Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Here’s a safe way to use AI to write performance reviews

Beware of these three risks to using free, public AI platforms.

[Photo: Getty Images]

Profile photo of Doug Dennerline

CEO · Betterworks

Lately, there’s been a lot of discussion on LinkedIn about using public large language models (LLMs) to write performance reviews, both self-reviews and manager-written reviews.

I’m a strong proponent of using AI. It often does a better job than managers at crafting performance and development feedback because most managers aren’t trained writers. Many struggle to deliver difficult messages clearly and fairly. And recency bias—the tendency to emphasize an employee’s most recent performance and behavior while discounting earlier performance—often skews reviews toward what happened last week rather than the full review period.

So let me be very clear: AI absolutely belongs in performance management. But where and how it’s used matters enormously.

Over the past few months, I’ve watched a growing trend of managers and employees using AI tools—particularly free, public LLMs—to write reviews, often without company guidance or approval.

It’s easy to see why. Review season is stressful. Writing reviews is time-consuming and the blank page is intimidating. Cultural differences can make phrasing even harder. AI promises speed, structure, and polish.

The problem is that using public AI tools for performance reviews introduces risks that many organizations aren’t yet fully considering—and HR leaders will be left to manage the fallout. Here are three risks managers face when using free AI programs to write performance reviews.

1. Privacy and data exposure

    Performance reviews are among the most sensitive documents a company creates. They include personal employee information, context around underperformance or behavior issues, and often proprietary business details.

    When that information is pasted into a public LLM, it may be used to further train the model unless safeguards are explicitly turned off—and most people don’t know how to do that.

    This creates a dual risk. Company and employee data can leak into external systems, potentially resurfacing where it doesn’t belong. At the same time, AI-generated content can be inaccurate, biased, or inappropriate. These are issues a manager may not catch before sharing the review. What starts as a technology shortcut can quickly become a trust, compliance, or legal issue.

    Public LLMs are trained on vast, generalized data sets. They aren’t trained in how managers and employees should discuss performance, growth, and development.

    When a model hallucinates or subtly introduces biased language into a review, the impact is real. An employee can legitimately say, “This doesn’t reflect my work.” In the worst cases, it can escalate into grievances or lawsuits.

    What concerns me most is how often managers accept AI output without question. There’s rarely a built-in checkpoint asking, Is this accurate? Does this reflect what I want to say? Without that pause, AI replaces judgment, and that’s where things break down.

    AI should always be reviewed by humans. That’s how we approach it at Betterworks. Instead of relying on public AI tools, we use secure, enterprise-grade models designed specifically for performance management. These private environments generate more relevant, professional, and less biased output, and they improve over time, without exposing sensitive employee data.

    3.   Eroding trust in the performance process

    Performance reviews only work when employees believe the feedback is real, thoughtful, and human. When reviews feel generic or machine-written, trust erodes. Employees disengage while managers check the box. And we recreate exactly what we were trying to move past: performance reviews that feel generic, thoughtless, and useless.

    There’s also a longer-term risk: AI summarizing AI summarizing AI. Over time, this flattens reviews and everything becomes vanilla. Everyone starts to sound the same. Feedback drifts further from reality, the opposite of what performance management is meant to achieve—meaningful feedback and career development.

    The right way forward

    The answer isn’t to avoid AI. It’s to use it intentionally and responsibly.

    Used well, AI can help eliminate bias, surface a complete view of an employee’s work, and save time while improving quality. But it must live inside systems designed for performance conversations, with guardrails, privacy protections and human validation built in.

    Employees and managers are already using AI. The question isn’t whether AI will show up in performance management, but whether HR will lead this responsibly or spend years reacting to avoidable problems.

    AI can make performance feedback better. But if we get careless about where and how it’s used, we risk breaking the very thing performance management is meant to build: trust.

    And once that’s gone, no amount of technology can fix it.

    Doug Dennerline is CEO of Betterworks.