Every November I disappeared into performance-review purgatory. I managed twelve people, each review ate 45 to 90 minutes, and I always started strong and finished the last three at 1am on the deadline, fueled by cold coffee and resentment. The reviews I wrote at midnight were never the ones I was proud of. They were either too soft because I was tired of being precise, or weirdly harsh because I'd run out of generous language. Neither version was fair to the person reading it, and both versions were going into someone's permanent record and getting read aloud in a one-on-one where their raise was on the line.
The worst part was that I knew the reviews mattered enormously and I was producing my lowest-quality work on them precisely because there were so many. It's the cruel math of management: the more people you're responsible for, the less attention each of them gets in the exact document that's supposed to prove you were paying attention.
The bottleneck was never the thinking
Here's the thing I finally admitted to myself: I already knew how everyone had done. I had 1:1 notes, project outcomes, a year of context sitting in my head. The actual work — judging performance — was done. What was killing me was the translation step: turning a bullet like "missed the launch date but recovered well, communication still spotty under pressure" into prose a calibration panel won't tear apart and the employee will genuinely find useful. That conversion is a writing task, and writing twelve careful documents in a fortnight is just a lot of writing, no matter how good your raw material is.
I'd tried templates. Templates make every review sound like the same beige person did everything fine. I'd tried just writing faster. Writing faster is how you produce the midnight reviews I'm embarrassed by. The problem needed a different shape of help.
What I changed
I ran the Fair Performance Review Summary from Messy Manager Notes prompt through Claude Opus 4.8. For each person I pasted my notes — the unfiltered, honest version with the typos still in — plus our four company values. What came back was structured exactly the way I wished my own brain worked: accomplishments tied to evidence, growth areas framed as directions instead of accusations, values alignment, and two or three SMART goals. Opus is genuinely good at this kind of careful, nuance-preserving rewrite; the XML tags in the prompt keep it from wandering off into generic HR mush, and the instruction to think through the evidence before writing means it doesn't just reword my notes — it reasons about them.
The part I didn't expect to love was the bias check. The prompt flags any note that reads as subjective and suggests objective wording. Across my twelve reviews it caught two: one where I'd written "comes across as junior" (a vibe, not a fact) and one where I'd leaned on a personality word instead of describing the actual behavior. Both flags were right. I would not have caught either at midnight, and either one could have turned an honest review into a defensible-on-paper, indefensible-in-spirit one.
- Average drafting time per review: under 10 minutes, down from 60 to 90
- Total review season: about 3 hours including my own proofreading pass
- Zero reviews kicked back by HR for calibration issues — a first in three years
- Two bias flags I'd have shipped without thinking, rewritten before anyone saw them
The one I was scared of
The review I dreaded was for someone whose year was genuinely mixed: brilliant in two areas, concerning in two others. Those are the hardest to write because the temptation is to round to a single story — either "great with caveats" or "struggling but trying" — and neither is true. The prompt helped me hold both at once. The growth areas read like a map, not a verdict, and the accomplishments didn't feel like they were there to cushion a blow. He told me afterward it was the most useful review he'd ever gotten, and asked to keep the development goals as his quarterly plan. That was the moment I stopped feeling sheepish about using AI for this. It didn't write my opinions. It made my opinions legible.
The honest caveat
The first output for my most senior report was too flattering, because my notes for her were thin and glowing — I'd coasted on "she's great" all year. I added one line to my input: "be specific, push for at least one real growth area." The second pass was sharper and surfaced a stretch goal she actually thanked me for. The lesson is that the model only reflects the evidence you feed it; if your notes are lazy, your review will be too, just faster. So I now keep better notes throughout the year, which has quietly made me a better manager independent of the prompt.
Grab the Fair Performance Review Summary from Messy Manager Notes prompt on Prompt Dock, paste your own ugly notes into it, and watch the season shrink. I pair it with my Performance Improvement Plan Draft — Kind, Specific, and Actually Defensible prompt for the rare case where a review surfaces a problem that needs a real plan rather than just a growth area. Together they cover the whole spectrum from "thriving" to "we need to talk" without me losing a single midnight.