The Feedback Loop from Hell: Escaping AI-Generated Performance Review Cycles

It is performance review season. A manager — let's call her Sarah — has twelve direct reports and two weeks to get evaluations done alongside her actual job. Her organisation now permits using tools to draft reviews. So she does. She pastes in each person's goals, pulls some project notes, and generates twelve polished evaluations in an afternoon.

Across the floor, her report — call him Marcus — opens his review. It also reads like it could have been written about anyone, despite being thorough and full of specific examples. He has no idea what Sarah actually thinks of him, whether he's on track for a promotion, or whether that rough quarter last year even registered.

So Marcus does what more employees are doing now: he writes his self-assessment with the same tools, says the right thing, and sends it back.

Nobody said anything real. Both walked away vaguely unsatisfied and neither could name why.

This Is Already Happening

JPMorgan rolled out a chatbot for performance reviews in late 2025, letting staff draft evaluations while keeping pay decisions with managers. Citi launched something similar around the same time. The pattern is spreading.

Only 2% of CHROs think their performance management system actually works, per Gallup. In Deloitte's 2025 Global Human Capital Trends report, 61% of managers and 72% of workers could not say they trust the process. Those numbers came before tool-assisted reviews became standard practice. They are not going to improve by themselves.

There is also something worth sitting with. Employees are getting good at spotting generated language in their evaluations. When someone suspects their review was drafted by a chatbot, it undermines the entire premise of the exercise.

What Gets Left Out

These tools are useful. They structure observations, cut admin time, and surface things you might otherwise miss. None of that is nothing.

What they cannot do is see the moment someone stepped up when nobody asked them to. They have no way of knowing a colleague's confidence has quietly shifted over six months — which matters more than any project metric.

They also struggle with something newer and more awkward: telling the difference between a person who has genuinely developed and one who has gotten better at using tools. Some managers are penalising AI usage in self-assessments. Others are rewarding it. Nobody has agreed on what they're actually measuring.

The data shows the gap clearly. Only 30% of employees strongly agree that someone at work encourages their growth, down from 36% in 2020. Only 46% clearly know what is expected of them, down from 56% in the same period.

A tool working from goals and project notes cannot close those gaps. A manager who pays attention can.

What Sarah and Marcus Could Have Done

Sarah's tool was not the problem. Using it instead of her own observation was.

A better version of that afternoon: she uses the tool to organise her notes and catch what she forgot to include. Then she reads what came back and asks one question — does this sound like someone I actually watched do work this year? Whatever does not pass that test gets rewritten in her own words.

For Marcus, the same logic runs the other way. His self-assessment is the clearest shot he has at putting his own account of the year on the record.

75% of employees support tool-assisted reviews as long as a human makes the final call, but that principle applies just as much to the person being reviewed. It is fine to outsource the structure, but the honest account of what actually happened has to come from the person who was there.

Review cycles in too many organisations have quietly become compliance exercises. These tools did not start that, but they will accelerate it if nobody draws a line. Drawing the line does not mean rejecting the tools. It means staying in the room and saying something real while you're there.

Quick Win This Week

If a review is coming up, try this before opening anything: write three sentences in your own words about what this period actually felt like.

• What surprised you.
• What you learnt.
• What you'd do differently.
• Those are the backbone. Let the tools handle the structure around them.

P.S. If you've been on the receiving end of a review that felt machine-shaped, hit reply and tell me what tipped you off. I'm collecting these patterns.

Close the gap and amplify what makes you human.

Start Your Human Skills Journey.

Sign Up

Email sent! Check your inbox to complete your signup.

No spam. Unsubscribe anytime.