Feedback Frameworks Made Your Managers Worse
The most dangerous manager in your organisation is the one who has been on the feedback course.
Not the one who avoids feedback entirely — you already know about them. They're a problem, but they're a visible one. The dangerous one is the manager who believes they're good at it. The one who learned SBI on a Tuesday, delivered it on a Wednesday, and has been ticking the box ever since — convinced that the conversation happened because the model was followed.
It wasn't a conversation. It was a recitation. And the person on the other side knew the difference immediately.
The framework becomes the performance
Here is what typically happens when an organisation decides to get serious about feedback culture. A framework is selected. Workshops are designed. Managers attend. They learn the model. They practise it — briefly, in a room, often by reading from a card. They leave with a handout and the belief that they now possess a skill.
What they actually possess is a script.
And the script works beautifully — right up until the moment the other person does something the script didn't prepare them for. The eyes that fill with tears. The defensive silence. The person who says, calmly, "I don't agree with any of that." The long-serving team member who looks at them and says, "After everything I've done for this team, this is what I get?"
The manager trained in SBI knows the next step is to describe the impact. But they don't know what to do when the impact they're describing has just landed on a person who is now sitting across from them in visible pain, and the room has gone very quiet, and the model has nothing to say about this part.
So they do what most people do. They rush. They soften. They over-explain. They fill the silence with words that undo the message they just delivered. They leave the room believing the conversation happened. The other person leaves the room believing it didn't.
The framework was followed. The feedback was lost.
Rachel delivers the model
Rachel is a team leader in operations at a mid-sized financial services firm. She's been managing people for four years. Last year, she attended a two-day feedback skills programme — well-designed, well-facilitated, well-received. She learned a model. She practised it once, in a pair exercise with a colleague named David, who played the role of an underperforming team member with the enthusiasm of someone waiting for lunch.
Three weeks later, Rachel had to deliver the real thing. Not to David. To Priya.
Priya had been on the team for seven years. Longer than Rachel. She was well-liked, reliable in a crisis, and had been coasting for the better part of a year. The quality of her case reviews had slipped. Her notes were incomplete. Two client complaints had been traced back to her work in the last quarter. Rachel's manager had asked her to address it.
Rachel prepared. She used the model. She wrote down the situation, the behaviour, and the impact. She rehearsed the opening in her head on the drive to work.
The opening went fine. "Priya, I wanted to talk to you about some of the case reviews from the last quarter. I've noticed that several have had incomplete notes, and two have led to client complaints. The impact is that the team is having to pick up the rework, and it's affecting our turnaround times."
Textbook. Clean. The model would be proud.
Then Priya said: "That's not fair. You know I carried this team when we were short-staffed last year. No one said anything then."
And Rachel had nothing. Not because she didn't know what to say in theory — but because no one had ever helped her practise the moment after the model ends. The moment when the other person pushes back, or gets hurt, or names something true and inconvenient. The moment the conversation actually begins.
Rachel softened. She acknowledged the short-staffing. She thanked Priya for her past contribution. She said "I just want to make sure we're on the same page going forward." She left the room feeling she'd handled it well.
Priya left the room believing nothing needed to change.
And now AI is making the same mistake faster
Here is the part that should worry anyone who cares about this: AI, used badly, accelerates exactly the same problem.
A manager opens ChatGPT and types: "Write feedback for a team member whose work quality has declined, using Radical Candour." In seconds, they have a beautifully structured paragraph. Clear. Empathetic. Balanced. The situation is named, the impact is described, the care is signalled. It reads like something a skilled manager would say.
It is also completely useless — because the paragraph is not the hard part.
The hard part was never composing the message. A reasonably thoughtful person can sit at their desk and write down what they need to say. The hard part is what happens when they say it out loud, to a real person, and that person reacts in a way the paragraph didn't anticipate. Priya doesn't respond to the written version. She responds to the live one — and the live one is where the skill either exists or it doesn't.
AI-generated feedback scripts are the framework problem with a turbo engine. They give managers even more polished language, even more confidence in their preparation, and even less awareness that the preparation has not touched the actual skill. The manager walks into the room with a beautifully crafted message and no ability to handle what comes after they deliver it.
