In brief
- In one recent survey, 78% of managers said they’d used AI to help with feedback or a performance review. Only 16% of employees said they’d been told AI played a role in their last review.
- AI didn’t create the problem. Managers have always struggled to find the words, and they’ve always borrowed them from somewhere.
- What’s missing is language from the company itself: what good looks like here, and what too little and too much look like.
Picture an employee reading their annual review. It’s polished. It’s well organized. It uses phrases their manager has never once said out loud.
They start to wonder. Did my manager write this? Did they think about my year at all, or did they paste in a few notes and let a tool fill in the rest?
That question is showing up in a lot of companies right now, and most leaders haven’t noticed. A survey of more than 1,000 full-time U.S. corporate professionals at companies with 100 or more employees from Highwire and Knopman Marks found that 78% of managers had used an AI tool to prepare feedback or a performance review in the past year. Only 16% of individual contributors said they’d been told AI played a role in their most recent review. The two numbers come from different questions and different groups, so they don’t measure how many reviews went undisclosed. But the distance between them is hard to ignore.
It would be easy to read that and blame AI. I don’t think that’s the real story.
AI didn’t create this problem
Performance reviews have always been hard for managers. Long before AI, there were whole books of ready-made review phrases. A manager who didn’t know how to describe someone’s year could flip to the right page and borrow a sentence that sounded professional.
Some managers bought those books. Some didn’t. Either way, the borrowed language came from outside the company, and it said nothing about what good work looked like there. It just didn’t seem like a big problem, because it was easy to miss.
AI is doing the same thing, at a much bigger scale. When most managers pull from the same tools, the reviews start to sound the same. The words are polished and generic, because they were never written for your company in the first place. AI isn’t causing the problem. It’s exposing one that was always there.
The survey shows both sides of that. In the same Highwire survey, 54% of respondents said AI-assisted feedback became more specific and actionable. A manager who struggles to put feedback into words can get real help from it.
But 34% said feedback became more generic, and 32% said it became less useful. And 40% suspected AI had been used without anyone confirming it, or weren’t sure.
That last number matters. A performance review is supposed to tell someone that a person who works with them paid attention to their year. When employees suspect the words came from somewhere else, and nobody will say either way, they stop trusting the review, and then the conversation that goes with it.
This isn’t a manager problem
It’s tempting to read those numbers and decide managers are cutting corners. That’s not what I see.
Most managers were never trained to write reviews well in the first place. They’re doing them on top of a full job, often for a lot of people at once. A tool showed up that makes the hardest part easier, and nobody told them how it should or shouldn’t be used. So each manager decided for themselves.
Highwire’s CEO, Liza Streiff, put it plainly: “AI is filling a gap, not addressing the issue.” Managers have always needed language to describe performance. The company rarely gave it to them, so they found it in a book, and now in a tool.
The language your managers need should come from you
Every company has a culture, whether anyone has named it or not. But when it’s named, as your company’s personality, something useful happens. You can describe what good work looks like here, in your company’s own voice.
That’s the language managers have been missing. Not “exceeds expectations” or a phrase from a book, but a clear picture of what each part of the job looks like when it’s done the way your company does things. And just as important, what it looks like when someone does too little of it, and when they do too much.
Too much is the part most review forms miss. It doesn’t mean someone is too good at their job. It means a strength your culture values is starting to crowd out the work. A company that prizes speed can end up with someone who moves so fast the quality slips. A company that prizes care can end up with someone who protects people so much they stop holding anyone accountable.
When managers have that language, they don’t need to borrow it. It drops right underneath the rating scale you already use, so “meets expectations” finally means something specific. I walk through how to turn what your company values into behavior people can see in How to Turn Your Company Values Into Behaviors People Can See, and the Culture Code™ is how that language gets built.
The harder question: what are you rating now?
AI is also changing the work being reviewed. In a monday.com study conducted by Nielsen of 1,700 directors and senior managers across seven countries, more than half of senior leaders said it had become harder to evaluate employees’ work after adopting AI, and 78% said they’d changed how they evaluate people.
If an employee used AI to draft a report, what part of it is theirs? Usually it’s the judgment. What they asked for, what they caught, what they fixed, and whether they stood behind the result. Those are the things worth rating. Output volume isn’t, because the tool can inflate it.
That means going back to what the job is actually for. I write about getting that clear in Skills vs. Behaviors, and it matters even more when part of the work is done by a tool. For a look at what AI should and shouldn’t handle in the review itself, see AI in Performance Reviews: Faster Ratings or Better Managers?
What to put in place
None of this calls for a ban. It calls for a few decisions, made once by the company, so managers aren’t left to guess.
| Decision | What it looks like |
|---|
| Language | Managers describe performance using the company’s own definitions of what good looks like, and what too little and too much look like |
| Disclosure | Employees are told when AI helped prepare their review, and how |
| Ownership | The manager owns every word and can explain any point in it |
| Evidence | Feedback is tied to real examples from the employee’s year, not general phrases |
| Privacy | Employee names and performance details don’t go into public AI tools |
| What good looks like with AI | Expectations say how AI can be used in each role and what the employee is accountable for |
These aren’t HR’s rules to make alone. They’re decisions about what kind of company you want to be when it evaluates people, and they belong with leadership. HR can then build them into how performance conversations actually happen.
The verdict
AI is getting a bad rap here. Performance reviews were a struggle long before it arrived, and managers were borrowing their words long before it arrived too. AI just made it obvious.
The fix isn’t banning the tool or buying a better one. It’s giving managers language that comes from your company: what good looks like here, and what too little and too much look like. Then AI can help them write it up without replacing what they mean. Trust is part of how people experience HR, and a review is one of the moments they experience it most. That’s the heart of HRX™, the HR experience.
If your reviews already feel like a box to check, this is a good time to look at the whole process. I lay out the most common gaps on my performance reviews page.
Questions worth asking
- Do you know how many of your managers used AI on their last round of reviews?
- If an employee asked whether AI helped write their review, would their manager know what they’re allowed to say?
- Now that AI helps with the work, what are you actually rating?
Read next: How to Run a Performance Conversation, Not a Performance Review
Frequently asked questions
Is it okay for managers to use AI to write performance reviews?
It can help, especially with organizing feedback and finding clear words. The risk comes when employees aren’t told, when the feedback turns generic, or when the manager can’t stand behind what it says.
Should employees be told when AI is used in their performance review?
Yes. Telling people protects trust. When employees suspect AI was used and nobody will confirm it, they start doubting the review and the conversation that goes with it.
How do you evaluate work when employees use AI?
Focus on the parts that belong to the person: their judgment, what they verified, what they changed and whether they stand behind the result. Output volume alone says less than it used to.
Who should set the rules for AI in performance reviews?
Leadership, with HR. The rules reflect how the company wants to treat people when it evaluates them, so they shouldn’t be left for each manager to decide.
