Stop hollow performance reviews when AI writes the scorecard.
Stop hollow performance reviews when most scorecards are ChatGPT-generated, so feedback stays authentic and continuous, not AI theater employees can smell.
Eleven reviews are due Friday. The manager pastes last year's bullets into ChatGPT, gets a clean paragraph about ownership and collaboration, and drops it in the form. The person who did the work reads it and hears nobody. Across the cycle, most scorecards arrive the same way. The meeting still happens. The feedback is already gone.
Stay on the hollow draft. This is not an Adapt rollout. This is not a PII wall. This is not a missing AI strategy, a manager change-layer design, or a tool bakeoff. The job is to keep feedback authentic and continuous when the scorecard itself became theater.
Employees do not need a detector. They need one sentence that could only have been written by someone who was there.
Friday, eleven reviews, one prompt.
Operator pattern, not a Boon benchmark. In some review cycles, about four in five scorecards arrive as a ChatGPT draft. That is a scar, not a target, and not an industry study. We will not dress it as ROI. The form still asked for prose. The calendar still asked for volume. The prompt did the only job the system rewarded.
A blank AI-fluency row on the same form is a different failure (why AI competency on the review form fails). Here the row is filled. The voice is just not the manager's. Practice with AI can belong between human conversations (how to practice difficult conversations with AI). Using it to finish the review is how you skip the conversation you owed.
Hanging on to last year's language.
Operator view: hanging-on mentality is what a yearly object produces. People cling to last cycle's adjectives because the form rewards a finished paragraph more than a live observation. Combat that by shortening the cycle, not by buying a nicer prompt library. Annual to quarterly is the counter. Continuous feedback is how the quarter stays honest.
Always-on without a split between talk and score will recreate the same theater every week (always-on feedback without turning every conversation into a rating). Managers who already will not have the conversation will not start because a model wrote it for them (managers who will not have the conversation). A 1:1 that never names the work is the same hang in miniature (weekly 1:1 with no tool question).
Not a tool ban. Recommendation, not a policy: you can let people outline. You cannot let a generated paragraph stand in for having seen the work. A ban without a cadence just moves the paste to a personal account.
Employees do not need a detector.
The tell is not the grammar. It is that the scorecard names none of the Tuesday work. The launch that slipped. The customer who waited. The peer who covered. People who lived that week do not run a watermark check. They read a stranger's paragraph and decide the review is a ritual.
Once they decide that, they stop bringing the truth into the room. Coaching next to that cycle has to stay growth work, not a ratings back-channel (integrating coaching with the performance cycle). A manager who cannot say the sentence out loud yet is a coaching candidate, not a better prompt user. GROW is the room where they practice the conversation they keep outsourcing to the form.
What this is not.
| If you are actually deciding | Go here instead |
|---|---|
| AI means something different to everyone | Stop thousand-flowers AI experiments |
| Legal blocked AI on people files | Lead people work when legal blocks AI on PII |
| Leadership committed, managers must own team AI adoption | Make managers the AI change layer |
| Which AI coaching product to buy | AI coaching: what it is and where it fits |
Shared language is thousand-flowers AI experiments. A data wall is the PII wall. Manager-owned adoption is managers as the AI change layer. A product comparison is AI coaching. This page will not refuse an AI coach, sell Adapt, or rank tools. Come back here when the scorecard is the hollow object.
Quarterly talk, not a prettier annual.
One sentence only a witness could write
Before the form locks, each scorecard needs a line that names Tuesday work. If a stranger with last year's bullets could have generated the rest, the review is hollow. The sentence is the test. The paragraph is not.
Move the talk off the annual
Operator pattern: hanging-on mentality lives in a yearly object. Shorten the cycle. Quarterly conversations, plus continuous feedback in the 1:1 you already claim to run. The annual can still rate. It cannot be the only time someone hears the truth.
Use 360 when the work is actually cross-functional
A 360 is a counter to one generated narrator, not a PDF ritual. Match scope to the job. A new manager with three reports does not need twelve raters. A cross-functional lead might.
A 360 is only useful when the work is actually shared (match 180 or 360 to the job). A coaching culture is managers who can say the sentence without a prompt (manager as coach). If the only change is a longer generated paragraph, you kept the theater (signs your managers need coaching).
What good looks like. A person can point to one line that named their Tuesday. The next conversation is already on the calendar, not next December. If the only upgrade is a better prompt, you still shipped theater.
Keep the review human when the form wants a draft.
Bring the cycle, the volume, and a hollow scorecard. We will map a witness sentence and a shorter cadence. We will not sell you a detector.
Book a strategy callFrequently asked questions
What do you do when ChatGPT writes most performance scorecards?
Stop treating a polished paragraph as a review. Require one sentence that could only have been written by someone who saw the work, then move the real talk off the annual form and into a cadence people already run. A generated scorecard is theater. Employees can smell it.
Is this the same as always-on feedback without killing the conversation?
Close, different job. Always-on is the regime you run instead of an empty annual. This page is the hollow draft: the form still exists, and most of it arrived from a prompt. Fix the draft first or always-on becomes a year of generated mini-reviews.
Should we just ban AI on performance reviews?
A ban is a policy, not a review. People will still paste last year into a prompt if the only artifact you reward is a finished paragraph by Friday. The job is authentic feedback and a shorter cycle, not a tool prohibition.
Is this an AI adoption or Adapt problem?
No. Shared language, a PII wall, managers as the AI change layer, and a coaching-led change program are different pages. This page stays on the scorecard people can smell.