A growing number of companies are folding AI usage volume into hiring and layoff decisions. If your intention is to spotlight your organization as an industry leader and a top trusted entity in your markets, poor measurement of AI use in your company is about the worst possible way to start. Here’s what to measure instead.
Companies are increasingly factoring AI usage volume into decisions about who gets hired and who gets riffed. The intent is understandable: leadership wants proof that the organization is adapting and using new technologies and achieving ROI on AI token spend. But AI usage volume, taken by itself, measures and shows the wrong thing. Focusing too heavily on it, without accounting for other, more important factors, can ultimately erode a company’s trusted reputation and industry credibility.
What "usage volume" actually measures
Last month, twenty-six current and former Meta employees filed suit alleging that the company had used an assemblage of internal AI systems, including AI-token-usage dashboards, keystroke and activity monitoring, and algorithmic performance ranking, to select employees for a 10 percent workforce reduction. Court filings describe internal dashboards that classified employees by AI-adoption stage, using labels like "AI Native," "AI First," and "AI Enabled." The suit further alleges that employees on medical, parental, or disability leave couldn't accumulate the same usage signals as their peers and were disadvantaged as a result. Meta disputes that AI usage drove its layoff decisions, but the case is reportedly among the first to target AI's role in firing decisions specifically, rather than hiring.
Meta is not an outlier. Nearly every Fortune 500 company now tracks AI usage rates at the group, role, or individual level, according to CNBC, as token costs have become a standard line-item business expense. And that is exactly what AI usage actually measures: the number of tokens that employees use within a given AI tool or resource. It's like counting prompts without analyzing the substance of the dialogue between the human and the AI agent. And the end result could be promotions for people who use lots of tokens to churn out useless, failed products and pink slips for high performers who let their tokens sit on the shelf.
Granted, effectiveness is much harder to measure than the hard number of “tokens consumed.” But a metric is not necessarily good just because it is more easily attainable. This gap shows up in the numbers: McKinsey research finds that 64 percent of companies say AI is driving innovation, yet 39 percent report a measurable earnings impact.
The cost to employees and to the work
Rewarding usage volume rewards the wrong behaviors. Employees learn to game the metric — for example, by padding prompts or running trivial tasks through AI that they could have completed just as quickly without it. This is Goodhart's Law in practice: once a measure becomes a target, it stops functioning as a good measure. As one Forbes analysis put it, employees respond to whatever is tracked, staying visibly online when presence is monitored, sending more messages when messaging is monitored, consuming more tokens when token usage is monitored, producing more activity without necessarily producing more value.
There's a second cost that generally goes unnoticed: trust. When AI usage becomes a surveillance metric tied to job security, the message employees receive is that the company has prioritized hoop-jumping over quality work.. By the time performance reviews roll around, employees will have a lot of hoops to boast about and little else. And as the Meta litigation illustrates, usage-based scoring can penalize employees whose lower activity has nothing to do with judgment or output, including those on protected leave, creating legal exposure that leadership may not have factored into the equation.
Also, employees talk. Social media and platforms such as Indeed and Glassdoor will capture and share their sentiments regarding value perceptions. Employees appreciate having the right tools to do their jobs, and if they’re going to be measured, it should be for how effectively they use them, not merely how often.
None of this means AI itself is bad for employees. Quite the opposite: When a company adopts AI deliberately, with a clear policy and real training behind it rather than a quota, employees benefit directly. They get tools that take tedious work off their plate, freeing up time for the innovation, judgment calls, and problem-solving that make a job worth doing. Real training also builds a skill employees carry with them, rather than a workaround they picked up to survive a review cycle. And when usage stops being the yardstick, employees are free to use AI where it actually helps and skip it where it doesn't.
Usage-chasing doesn't even hold up on the company's own financial terms. KPMG found that nearly half of organizations have already scaled back AI rollouts after costs outweighed the expected value. And when vendors shift to usage-based pricing, the bills climb quickly. When GitHub Copilot moved to full usage-based billing in June 2026, one developer's projected monthly AI cost jumped from roughly $77 to $1,115. A metric that's expensive to chase and disconnected from output quality is a poor foundation for personnel decisions.
What sound AI adoption looks like
Leaders who want to assess AI fluency should evaluate judgment directly rather than inferring it from usage logs.
For existing employees, dedicated training in AI functions and functionality will go a long way. It’s understandable for leadership to want to get the company’s money’s worth, but they must be careful not to obsess over that. Instead, make sure the people doing the hiring and firing understand the AI investments themselves. Then they can truly evaluate their employees’ usage rather than just count beans.
Done well, this pays off for the company as much as the employee. A clear AI usage policy spells out which tools are approved, what data can and can't be shared with them, and how AI-assisted work gets reviewed. This closes off the compliance and IP exposure that comes with ad hoc adoption, including the kind of exposure the Meta litigation illustrates. Pair that policy with real training, and the learning curve across the organization smooths out, so gains in speed and quality show up in the work itself instead of in a token dashboard.
Companies that invest here tend to see it in retention and recruiting, too. Employees stay longer when they trust how they're being evaluated, and it gets around that a company treats AI as a tool for judgment rather than a stopwatch on activity. This adds up to a solid advantage in a competitive market for skilled talent.
In new hire interviews, ask candidates which AI tools they favor and why. A prospective hire will reveal a lot by whether he responds fluently or sounds as though AI wrote his answer.
A good interview will focus on how candidates have used AI, not whether or how much. Prospects should walk through a work product they built with AI, and specifically where the AI output ended and their own judgment took over.
Ask when they would or did choose not to use AI. A candidate who can name where AI doesn't belong in the role demonstrates more judgment than one who scrambles to find a role for it.
All of these moves will help leadership fight the tendency to “outsource” personnel management to a program. People should be evaluated by people. AI functions can help with this, but the moment they become more than subordinate, soulless processes will produce soulless (and ultimately unproductive) employees.
Measuring what matters builds trust
Obsessing over AI usage frequency can look like a company embracing forward-looking technology and seizing opportunities. What it becomes in practice is a shortcut or a crutch, with real human consequences for both employees and the company’s bottom line.
The inverse is just as true. Companies that pair AI adoption with clear policy and genuine training tend to land closer to the ROI leadership is chasing in the first place — faster onboarding to new tools, fewer costly missteps with sensitive data, and employees who reach for AI where it adds value and set it aside where it doesn't. That combination, not the token count, is what eventually shows up in earnings. It compounds, too: A workforce trusted to exercise judgment becomes a recruiting and retention advantage, which is its own form of industry credibility.
The pressure to demonstrate rapid AI adoption is real. In many — perhaps even most — scenarios, adoption is the right course. But sustainable growth and industry credibility depend on measuring outcomes that hold up over time — not the ones that are simply easiest to count.
