Nearly half of leaders (47%) are already factoring AI usage into performance evaluations. Considering how recently most organizations were simply trying to get employees to use AI at all, that’s a significant change.
It also raises a much harder question: What exactly counts as good AI use?
Tracking whether employees are using AI can help organizations understand adoption. But once that data starts influencing performance reviews, promotions, or compensation, simply measuring how much someone uses AI can quickly become a problem. An employee who opens an AI tool 20 times a day isn’t necessarily getting more value from it than someone who uses it twice to solve the right problems.
If AI is going to become part of how performance is measured, organizations need to get much smarter about the kind of AI use they’re actually rewarding.
More AI use doesn’t necessarily mean better work
There are plenty of ways to measure AI activity: logins, prompts, time spent in a tool, tasks completed, even the number of tokens an employee consumes. Those metrics can tell you whether AI is showing up in the workflow, but they can’t tell you whether it’s improving it.
That’s an important distinction, especially once a metric becomes attached to individual performance. People naturally respond to what they’re measured against. If the expectation is simply to use AI more, employees have every incentive to find more opportunities to use it, whether or not the work actually benefits.
The result could be an influx of AI activity without noticeable AI impact. Employees may use AI for tasks that don’t need it, prioritize visible activity over useful application, or move faster only to spend that saved time correcting mediocre output.
Usage can be a useful signal of adoption, but it shouldn’t be mistaken for a signal of performance.
Measure the impact, not the activity
Instead of asking how much an employee used AI, start by asking what changed because they used it.
That answer will look different across roles. Saving two hours on a recurring reporting process may be meaningful for one employee. Another might use AI to uncover an insight that improves a customer experience or helps a team make a better decision. Someone else might redesign an entire workflow.
A more useful approach to evaluating AI use should consider:
| Quality | Did AI help improve the final work? |
| Judgment | Did the employee understand when and how to use AI, and critically evaluate its output? |
| Efficiency | Did AI meaningfully reduce manual work or time spent on a task? |
| Problem-solving | Did it help the employee approach a challenge differently or solve something more effectively? |
| Workflow improvement | Has AI changed how work gets done beyond an occasional prompt? |
| Business impact | Did that work contribute to a meaningful outcome for the team, customer, or organization? |
Not every measure will be relevant to every employee, and it shouldn’t be.
What meaningful AI use looks like will depend on the role, the work, and the outcomes an employee can reasonably influence. For some, the biggest impact may come from saving time on repetitive tasks. For others, it could mean improving the quality of their work, making better decisions, or rethinking how a process gets done.
The point isn’t to replace one universal AI metric with a list of new ones. It’s to give managers a better way to understand whether employees are developing the judgment and capability to use AI well—and whether that capability is actually making a difference in their work.
Sometimes, the smartest AI use is no AI at all
This is where measuring volume gets especially messy.
An employee who constantly uses AI isn’t necessarily more AI fluent than someone who uses it selectively. In some situations, choosing not to use AI may be the better decision.
A task may require original thinking, sensitive information, human nuance, deep subject matter expertise, or simply a level of attention that isn’t improved by adding AI to the process. Recognizing that is a skill, too.
As organizations build AI capability, employees need to understand more than how to write a good prompt or navigate the latest tool. They need to evaluate outputs, recognize limitations, protect company and customer data, and decide when human judgment should lead.
The goal isn’t maximum AI usage. It’s appropriate AI usage.
That’s a much harder thing to capture on a dashboard. It’s also much closer to the capability organizations actually need.
Leaders can’t evaluate what they don’t understand
There’s another side to putting AI into performance reviews: the people conducting those reviews need to know what they’re looking for. And our research suggests many leaders are being asked to evaluate AI capability while they’re still building their own.
In the past year, 80% of directors completed leadership-specific AI training, compared with just 55% of VPs. Among VPs, only 58% said they felt confident using AI without compromising company data, while at the same time, nearly half of leaders are already factoring AI use into employee evaluations.
This creates a tricky dynamic. As AI becomes part of performance expectations, leaders need the skills to recognize the difference between thoughtful application and surface-level usage.
This is especially important for middle managers, who sit at the intersection of organizational AI strategy and the employees expected to put it into practice. They need enough AI fluency to have better conversations about where AI improved the work, where it introduced risk, what an employee learned, and where they could apply it more effectively next time.
Done well, performance management becomes another way to develop AI capability, rather than simply track AI activity.
Make AI use worth measuring
As AI becomes part of everyday work, including usage and capability in conversations about employee performance makes sense. But what organizations choose to measure will shape how employees use it.
Measure activity alone, and people will learn to optimize for activity. Measure judgment, improvement, application, and impact, and you encourage something different: employees using AI thoughtfully to improve how they work. AI use becomes less about proving that someone opened the tool and more about showing what they were able to do differently because of it.
For organizations, that means setting clear expectations for responsible AI use and defining what meaningful application looks like across different roles. It also means making sure leaders know what good AI use looks like and employees have the skills to put it into practice.
The goal isn’t “use AI as much as possible.” What matters is whether people can recognize where AI improves the work, where it doesn’t, and use their judgment accordingly.
That’s how AI use starts translating into better work.