two versions of the same feedback.
| Vague version | Specific version |
|---|---|
| "Great team player, always willing to help." | "Three colleagues independently mentioned bringing work to her for review before sending it. That is a reliability signal you cannot self-report, and it saved the March client deck from going out with last quarter's numbers." |
| "Needs to work on communication." | "On the Q1 migration, the four-day slip was known internally on Tuesday and shared on Friday. Two people spent Wednesday and Thursday planning against a date that had already moved. For any project over two weeks: a three-line status every Friday, and flag a slipped date the day it slips." |
| "Needs to be more proactive." | "In the vendor renewal, I would have wanted you to bring me the two options and a recommendation rather than asking which one to pursue. Next time a decision like that lands, come with a proposed answer and I will tell you if I disagree." |
Every specific version does the same three things: names an incident with a time reference, states the actual effect on someone else's work, and describes the different behavior in terms the person could follow tomorrow. That is the whole technique. Our performance review template has the full form plus phrase banks organized by rating, if you want the structure to write into.
How many development areas?
One or two. Not five. A review listing five things to improve produces zero improvements, because nobody prioritizes on the employee's behalf and so nothing gets sustained attention. Pick the one that would most change their effectiveness and say it clearly, then leave the smaller items for one-to-ones.
Same logic on goals for the next period: three to five, each with what will be true, how it is measured, by when, and what support is needed from you. The last part is skipped constantly and it is the part that makes the goal fair. If the goal requires a budget or an introduction or your time, name it in the goal.
Ratings, and the calibration step everyone skips
Use a three-point or five-point scale labeled in words, and then actually use the range. The common failure is a five-point scale where nearly everyone is rated a four, which happens when ratings are tied directly to pay and managers avoid hard conversations by inflating.
Before ratings are final, sit the managers down together and compare: does a "meets expectations" from one manager mean the same as from another? Ten minutes per employee is enough on a small team. Without it, the lenient manager's team gets the raises, and everybody notices within about two cycles.
One related discipline: separate the pay conversation from the development conversation by a couple of weeks. When both happen in the same meeting, the employee hears the number and stops processing anything else, which is a rational response and it wastes the development half.
Documentation, and why inflated ratings come back
Reviews are records, and records get read at the worst possible moment. Write to facts: dated incidents, observable behavior, effects on work. Never to personality, and never to anything touching a protected characteristic.
The most common documentation problem in small companies is three years of "exceeds expectations" followed by a termination for performance. That contradiction is created by managers rating generously to avoid a difficult conversation, and it is the version that reads worst later. Rate honestly in the moment and the record takes care of itself. The review should also line up with what your employee handbook says about conduct and discipline, since a policy and a review that describe different standards are a problem in the same file. Our guide to what to include in an employee handbook covers the conduct and discipline sections a review has to stay consistent with.
Delivering it
Hand over the written review at the start of the conversation rather than days ahead. Sending it early turns the meeting into a defense of a document. If the review is long, give the person ten minutes to read it in the room, then talk.
Start with the goals from last period and the outcomes, because that is factual ground and it sets a shared basis. Then strengths, then the development area, then next period. Leave real silence after the development point; the instinct to fill it with softening is what turns clear feedback into something the employee later says they never heard.
The test of the whole thing is the surprise test. If anything in the review is news to the employee, the problem is the year, not the review. Monthly one-to-ones are what make an annual review boring in the right way, and boring is the goal.
Running the cycle without losing a month to it
The administrative weight is real for a small office: scheduling every pairing, sending self-evaluations a week ahead, chasing the people who have not submitted, collecting manager drafts, booking the conversations, filing signed forms, and setting reminders for the goals agreed in each one. None of that requires judgment and all of it requires somebody, which is why review cycles slip.
Officeagent handles that layer: it sends the forms, chases the stragglers, books the meetings, and files the signed reviews where they belong, while the goals from each review become tracked items with owners and due dates rather than a line in a document nobody opens until next year. Every message and filing waits on a human approval, which is the right control on anything landing in a personnel file. If the same instinct applies to how you evaluate the organization rather than the individual, scoring the process and culture side systematically is the natural companion exercise.
Frequently asked questions
How long should a performance review be? One to two pages of substance. Length is not quality; a page where every rating carries an example beats four pages of adjectives. The written review is the summary, and the conversation is where most of the value is.
What if the employee disagrees with the review? Ask them to record their disagreement in the employee comments section, so both positions sit in the same record. A signature acknowledges receipt and discussion, not agreement, and your form should say so on that line.
How often should performance reviews happen? A full documented review once or twice a year, a lighter check-in against goals each quarter, and monthly one-to-ones where feedback is immediate. Annual-only reviews are too infrequent to change behavior.
Should I write the review before or after the self-evaluation? After. Send the self-evaluation a week ahead, read it, then draft. Writing first and reading second means you spend the meeting defending a position rather than incorporating information you did not have.
Can I use AI to write performance reviews? To structure and tighten your own notes, yes. To generate the content, no. A review generated from a job title and a rating produces exactly the generic sentences that make reviews useless, and it puts words in a personnel file that no human verified. The evidence has to come from you.