This is not a minor distinction. It is the whole distinction.
The structure of the feedback matters. Of course it does. Clarity matters, specificity matters, care in how the message is framed matters. But the structure is the opening act. The real performance begins the moment the other person responds — and responses do not follow scripts.
What if the person is confused and asks you to explain what you mean, and your beautifully prepared paragraph suddenly has to become an improvised, honest, specific answer? What if they disagree — not rudely, but firmly and articulately — and you have to hold your position without becoming defensive? What if they go quiet, and you're sitting in a silence that feels endless, and every instinct you have is screaming at you to fill it with reassurance? What if they simply nod, say "fine," and you know — you can feel it — that they haven't heard a word, and you have to decide in that moment whether to let it go or press further?
No generated paragraph prepares you for any of that. No framework does either. The only thing that prepares you is having been there before — not once, but many times, with many different people responding in many different ways, until your nervous system has learned that the discomfort after delivery is not a signal to retreat. It is the signal that the conversation has finally started.
The part nobody teaches
The problem with feedback frameworks is not that they're wrong. SBI is a perfectly reasonable structure for organising a message. Radical Candour's two-axis model is a useful diagnostic. The feedback sandwich — well, the feedback sandwich is genuinely bad, but that's a separate matter.
The problem is that the framework covers the first forty-five seconds of a conversation that lasts fifteen minutes. And the other fourteen minutes and fifteen seconds are where the skill actually lives.
What happens after you deliver the message? That is the conversation. Not the opening. Not the model. The bit after — where the other person reacts, and you have to stay in the room, and adjust, and hold the line without being cruel, and make space for their response without abandoning yours.
That part requires something a framework cannot provide. It requires practice. Not one paired exercise in a workshop with a colleague who half-commits to the role. Repeated, private, pressurised practice — the kind where you get to feel the silence and learn to stay in it. Where you deliver the message, hear the pushback, and discover whether you can hold your ground or whether you instinctively retreat. Where you find out what you actually do when someone's face changes — not what you think you'd do, not what the model says you should do, but what you do.
That discovery is the beginning of real skill. And it almost never happens in a workshop.
The fluency trap
There is a specific kind of incompetence that only training can produce. It's the incompetence of someone who has the language but not the capability — who can describe what good feedback looks like but cannot deliver it when the room gets tense.
This is worse than having no framework at all. The untrained manager who avoids giving feedback at least knows they're avoiding it. They might seek help. They might ask someone else to have the conversation. They might, in a moment of honesty, admit they don't know how.
The manager who has completed the course believes they do know how. They have evidence — a certificate, a model, a memory of a paired exercise that went well. And now, possibly, a perfectly worded paragraph generated by AI that they've read three times and mistaken for readiness. They are inoculated against the realisation that they're still not able to do the thing. The framework — or the generated script — has given them the confidence of competence without the competence itself.
This is the fluency trap. Fluency in the language of feedback is not the same as skill in the act of it. And every tool that makes the language easier to produce without making the act easier to perform deepens the trap. Organisations that invest heavily in frameworks, and managers who lean on AI to polish their words, are producing ever-more-articulate people who still cannot sit in silence after saying something hard.
The feedback culture they designed on paper doesn't exist in reality. Not because the words are wrong. Because the skill was never built.
What repetition actually builds
The skill of giving feedback that lands — genuinely lands, not just gets delivered — is not a knowledge skill. It is a performance skill. And performance skills are built the way every other performance skill is built: through repetition under conditions that approximate the real thing.
A musician does not learn to perform under pressure by studying music theory. They learn by performing, badly, many times, until the thing that once required all their conscious attention becomes something their hands and voice can do while the rest of them stays present and responsive. A surgeon does not learn to operate in a crisis by reading about crisis management. They learn by being in simulations where the crisis unfolds and they have to respond — not once, but dozens of times — until their response is reliable.
