Here is an uncomfortable question for any CEO this review season. How many of the performance reviews in your company were written by the person who signed them?
You probably don’t know. Neither do most of your employees, which is exactly the problem.
Surveys this year put the share of US managers using AI to prepare feedback or reviews at around eight in ten, while fewer than one employee in five says they were told. The tools are already inside the most personal document a manager writes all year. Whether that helps or hurts your business depends on decisions most leadership teams haven’t made.
The experiment every leader should know about
In 2021 the Strategic Management Journal published a field experiment from a large financial services company. Siliang Tong, Nan Jia, Xueming Luo and Zheng Fang randomly assigned 265 employees to receive performance feedback generated either by AI or by human managers, and separately varied what employees were told about its source.
The AI feedback was more accurate and more consistent, and it improved performance. Then came the twist. Once employees learned a machine had produced it, much of that gain disappeared. The researchers called it the disclosure effect. Employees who had been in the job longer reacted less sharply.
Read that carefully and a tempting conclusion appears: use the tool and keep quiet. That would be a mistake, and an expensive one.
Why secrecy is the costly option
Employees are already suspicious. In a Highwire survey of 1,034 US corporate professionals, 40% of individual contributors either suspected AI had shaped their last review or weren’t sure. Suspicion without an answer does more damage than open use, because people fill the gap with the worst version of the story.
There is also a sincerity cost. Research by Peter Cardon and Anthony Coman found that professionals rated a supervisor sincere 83% of the time when a message used light AI help, but only 40% to 52% of the time when it used heavy help. A review is a trust document. Lose sincerity there and you lose it in the conversations that follow.
Then there is the exposure nobody budgets for. A Predictive Index survey of 399 managers found 44% of managers had entered employee names or performance details into public AI tools. If a review built on those notes is later challenged, the employer answers for the decision. The chatbot does not.
Think of a calculator, not a ghostwriter
A calculator helps the student who knows the method. Hand one to a student who doesn’t, and it produces a confident wrong answer with every decimal place in order.
AI behaves the same way with a performance review. Give it a year of dated evidence and a rating the manager has already decided, and it saves hours of drafting. Give it a page of last month’s impressions and ask for a review, and it will invent a coherent story. Sometimes that story praises someone for a project they barely touched. Sometimes it describes a quiet high performer as “lacking presence”, because that is the pattern the model has learned to attach to quiet people.
The tool reflects the manager’s discipline back at them, polished and enlarged. That is why the fix sits with management practice, and why it belongs on the CEO’s desk.
What the leadership team should decide
Approve the tools managers may use for people data, and say clearly that public chatbots are off limits for names and performance details. Set the order of work, so the rating is decided from evidence before any drafting tool is opened. Make disclosure normal, with a single plain sentence in every review that tells the employee how AI was used and confirms the judgment is the manager’s own. Finally, ask HR to sample a handful of reviews each cycle and check that every claim has a record behind it.
Put those decisions on a single page with the CEO’s name at the bottom, and send it out before the next review cycle opens. Sent after the first complaint, the same page reads as damage control, and managers treat it that way.
None of this needs new software. It needs managers who know where the line sits, which is a skills question.
How you will know it is working
Ask HR for two numbers after the next review cycle. The first is the share of reviews in which the employee was told how AI was used. Anything short of all of them means the rule has not yet become habit. The second comes from the sampling: the share of claims in those reviews that can be traced to a dated record.
A manager who scores badly on the second number needs coaching more than a warning. The usual cause is a year of thin notes, not bad faith, and no tool can draft its way out of an empty file.
Watch the appeals as well. A rating that rests on evidence is easier to explain, so when an employee does challenge one, the manager walks into that meeting with dates and examples to point at instead of a feeling. HR spends less time refereeing, and the employee leaves knowing exactly what the rating was based on, even if they still disagree with it.
Training catches up
Few training programs have caught up with the way managers actually work. The Institute for Performance Management has just added two modules to its performance management training course to close that gap. Module 9, AI in Performance Management, gives managers seven working rules, from deciding the rating before opening a tool to holding the conversation themselves, and tests them against a case study of a manager who writes eight reviews in a single evening. Module 10 covers managing performance fairly in hybrid and remote teams.
The course now has ten modules, takes about twelve hours online at the learner’s own pace, and ends in an exam of 20 questions drawn at random from a bank of 30. Pass it and you hold the CPMP from the Institute for Performance Management.
Your managers have already adopted AI. The only open question is whether they do it with judgment or without it.
Sources: Tong, Jia, Luo and Fang, Strategic Management Journal (2021); Highwire survey of 1,034 US corporate professionals, reported by HR Dive and HCAMag (October 2026); Coman and Cardon, International Journal of Business Communication, reported by the University of Florida (August 2025); The Predictive Index survey of 399 managers, reported by HR Dive (August 2026).
Mike Kelly leads the Institute for Performance Management, which trains managers in commercial organizations to run fair, evidence-based performance management and awards the CPMP certification. The Institute’s syllabus is reviewed by its Industry Advisory Council.
