AI can help employees summarize accomplishments, turn managers’ notes into feedback and organize reviews before compensation discussions.
But an easier-to-write review is not always a more accurate one.
A draft may overstate a contribution, turn one missed deadline into a pattern or leave out context that changes the assessment. Those errors matter more when reviews influence pay, promotions or whether someone keeps their job.
AI can make an unsupported judgment sound as credible as a well-supported one. HR needs a review process that can tell the difference.
The legal stakes behind the review
Federal employment discrimination laws apply when employers use AI, just as they do to other employment practices. Using AI to support a decision does not remove the employer’s responsibility for its outcome.
States are making those expectations more explicit. California’s regulations, effective October 1, 2025, clarify how existing discrimination protections apply to automated-decision systems. Illinois’s AI employment provisions, effective January 1, 2026, prohibit discriminatory AI use in covered employment decisions and require employee notice of covered uses.
For HR, the practical step is to work with Legal to distinguish tools that help write feedback from tools that influence ratings, pay or promotions, and understand which requirements apply.
Five risks to catch before ratings are final
You may not know when or how AI was used. Focus on whether the review process can catch these problems before they shape decisions.
1. Polished writing can overshadow performance
Two employees contribute equally to a project. One uses AI to write a detailed, polished self-review. The other submits a brief account. Similar work suddenly looks different on paper.
A self-review offers the employee’s perspective. It should not determine the rating on its own.
Ask managers to look past the presentation:
- What did the employee accomplish?
- How did that work compare with expectations?
- What was their individual contribution?
“They led a successful project” needs a closer look if the manager cannot explain what the employee actually led.
The reverse matters, too. A short or understated review does not mean someone accomplished less. Managers should consider work an employee may have overlooked or struggled to describe.
The goal is to evaluate the work, not how convincingly someone writes about it.
2. AI can turn one incident into a pattern
A manager enters “missed a deadline” into an AI tool. The draft describes an employee who “consistently struggles with reliability.”
One event has become a broader judgment.
Managers need to check whether AI changed the meaning of their notes, not just the tone. Words such as “consistently” and “repeatedly” suggest a pattern. Labels such as “difficult” or “not committed” add judgments the original notes may not support.
Give managers one simple question to ask before submitting feedback: Does this say anything stronger or more certain than my examples support?
If the answer is yes, they need to correct the draft. If that broader judgment also influenced the rating, they need to revisit it.
Softer wording will not fix an unsupported rating.
3. Uneven reviews can skew comparisons
AI can summarize dozens of reviews quickly. But a detailed review and three vague sentences do not provide an equally useful basis for comparing employees.
When one manager documents thoroughly and another barely records anything, employees can end up being compared on the quality of their managers’ notes. An AI summary can make that uneven information look ready for a decision.
If HR uses AI to prepare for compensation or talent discussions, check:
- Did the summary leave out examples or context that could change the assessment?
- Did it treat missing information as poor performance?
- Are employees being compared against the expectations for their roles?
Before acting on a recommendation, return to the original reviews and ask managers to fill important gaps. A confident summary should not substitute for a fair comparison.
Also, keep the tool’s purpose clear. Approval to help draft feedback should not extend to scoring employees or recommending promotions. Review that change in use with Legal and IT.
4. Leave or accommodations can be mistaken for poor performance
An employee takes protected leave. A productivity summary shows lower annual output, and a manager’s draft describes the employee as less committed.
Before ratings are finalized, ask managers to check references to attendance, availability and output. Does the assessment appropriately account for protected leave or an approved accommodation?
Also question what managers mean by terms such as “commitment,” “flexibility” or “culture fit.” Are they describing specific job-related behavior? Or are they rewarding longer hours and constant availability without considering the circumstances?
AI can make those assumptions sound objective. HR still needs to question them.
If a recommendation counts protected leave or an accommodation against an employee, pause it and involve Legal before moving forward.
5. Ratings can become hard to explain or correct
An employee challenges a lower rating. The manager points to the review. HR points to an AI-generated summary. Neither can explain what supports the rating.
Managers remain responsible for the reviews they submit. They should be able to connect each rating to job expectations and specific examples.
When an employee raises a concern:
- Pinpoint what they believe is wrong.
- Ask the manager what supports the assessment.
- Consider missing information the employee provides.
- Revisit the rating and any related recommendation if an error affected them.
Use your company’s existing process for reviewing concerns, and make sure employees know where to go if their manager cannot resolve the issue.
For HR’s own AI use, work with Legal and IT to determine what notices and records are needed. A final summary may not explain how a recommendation was reached.
Make better decisions, not just better drafts
Clear rules matter, even when HR cannot see every use of AI.
Make complaints and confidential customer information.
Build three questions into manager guidance and review discussions:
- What work and examples support this rating?
- What information could change it?
- How can the employee raise a concern or correct an error?
AI can help managers find the words. They still need to explain the judgment.
