Managers Are Using AI to Write Performance Reviews, and Employees Can Tell

Managers are increasingly turning to AI tools to draft performance reviews, and a new survey suggests the results are noticeable to the people receiving them. The findings point to a growing divide over whether automated help improves feedback quality or strips it of the specificity employees need to actually improve. For AI vendors and the companies deploying their tools, the survey lands at an awkward moment: performance management is one of the most sensitive places an algorithm can show up.

The core tension the survey captures is straightforward. AI can smooth tone, cut the time managers spend staring at a blank form, and reduce the odds of a review being accidentally harsh or vague. But reviews are supposed to be personal, tied to concrete examples of what someone did well or badly over the past quarter, and generic language is exactly what a language model produces when it lacks that context. Employees who sense their feedback was generated rather than observed tend to discount the whole exercise.

That reaction matters because performance reviews feed directly into promotions, raises, and in some companies, terminations. If workers stop trusting the process, the downstream effects show up in retention and engagement numbers long before they show up in any HR dashboard. The survey adds to a pattern that has been building for two years: generative AI has quietly moved from coding assistants and marketing copy into the administrative plumbing of the workplace, often without a formal rollout or policy.

The vendors selling into this space have an obvious incentive to frame AI as a co-pilot rather than a replacement. Microsoft has pushed Copilot into Word and Outlook, where performance write-ups are often drafted, and a cluster of HR-tech startups now market review-writing assistants directly to people managers. What none of them can do is supply the manager's memory of a specific project, a missed deadline, or a moment where someone stepped up. That gap is where the credibility problem lives.

For now, the practical advice circulating among HR leaders is to treat AI as a first-draft tool and require managers to add at least one concrete, verifiable example before a review goes out. Some organizations are going further, logging when AI was used in the review process so employees know what they are reading. Whether that transparency helps or simply makes workers more suspicious is an open question, and the survey does not settle it.

The next thing to watch is whether review-writing tools start integrating with project management and code repositories, giving models the concrete evidence they currently lack. If that happens, AI-assisted reviews could get more specific rather than less, which would flip the current complaint. Until then, the burden falls on managers to decide how much of their judgment they are willing to outsource, and on employees to decide whether they trust what comes back.