AI in Performance Reviews: Define the Rule Before You Rate

Written By Breakfast Leadership

AI in performance reviews is already happening in most organizations, whether or not anyone has written a policy. A manager notices who drafts faster, who produces cleaner summaries and who seems to lean on a tool, and those impressions shape ratings, assignments and promotions. For leaders at nonprofits, associations, healthcare groups and professional services firms, the risk goes beyond fairness. When expectations about AI are unwritten, employees guess, and in a regulated setting a wrong guess can put confidential data into an unapproved tool.

A Checkr survey of 3,000 employed U.S. adults, fielded February 9 to 12, 2026, found that 58% of managers agree AI use is becoming an unspoken performance requirement at work. Only 29% of employees agree, and 37% say they are genuinely unsure. Managers believe a rule exists. Employees cannot see it.

This article is written for leaders who have bought or piloted AI and cannot yet show measurable adoption. It explains why the rule gap forms, what Gallup's 2026 data says about the conditions that change behavior, and the decisions that turn an unspoken expectation into one your people can actually meet.

AI in Performance Reviews Starts With a Rule Nobody Has Written Down

The Checkr finding is easy to misread, so it helps to be precise about what it measures. It reports agreement that AI use is becoming an unspoken performance requirement. It does not report how many organizations rate AI use formally. What it does show is a 29 point gap between how managers and employees read the same environment, and a further 37% of employees who cannot tell which reading is right.

Both groups are behaving sensibly. A manager who believes AI fluency is now expected will notice it, reward it and quietly worry about people who seem to avoid it. An employee who has never been told what is expected has two safe options: use the tool quietly and say nothing, or avoid it and hope that is acceptable. Neither produces the open practice sharing that real adoption depends on.

For a leader, the cost is invisible until a review cycle exposes it. Ratings drift toward whoever is visibly experimenting. Cautious staff, often the people with the deepest professional judgment, read the silence as a warning. In regulated work, the employees most careful about data are the ones most likely to stay away from the approved tool.

The fix is smaller than most executives expect. For each role, write one plain statement of what AI is expected for, what it is permitted for and what it may not touch. A single paragraph per role is enough, as long as every manager can say it aloud and every employee has read it.

This quarter, ask your direct reports to write that paragraph for their own function and bring it to your next leadership meeting.

Managers and Employees Are Describing Two Different Workplaces

The same survey shows how far apart the two groups sit on basic facts. Among managers, 45% believe their colleagues use AI frequently or constantly. Among employees, only 18% believe the same about their peers, and another 16% say they are not sure.

The two groups also disagree about where the push is coming from. Checkr reports that 34% of employees say they do not know where the push to adopt AI is coming from. Only 8% of employees name managers as the driver, while 21% of managers see themselves in that role.

Read together, these figures describe a middle layer that believes it is leading adoption and a workforce that cannot see it happening. That is an observation about perception, and it matters for a practical reason. If a manager thinks everyone is already using AI, that manager will not explain it, demonstrate it or ask what is getting in the way. The conversation that would close the gap never starts.

You can test this in your own organization in an afternoon. Ask managers and staff the same two questions: how often do people on your team use the approved AI tool for real work, and who decided that they should? Compare the answers by team. A wide spread tells you where the unspoken rule is doing the most damage, and it costs far less than another round of training.

Treat the result as a diagnostic, not a verdict.

What Gallup Shows About Plans, Manager Support and Engagement

Gallup's 2026 workforce surveys give leaders a clearer picture of which conditions are associated with better outcomes. The data comes from surveys fielded February 4 to 19 and May 6 to 20, 2026, covering 43,262 U.S. adults who work full or part time. The AI figures below use the subsample of 21,724 employees whose organizations have integrated AI tools.

Among those employees, 43% of those who say their organization has a clear plan for integrating AI are engaged, compared with 28% of those who say it does not or do not know. Manager support shows a similar pattern: 48% of employees who strongly agree their manager actively supports their team's use of AI are engaged, compared with 30% of those with lower agreement. Gallup also reports that only 36% of employees in AI-integrating organizations strongly agree that their manager actively supports it.

These are associations, not proof of cause. Even so, the direction supports a management conclusion: the plan and the manager's visible backing are the two conditions leaders control most directly.

Frequent AI use is also associated with higher engagement. Gallup found 39% of weekly users engaged against 31% of less frequent users and non-users. That is a reason to remove barriers to use, not a reason to mandate it. A review system that forces visible usage will inflate the frequency number without improving the work.

