Should AI write performance reviews?

Confused in Cleveland, this is exactly where an AI champion earns the title. A champion should not cheer every use of the technology. Your job is to help the company use it well, including recognizing when a system is costing more in trust than it saves in time.

Managers love the time savings for an obvious reason: review season is brutal, and anything that drafts paragraphs feels like oxygen. Employees whisper about the “same voice” for another obvious reason: sameness is what people hear when institutions stop seeing them. Both reactions are political. One group is protecting capacity. The other is protecting fairness. People ops sits in the middle—expected to cheer adoption while quietly holding the trust account.

Everyone involved has a reasonable priority. Managers want relief from a painful administrative process. HR wants consistency. Leadership wants evidence that the company is adopting AI. Employees want to know that someone actually saw their work, understood the circumstances, and cared enough to offer useful feedback. The tension begins when hours saved for one group create distance and uncertainty for another.

Calendar patterns and message summaries can show activity, but activity is not the same as performance. A machine can organize observations and improve a rough draft. It cannot know whether someone calmed an angry client, helped a colleague through a difficult week, or quietly prevented a project from failing. Those details often carry the most useful feedback, and they live in relationships more than in the metadata.

We have found the metadata is very valuable for learning about baselines, building chronologies for disputes or benchmarking, and taking retrospective looks. It is not always as good at measuring culture in real time.

I would call this out with evidence instead of opposition. Maybe start with a survey or feedback session to get real data from employees about effectiveness and sentiment. Show employees you are listening and care by actually measuring the feedback. That gives you both the cultural boost of active listening and the data-driven, closed-loop approach any AI champion should want. Then you can say: “The tool is saving managers time, but employees are telling us that the reviews are becoming less personal and less useful. Can we narrow the role of AI so it helps managers write better feedback without replacing their judgment?” That makes you the person improving the implementation, rather than the person trying to stop it.

If you decide to go forward, perhaps a reasonable boundary would require the manager to decide the rating and provide specific examples before AI touches the review. The tool can then organize the draft, improve clarity, and identify vague language. Every review should still explain what the person did, why it mattered, and what would help them improve. Employees should also know which information was used and have a way to correct factual errors. Before activity data influences employment decisions, HR and counsel should confirm that the process, consent, and retention rules are appropriate.

Measure more than the hard KPIs. Review a sample of the output for specificity, usefulness, factual accuracy, and employee trust. If every employee sounds the same, the tool is removing the very context that makes feedback valuable.

Being a human and culture champion is part of being a credible AI champion. The strongest advocates are the people who help the organization learn where AI creates value, where it creates risk, and where a human still needs to look someone in the eye and own the words.