We Let ChatGPT Write Our Performance Review and It Was Somehow Both More Accurate and More Insulting Than Our Actual Boss

It is performance review season, that magical time of year when your manager schedules a 30-minute Zoom call to read you a document they wrote in 11 minutes the night before, containing feedback so vague it could technically apply to any carbon-based employee at any company on earth. "Strong contributor. Room for growth in stakeholder communication. Continue to leverage existing strengths." Incredible. Deeply meaningless. A workplace haiku about nothing.

This year, something changed. Companies across the country — startups, mid-size firms, enterprises large enough to have a Vice President of Synergy — started quietly, then not so quietly, plugging employee data into AI tools and letting the algorithm take a crack at the annual review. Some are using dedicated HR software with AI baked in. Some are just copy-pasting peer feedback into ChatGPT and hitting enter. Some are doing it and not telling anyone, which is fine, totally fine, no ethical questions there.

We decided to investigate by doing what any responsible satirical publication would do: we ran the experiment ourselves, compared the outputs, and came away with feelings we will need to process in therapy.

The Setup: Human Manager vs. The Machine

For science — and we use that word loosely, the way a wellness influencer uses "neuroscience" — we fed ChatGPT a fairly standard set of inputs: job description, a list of projects completed, some peer feedback pulled from a real (anonymized) review cycle, and one (1) meeting transcript where the employee in question said "circling back" four times in eight minutes.

We asked it to write a performance review in the style of a mid-level corporate manager.

What came back was, and we cannot stress this enough, eerily accurate in all the wrong ways.

Side-by-Side: What the AI Said vs. What the Human Manager Said vs. What Was Actually True

On communication skills:

AI version: "Demonstrates consistent effort in cross-functional communication, though stakeholder alignment could benefit from increased proactive outreach in earlier project phases."

Human manager version: "Great communicator! Real team player."

What was actually true: This person sent one Slack message that derailed an entire sprint because they CCed the wrong VP and nobody wanted to say anything about it for three weeks.

The AI, somehow, got closer to the truth. It didn't name the VP incident — it doesn't know about the VP incident, the VP incident lives in the trauma center of the team's collective memory — but it identified the pattern. The human manager, meanwhile, wrote "great communicator" because writing anything more specific would require a conversation they are not emotionally prepared to have.

On project execution:

AI version: "Delivered key deliverables within established timelines on the majority of assigned projects, with some variance in Q3 that may warrant further discussion regarding prioritization frameworks."

Human manager version: "Crushing it! So much growth this year!"

What was actually true: Q3 was a disaster. Everyone knows Q3 was a disaster. Q3 has a nickname among the team that cannot be printed here.

Again: the AI gestured at the truth through the fog of corporate language. The human manager wrote "crushing it" and moved on with their life.

Why the AI Is Weirdly Better at This (And Why That's Still Bad)

Here's the uncomfortable thing: AI performance review tools are, in a narrow, technical sense, more consistent than human managers. They don't have favorites. They don't give glowing reviews to the guy who always laughs at their jokes. They don't mysteriously rate women's communication skills lower while praising identical behavior in male colleagues. They don't skip over the Q3 incident because talking about it would make the 1-on-1 awkward.

They are, in this limited way, more fair.

But — and this is the part where we need you to stay with us — "more consistent than a human with unconscious biases and a conflict-avoidance problem" is an extraordinarily low bar that should not be mistaken for good. The AI doesn't know why Q3 happened. It doesn't know that the reason the employee "showed variance in prioritization" was because leadership changed the project scope three times and then acted like none of that occurred. It doesn't know that "proactive outreach" wasn't happening because the last person who sent proactive outreach emails got quietly frozen out of the interesting projects.

The AI sees patterns in the data it's given. It has no idea what the data means, who's responsible for it, or whether the employee it's reviewing is a person who is trying very hard in a broken system or a person who is genuinely coasting. It cannot tell the difference. It doesn't have the apparatus for that distinction.

The Passive-Aggressive Uncanny Valley

One of the most fascinating — and by fascinating we mean mildly horrifying — features of AI-generated performance feedback is the tone. There's a specific register that these tools land in, something we've started calling Corporate Passive-Aggressive Neutral, that manages to sound both inoffensive and vaguely accusatory simultaneously.

"May warrant further discussion." "Could benefit from additional attention." "Opportunities exist for continued development in this area."

Every one of these phrases is technically a compliment-shaped container that has been filled with mild disappointment. It's the linguistic equivalent of your manager saying "interesting choice" about your presentation slide deck. You know something went wrong. You cannot identify what. You will think about it at 2am.

Human managers do this too, of course. But when a human does it, there's at least the possibility of a real conversation — of asking "hey, what did you actually mean by that" and getting a human response, however uncomfortable. When the AI does it, there's nobody home. You're arguing with a mirror that doesn't know your name.

What This Actually Means for Your Career (Nothing Good)

If your company is using AI to generate or assist with performance reviews — and statistically, there's a growing chance they are — here's the practical reality: you are now being evaluated by a system that is optimizing for pattern consistency rather than accurate human judgment, which was already not great, and has now been replaced with something that is consistent in its inability to understand context.

Your manager, flawed and conflict-avoidant as they may be, at least knows you exist as a specific person. The AI knows you as a dataset. The difference matters enormously when it comes to whether your "room for growth" comment reflects genuine feedback or just the algorithm noticing that you used the word "challenging" in a self-evaluation and flagging it as a sentiment indicator.

The real scandal here isn't that AI is writing performance reviews. It's that the reviews were already so empty, so carefully drained of anything specific or useful or honest, that the AI slot in almost perfectly. The machine didn't lower the bar. The bar was already on the floor. The machine just picked it up and held it there more efficiently.

Congratulations on your continued contributions. There is room for growth. Please circle back.