the review that wrote itself

picture a manager in late october with a page of rough notes, 8 reviews to write and a friday deadline. a chat window will turn those notes into something that reads like a finished performance review in under a minute, as if the review wrote itself. we get the pull.

plenty of managers have already given in to it. a survey published on october 6, 2026 asked 1,034 people at us companies about feedback at work¹. 78 percent of the managers said they had used an ai tool for employee feedback or a performance review in the past 12 months. only 16 percent of the employees surveyed said they had been told ai played a role in their own last review.

and it isn't only managers working alone. in october 2025, bloomberg reported that citi was bringing in ai to help its managers write performance reviews². once an employer does that on purpose, this stops being a personal shortcut. it becomes a policy, whether or not anyone established one.

what a review is for

a review does several jobs at once. it feeds pay. it sits in a file that can come up in a promotion case or a dispute. and it is one of the few places a person sees, in writing, where they stand with their manager.

a tool can't do that last job. it only knows what's in your notes, and the notes miss most of the year: the meeting in march when the project slipped, who stayed late to save the launch, who quietly carried a struggling teammate. it will still write smooth sentences about all of it.

people notice. in the same survey, a third of all respondents said feedback had become more generic since ai came into it¹.

the law has already noticed

if you have staff in europe, this is already more than a question of taste. the eu ai act, the european union's law on artificial intelligence, puts ai systems "intended to be used to monitor and evaluate the performance and behaviour" of people at work on its high-risk list³. the same law says an employer using a high-risk system at work has to tell the affected workers, and their representatives, before it goes in³.

whether one manager pasting notes into a general chatbot counts is not settled. a company tool built to draft reviews is much closer to the line. our read: the law is moving toward telling people when a machine helped evaluate them.

write the part only you can write

before opening any tool, write down 3 things only you could know about that person's year. a moment they handled well. one they didn't. what you want from them next. start the review from those 3.

then use a tool on the wording if you want. tighten the sentences, fix the tone, catch the typos. if it helped, say so in one line at the end. something like: "i wrote the assessment in this review myself, and used an ai tool to help with the wording."

where ai helps, and where it doesn't

a review should feel genuine. the person reading it should be able to tell that their manager wrote it, because their manager knows them.

that still leaves ai real work to do. it can pull a year of scattered notes into one place, line them up against the goals you set, and suggest angles you can give the person on how to improve. you read those, keep what's true, and say it in your own words. what it can't do is make the judgment call, and it shouldn't be the one deciding what the person is worth.

if you run hr

most of the managers in that survey were making this call on their own. the fix is a rule short enough to fit on one page, written before the next cycle starts.

it needs 4 lines. what a manager may use a tool for: pulling notes together, suggesting angles for growth, and wording. what they may not: the rating and the assessment. how the employee is told, in one approved sentence like the one above. and what never goes into a tool at all: health information, accommodations, anything someone said in confidence. if you have people in europe, add one question for legal: does our review tool count as a high-risk system under the eu ai act, and if so, have we told the people it evaluates? get the answer in writing.