An employee reads her annual performance review and finds credit for a project she never worked on. The explanation is worse than the error. Her manager has drafted her reviews with an AI tool for two years, edited the output lightly, and delivered it as a personal assessment. She is less angry about the AI than about what it implies: nobody has been watching her work. That was the scenario HR Voices host Rebecca Taylor put to Jaime Akers, Vice President of Team Member Relations at Hilton Grand Vacations.
Jaime’s first reaction was a greeting: “Welcome to HRing in 2026.” Her second was to slow down. Before HR decides whether this is a performance management failure, a documentation integrity problem, or both, it has to establish what the company allowed, what the manager was trained to do, and what happened. Her name for that discipline is “freedom in a framework.”
Why the Investigation Starts With the Investigator
Jaime’s first step is not an interview. It is a pause. A case like this arrives with emotion attached, and she wants HR practitioners to notice their own thinking before they act on it. Researchers call the habit reflexivity. In practice it means naming the bias that surfaces first and setting it aside.
Then she gets curious. The manager may work in a second language and lean on the tool to sound professional. The manager may have been trying to get the words right. Jaime wraps the whole process in “kindness and curiosity,” and she still expects to get to the bottom of it.
The first evidence she pulls is the company’s written policy on AI and acceptable use, along with the training behind it and any IT governance council decisions. Without that document, she argued, HR runs the case on personal values. “Consistency breeds credibility,” she said, more than once. Outside guidance exists for companies still writing theirs, including the voluntary AI Risk Management Framework from NIST.
What the Hallucinated Project Might Actually Be
Jaime offered two readings of the phantom project. In the first, the manager followed the rules. Approved tools often come with an instruction to redact names or swap them for a private key before anything is uploaded. A substituted project name that never got swapped back would look exactly like a hallucination.
In the second, the manager asked the tool to read a year of email, accepted what came back, and never checked it. That is a different conversation, and only fact-finding separates the two. Jaime would pull the manager’s tenure, time in a leadership role, training logs, and signed policies, then ask the manager’s own leader what coaching has already happened.
Neither reading removes the oversight problem. “With great capability comes great accountability,” Jaime said, and she put a number on it: a leader has to care enough to spend twenty minutes checking the output. The stakes run past one error, because reviews feed pay, performance improvement plans, and internal mobility. Gallup has reported that only 14 percent of employees strongly agree their reviews inspire them to improve. A review nobody wrote will not raise that figure.
What Actually Works
Run Every Case Through the Same Framework
Jaime’s process is standard by design. HR makes an initial assessment, assigns the right investigator, and recuses anyone with a relationship to the people involved. Intake captures dates, times, people, and departments. Scope is defined before the plan is written, and the manager’s leader usually hears that HR will be in the area.
The output is modest on purpose. HR is not judge and jury. Its job is to decide “what is likely to be true,” line that up against the policy, and stand behind the result. Companies that run investigations the same way each time also spare their teams the decision fatigue Rebecca described.
Separate the Coaching Moment From the Cover-Up
Jaime draws the line at what happens after the mistake. An employee who uploads something by accident, flags it, and calls the right team has shown the behavior a company wants. One who hides or deletes it has created the larger problem, and possibly legal exposure under GDPR and its HR obligations or state privacy law. The regulation itself requires personal data to be processed with appropriate security.
Where discipline applies, she wants the corrective action documented and paired with coaching and retraining. Telling someone what not to do is half of it; the other half is the skill that prevents a repeat. Intent counts, in her view, but HR measures impact and checks how the same policy was applied before.
Keep Policies Short Enough to Be Read
Jaime was blunt about policy length. Once a policy runs past one page, the only readers left are compliance and legal. Her fix is to decide what has to be policy, move the rest into standard operating procedures that can change as the tools do, and communicate every revision with training and a place to ask questions.
Where Employee Relations Fits
The first message in this scenario probably lands with an HR business partner, and it is about trust before it is about technology. Jaime calls a performance review a moment of connection between a leader and a team member. Whatever the finding, the manager owes this employee a repair conversation.
Rebecca added the record-keeping point. AI rules change monthly at some companies, so each case becomes precedent for the next one. A violation that repeats nine months later reads differently when the first one sits in one documented case file instead of someone’s memory.
Frequently Asked Questions About AI-Written Performance Reviews
It depends on your company’s acceptable use policy. Jaime’s view is that an approved tool can help organize notes, with names redacted. The manager still has to review the output and own every word of the assessment.
Start with the policy. Jaime pulls the AI governance and acceptable use language, the training the manager received, and signed acknowledgments before she interviews anyone. Those documents define what the manager was allowed to do.
It can be either. Jaime separates a mistake that was flagged or honestly missed, which calls for coaching, from one that was hidden or deleted, which calls for documented corrective action.
Yes, when the data identifies a person and the tool or the upload is not approved. Jaime pointed to GDPR and state privacy laws, and to the trail left when work is sent to personal devices.
By treating the review as a relationship, not a document. Jaime recommends an honest conversation with the employee, a corrected record, and feedback that reflects the work the manager has observed.
The Bottom Line for HR Leaders
Jaime Akers’s framing came back at the end of the conversation: freedom in a framework. HR rarely has every answer when a case opens, and AI cases arrive faster than policy can anticipate them. A consistent process lets a team decide reasonably on partial information and explain the decision afterward.
Her last instruction applies to managers and investigators alike. If the rule is unclear, “pause and get the answer first.”
See how AllVoices helps HR teams keep every intake, conversation, and decision in one documented trail.