Feedback is no different. The skill is not in the structure of the message. It is in the manager's ability to stay composed when the message lands badly. To sit in silence for five seconds — which feels like thirty — after saying something the other person didn't want to hear. To hear a defensive response and neither capitulate nor escalate. To adjust their tone without adjusting their message. To recognise when the other person needs a moment, and to give it to them without filling it with nervous reassurance.
These are not things that can be taught in a model. They are things that emerge from doing — from delivering feedback to many different people and navigating many different responses. The person who gets confused and needs you to be clearer than you were. The person who disagrees and needs you to hold your ground without turning it into an argument. The person who goes quiet and needs you to let the quiet sit there without rushing to rescue them from it. The person who deflects, who gets angry, who cries, who laughs it off, who turns it back on you.
Each of those responses demands something different from the person delivering the feedback. And the only way to develop that range — that flexibility under pressure — is to have encountered each one enough times that it stops being a surprise and starts being something you can navigate.
This is what AI roleplay makes possible in a way that nothing else has. Not the polished paragraph — forget the polished paragraph. The conversation. The full, messy, unpredictable conversation where you deliver the message and then the other person reacts and you have to respond, in real time, without a script. And then you do it again. With a different reaction. And again. With a person who cries. And again. With a person who pushes back hard. And again. With a person who says nothing at all.
No audience. No colleague watching you freeze. No facilitator whose assessment you're managing alongside the conversation itself. Just the conversation, repeated, until the thing that felt impossible starts to feel manageable.
That is not a technology story. It is a practice story. The technology is the mechanism. The outcome is a manager who can stay in the room.
The person Priya deserved
Go back to that room. Priya, seven years on the team, being told her work has slipped. She deserved a manager who could deliver that message clearly, hear her response fully, and hold both truths at once — her years of loyalty and her recent decline in quality. She deserved someone who had practised not just the opening, but the whole conversation. Someone who had heard "that's not fair" before — not from Priya, but in practice — and had learned how to respond without retreating.
She didn't get that. She got a manager who followed a framework, lost her footing at the first moment of resistance, and defaulted to reassurance. The message was lost. The performance continued to decline. Three months later, Rachel had to have the same conversation again — except now it was worse, because Priya believed the issue had already been resolved.
That second conversation is the real cost of the fluency trap. Not just the first one that failed — but the harder one that followed because the first one wasn't real.
The question underneath the question
Organisations love to ask: have our managers been trained on feedback? It is the wrong question. The right question — the uncomfortable one — is: can they do it?
Can they sit across from someone who has just heard something painful and stay present? Can they hold their message when the other person pushes back? Can they tell a high performer that one aspect of their behaviour is damaging the team, without the high performer's value making them lose their nerve? Can they do it not once, in a workshop, with a colleague who is half-acting — but reliably, repeatedly, when it counts?
If the answer is no, then the framework didn't fail. It did exactly what it was designed to do: it transferred knowledge. And the AI-generated script didn't fail either. It did exactly what it was asked to do: it produced better words. Neither of them was ever going to close the gap — because the gap was never about the words.
The gap is between knowing what to say and being able to say it when the room goes quiet and the other person is looking at you and the model has run out of steps and the paragraph you prepared sounds nothing like your actual voice and all that's left is you, in the silence, deciding whether to hold your ground or retreat.
That gap is not closed by a better framework. It is not closed by better-generated language. It is closed by practice — private, repeated, unglamorous practice against the full range of human responses, the kind that builds the muscle memory of composure, the kind that teaches a person what they actually do under pressure and gives them the chance to do it differently.
The question worth sitting with is not whether your managers have been taught to give feedback. It is not whether they can produce a well-structured message. It is whether they have ever practised the part that comes after they give it — the part where the other person responds, and the script ends, and the real conversation begins.
Because that is where feedback lives or dies. And right now, almost nobody is practising it.
About Real Talk
Real Talk is AI-powered roleplay for the conversations that matter most. Not scripts. Not frameworks. Practice — private, repeated, and built around the moment the other person responds and the real conversation begins.
If this article described a problem you recognise, Real Talk is where your managers start solving it.