AI can help employees summarize accomplishments, turn managers’ notes into feedback and organize reviews before compensation discussions.
But an easier-to-write review is not always a more accurate one.
A draft may overstate a contribution, turn one missed deadline into a pattern or leave out context that changes the assessment. Those errors matter more when reviews influence pay, promotions or whether someone keeps their job.
AI can make an unsupported judgment sound as credible as a well-supported one. HR needs a review process that can tell the difference.
The legal stakes behind the review
Federal employment discrimination laws apply when employers use AI, just as they do to other employment practices. Using AI to support a decision does not remove the employer’s responsibility for its outcome.
States are making those expectations more explicit. California’s regulations, effective October 1, 2025, clarify how existing discrimination protections apply to automated-decision systems. Illinois’s AI employment provisions, effective January 1, 2026, prohibit discriminatory AI use in covered employment decisions and require employee notice of covered uses.
For HR, the practical step is to work with Legal to distinguish tools that help write feedback from tools that influence ratings, pay or promotions, and understand which requirements apply.
Five risks to catch before ratings are final
You may not know when or how AI was used. Focus on whether the review process can catch these problems before they shape decisions.
1. Polished writing can overshadow performance
Two employees contribute equally to a project. One uses AI to write a detailed, polished self-review. The other submits a brief account. Similar work suddenly looks different on paper.
A self-review offers the employee’s perspective. It should not determine the rating on its own.
Ask managers to look past the presentation:
- What did the employee accomplish?
- How did that work compare with expectations?
- What was their individual contribution?
“They led a successful project” needs a closer look if the manager cannot explain what the employee actually led.
The reverse matters, too. A short or understated review does not mean someone accomplished less. Managers should consider work an employee may have overlooked or struggled to describe.
The goal is to evaluate the work, not how convincingly someone writes about it.
2. AI can turn one incident into a pattern
A manager enters “missed a deadline” into an AI tool. The draft describes an employee who “consistently struggles with reliability.”
One event has become a broader judgment.
Managers need to check whether AI changed the meaning of their notes, not just the tone. Words such as “consistently” and “repeatedly” suggest a pattern. Labels such as “difficult” or “not committed” add judgments the original notes may not support.
Give managers one simple question to ask before submitting feedback: Does this say anything stronger or more certain than my examples support?
If the answer is yes, they need to correct the draft. If that broader judgment also influenced the rating, they need to revisit it.
Softer wording will not fix an unsupported rating.
3. Uneven reviews can skew comparisons
AI can summarize dozens of reviews quickly. But a detailed review and three vague sentences do not provide an equally useful basis for comparing employees.
When one manager documents thoroughly and another barely records anything, employees can end up being compared on the quality of their managers’ notes. An AI summary can make that uneven information look ready for a decision.
If HR uses AI to prepare for compensation or talent discussions, check:
- Did the summary leave out examples or context that could change the assessment?
- Did it treat missing information as poor performance?
- Are employees being compared against the expectations for their roles?
Before acting on a recommendation, return to the original reviews and ask managers to fill important gaps. A confident summary should not substitute for a fair comparison.
Also, keep the tool’s purpose clear. Approval to help draft feedback should not extend to scoring employees or recommending promotions. Review that change in use with Legal and IT.
4. Leave or accommodations can be mistaken for poor performance
An employee takes protected leave. A productivity summary shows lower annual output, and a manager’s draft describes the employee as less committed.
Before ratings are finalized, ask managers to check references to attendance, availability and output. Does the assessment appropriately account for protected leave or an approved accommodation?
Also question what managers mean by terms such as “commitment,” “flexibility” or “culture fit.” Are they describing specific job-related behavior? Or are they rewarding longer hours and constant availability without considering the circumstances?
AI can make those assumptions sound objective. HR still needs to question them.
If a recommendation counts protected leave or an accommodation against an employee, pause it and involve Legal before moving forward.
5. Ratings can become hard to explain or correct
An employee challenges a lower rating. The manager points to the review. HR points to an AI-generated summary. Neither can explain what supports the rating.
Managers remain responsible for the reviews they submit. They should be able to connect each rating to job expectations and specific examples.
When an employee raises a concern:
- Pinpoint what they believe is wrong.
- Ask the manager what supports the assessment.
- Consider missing information the employee provides.
- Revisit the rating and any related recommendation if an error affected them.
Use your company’s existing process for reviewing concerns, and make sure employees know where to go if their manager cannot resolve the issue.
For HR’s own AI use, work with Legal and IT to determine what notices and records are needed. A final summary may not explain how a recommendation was reached.
Make better decisions, not just better drafts
Clear rules matter, even when HR cannot see every use of AI.
Make complaints and confidential customer information.
Build three questions into manager guidance and review discussions:
- What work and examples support this rating?
- What information could change it?
- How can the employee raise a concern or correct an error?
AI can help managers find the words. They still need to explain the judgment.
