Colleagues applauding a seated coworker around an office meeting table.

Using AI for Employee Feedback: What Employers Should Check

Leave a Comment / HR & AI / By

By: Kristina Kovacevic, Principal Consultant, People & Culture

It’s Thursday afternoon, a performance review is due tomorrow, and a manager is trying to turn scattered notes from the past year into thoughtful feedback before their next meeting. 

With limited time and a blank document in front of them, using AI for performance reviews can feel like an obvious, and helpful, place to start.

AI can be genuinely useful here. It can also quietly create problems that don’t show up until months later, when an employee asks where a particular sentence came from. This article covers where AI helps, where it doesn’t, and what to put in place before your managers start experimenting on their own.

What Should AI Be Used For in Employee Feedback? 

The best way to think about it is that AI is an editor, not a witness. It can help a manager say something well. It cannot tell the manager what happened.

Reasonable uses include:

  • Organizing a manager’s own notes into strengths, development areas and next steps
  • Tightening wording and cleaning up grammar
  • Flagging vague phrases like “needs to be more proactive” and prompting the manager for a real example
  • Helping set up measurable goals, which the manager then approves
  • Practising a difficult conversation using made-up or fully de-identified details

Remote, a global HR company, reportedly finished a review cycle for around 1,800 employees in 48 hours. The detail worth noticing is that their AI drafts were built from months of feedback that had already been recorded, alongside manager training and calibration sessions. The AI summarized work the managers had been doing all year. It wasn’t a shortcut around doing it.

Where Does It Go Wrong?

The risk most people worry about is AI inventing facts. That happens, but it’s the easier problem to catch. The harder one is that AI makes weak evidence sound strong.

Here’s how that looks in practice:

What the manager knows What the AI draft might say What to check
Two reports were submitted late “You consistently fail to meet deadlines” Do two incidents make a pattern?
A client questioned one estimate “Clients have repeatedly complained about your work” How many clients? Was it a complaint at all?
Needs support presenting to executives “Lacks leadership potential” Has a skill gap turned into a career verdict?
Missed some evening meetings “Shows insufficient commitment” Were evening meetings expected? Is there an approved arrangement or an accommodation involved?

Proofreading for grammar isn’t enough. For each important statement, the manager should be able to say where it came from, what the context was, and whether the wording claims more than the evidence supports. 

A simple test: can you explain and defend every sentence without mentioning what the AI suggested? If not, it isn’t ready.

What About Employee Privacy When Using AI?

Think about what goes into the tool before you think about what comes out.

A manager who pastes details into a free personal account has just handed employee information to a third party, possibly without anyone at your company knowing. 

Removing the name doesn’t fix it. Imagine a prompt that says “the only payroll specialist in our Kamloops office, who is returning from cancer treatment.” There’s no name, and everyone who works there still knows exactly who it is. (That’s a made-up example, but you can see how easily it happens.)

Private-sector employers in BC generally fall under the province’s Personal Information Protection Act, and Canadian privacy regulators have been clear that organizations stay responsible when AI is used to support decisions about people. 

Before approving any tool, ask:

  • Is this specific product, account type and use approved for employee information?
  • Can the provider use prompts or uploaded documents to train its models?
  • Who can access the data, and where is it stored and processed?
  • How long is it kept, and how is it deleted?
  • Could the tool pull in information the manager never meant to include?

Also worth saying: don’t upload a whole personnel file when a few relevant lines would do. And don’t let anyone ask AI to guess at things like someone’s mental health, personality or motivation. Those guesses are still personal information about a real person, and they can lead straight to discrimination.

Do Employees Care Whether AI Wrote It?

It turns out they do, and it affects how the feedback lands.

A 2025 study of 1,100 professionals by researchers at the University of Florida and the University of Southern California found that perceived supervisor sincerity dropped from 83% when AI assistance was low to somewhere between 40% and 52% when it was high.

A 2026 experiment with 192 employed participants in Hungary is closer to home. Researchers kept the feedback text identical and changed only who was said to have written it. Feedback attributed fully to AI was received less favourably. Feedback written by a human and polished by AI did about as well as feedback written entirely by a human. Same words, different reaction, depending on how much the manager seemed to be involved.

Overall, what this tells us, is that people want to know what their managers actually think about them.

Who Should Review the Draft?

The manager reviews everything, every time. Beyond that, the level of HR involvement should rise with the stakes.

  • Routine coaching or recognition: the manager.
  • Formal reviews affecting pay or promotion: the manager, plus your usual HR or calibration process.
  • Written warnings and performance improvement plans: the manager and HR, before anything is delivered.
  • Anything involving accommodation, leave, harassment, discrimination or a disputed allegation: HR, and legal counsel where appropriate.

Why the Conversation Still Has to Be Human

AI can help a manager prepare. It can’t have the conversation.

That conversation is where the manager explains what they’ve seen, hears the employee’s side, fixes anything that’s wrong or missing, spots where support or accommodation is needed, and agrees on next steps. An emailed assessment that a manager skimmed and clicked “approve” on doesn’t do any of that.

What Should a Workplace AI Policy Cover?

Put a short written policy in place before a problem forces you to. At minimum, it should cover:

  • Which tools are approved, and which are not
  • What employee information may and may not be entered
  • That a manager must verify every statement and stand behind it
  • When HR review is required
  • That AI never decides ratings, discipline, pay, promotion or termination
  • That the manager delivers the conversation personally

Then train your managers on it. A policy nobody has read is just a document.

Need Help Putting an AI Use Policy in Place?

AI is not going away, and neither is the responsibility that comes with managing people. 

Trusted by 120+ leading organizations, Pivot HR Services helps employers develop practical AI use policies, train managers, and navigate sensitive employee matters.

Whether you’re setting rules for AI-assisted feedback or reviewing a draft that could affect an employee’s pay, promotion or continued employment, we can help you identify the risks and clarify the next steps. Book a free consultation to discuss your organization’s needs.

Sources 

 

Leave a Comment