This is why the problem belongs in an operating system for the organization rather than in a launch email. A Leadership OS review looks at exactly this layer: who owns the plan, how managers reinforce it and where the routine breaks down.

This quarter, put one question on every manager's monthly agenda: what did your team do differently because of the tool?

How to Rate AI Use Without Rewarding the Wrong Behavior

Some companies have responded to the gap by adding AI to the review form. Korn Ferry reported in March 2026 that firms are measuring AI use in performance reviews, in some cases tying it to promotions and raises, and that one manager found a "technology utilization" line sitting under quality of work. Korn Ferry's Bryan Ackermann called this an over-rotation and said it "isn't a measure of learning agility."

The pressure explains the temptation. Checkr found that 64% of managers feel at least some pressure to adopt AI tools to stay competitive, while 38% of employees feel meaningful pressure and 36% feel none at all. A leader under that pressure wants a number to show the board. Usage counts are the easiest number to produce and the weakest evidence of value.

A better review rates what the work shows. Ask each employee to bring one example where AI changed how a task was done, what they checked before relying on the result, and what they chose not to delegate to the tool. This rates judgment, which is the scarce skill, rather than activity. It also gives a cautious employee a legitimate way to demonstrate good practice, because declining to use AI on a sensitive task is itself a documented decision.

Be explicit about what the review does not do. It should not penalize people for time spent verifying output, and it should not treat tool logins as performance. State both in writing before the cycle opens so that employees are not discovering the criteria in the meeting.

This quarter, redesign one review question around outcome and judgment and pilot it with a single team before changing the form for everyone.

Trust Is the Compliance Question Inside Every AI Review

The sharpest finding in the Checkr report concerns trust. It reports that 40% of managers trust AI outputs often or almost always in their daily work, compared with just 9% of employees. Fifty-nine percent of employees rarely or never trust them. At the same time, 80% of managers believe AI outputs are trusted too much in decisions at least sometimes, and 68% of employees agree AI is over-relied upon.

Employees do not trust the output of the tools they are told to use, and both groups worry that other people trust it too much. Nobody has defined how much checking is enough, so each person sets their own standard.

In a regulated environment, that is a governance gap. A clinical administrator, a grants officer or a consultant needs to know which outputs can be used with light review and which require a named person to confirm them before they reach a patient, a member or a client. The answer should depend on the consequence of an error, not on individual comfort.

The review process is where that standard becomes real. When managers rate verification as part of good work, employees stop treating checking as wasted time. When the organization records what AI may draft and what a human must decide, the review has something concrete to measure against, and the board gets evidence of oversight rather than a promise of it.

If your organization has purchased or piloted AI without measurable adoption, the AI adoption advisory is built to design that standard, the scorecard and the manager role together.

This quarter, publish a one-page list of AI outputs that need human confirmation, and name who confirms each.

Conclusion

The unspoken rule is the problem. Managers believe AI use is becoming a performance requirement, employees cannot see the requirement, and the gap between the two produces hidden use, avoidance and uneven ratings. Gallup's data points to the conditions that help: a clear plan, a manager who visibly backs the team, and room to use the tool on real work.

Four decisions close the gap. Write one plain AI expectation for each role. Compare how managers and staff describe adoption and act where they disagree. Rate outcome and judgment rather than usage. Define which outputs a human must confirm, and treat that confirmation as part of good work.

They require an executive to decide that AI expectations belong in the same system as every other expectation. If you lead an organization where AI has been bought but not adopted, the fastest next step is a conversation about where your own rule is breaking down. You can book an intake call with Michael and decide what to fix first.

One manager who explains the AI rule helps one team. A system that makes every manager explain it changes the review cycle.

The practices in this article are the right ones: a written expectation per role, a shared picture of adoption, reviews that rate judgment and a clear list of what a human must confirm. They are also fragile, because each depends on a leader remembering to reinforce them under pressure, and they erode through drift, unwritten exceptions and ad hoc fixes the moment a review cycle gets busy.

Leadership Operating System

An unspoken performance rule is rarely a discipline problem; it is a structural gap between what leadership expects and what the review system actually states. See how a Leadership OS review works.

AI Adoption Advisory

If your AI tools are licensed but employees cannot tell what is expected of them, adoption needs its own owner, scorecard and oversight design. See how fractional AI adoption advisory works.

Book a call Either track starts the same way.

Michael D. Levitt is the founder of Breakfast Leadership Network and the author of Burnout Proof and Workplace Culture. More writing at the Breakfast Leadership blog.

Additional Resources

Sources

Breakfast Leadership
Next

Why Workplace Wellness Programs Fail and How Leaders Fix It