The Illusion of Training in the Age of AI

The illusion of training is the belief that giving people information will change how they behave. It’s an old problem, and twenty years of research says it doesn’t work. What has changed is that AI has now made the illusion harder to spot, because the information looks more polished than ever.

A manager I was coaching recently came to a session looking deflated. He had a difficult conversation to have with one of his team, a long-standing performance issue that had finally come to a head.

Before the meeting, he’d done what a lot of managers now do. He asked ChatGPT how to approach it, and got a thoughtful, well-structured script. After memorising the key phrases, he went into the room.

The conversation went badly. Not because the script was wrong, it was a perfectly good script. It went badly because the team member sitting across the table from him could feel that the words were out of sync with the tone, body language and facial expressions. The manager was performing feedback. He wasn’t genuinely providing it.

When we unpacked it together, something became clear. He knew what good feedback looked like. He could articulate it back to me with more fluency than he could have done a year ago. But knowing what it looks like and being able to do it are not the same. And in the age of AI that gap is getting wider, not narrower.

The old problem, in a new form

Twenty years ago, the Institute of Leadership and Management published numbers that ought to have caused a large debate in the leadership community.

Only 5% of the principles developed on Leadership and Management courses resulted in sustained improved performance back in the workplace. The average competence of UK managers was 54%. Most courses didn’t account for that, so half the material was covering ground delegates already knew. The effective impact of most management training was around 2.5%, rendering it practically useless.

These numbers are old. The picture hasn’t materially improved. If anything, the arrival of accessible AI has clarified what was always true: information is not the bottleneck. Content is now free, abundant, and on demand. A manager can get a better explanation of how to delegate, give feedback, run a meeting, or hold a difficult conversation in two minutes from ChatGPT than they would have got from a three-day course in 2005.

So why doesn’t behaviour change?

What’s actually driving behaviour

Behaviour is the only visible part of who we are. The rest sits underneath. Values, beliefs, how we see ourselves, what we think the ultimate purpose of our role is, none of this is observable. Most people find it difficult to recognise, let alone articulate. But it is what’s actually driving the behaviour, whether we’re conscious of it or not.

Robert Dilts described this in 1983 in his Logical Levels model. The levels run from environment at the bottom, the context we’re in, up through behaviour, capability, beliefs, values, identity, and finally purpose at the top.

• Purpose: the ultimate reason we are the way we are
• Identity: how we see our role, our sense of who we are being
• Values: what is most important to us
• Beliefs: what we hold to be true
• Capabilities: how we approach what we do
• Behaviour: what we do and say
• Environment: where we are, and with whom

The model still holds. What’s changed is what AI has done to the level of capability.

The capability illusion

AI is extraordinary at generating content for the capability-level above. It can give you the script, the framework, the model answer, the well-structured plan. It can produce something that looks like the output of a competent practitioner.

What it can’t do is shift you at the levels above capability. It can’t change what you believe about the person you’re about to give feedback to, it can’t change what you value about the relationship, and it can’t change how you see yourself in the role. And it has no effect on how you see the purpose of your role. Without these shifts, the script is just empty words.

This is the new form of an old problem. The manager I was coaching had done the work at the wrong level. He’d loaded up on capability content and tried to deliver it from inside a set of beliefs that hadn’t moved. He still believed, somewhere underneath, that having the conversation might damage the relationship. He still valued being liked over being respected, and still saw himself as someone who got on with people, not someone who upheld standards. The script collided with all of that and got derailed.

Why training has always struggled

The Logical Levels model explains why most training fails, and it explains why throwing more content at the problem makes it worse, not better.

The key principle is this: to change behaviour, you need a shift at the level above the one where the change needs to happen. That is an important point that is often missed. Skills don’t sustain themselves on their own. Practising new skills takes effort and motivation, and motivation comes from belief: what you hold to be true about the situation. Belief is sustained by what you value most. Values are anchored in how you see yourself. And identity is held in place by purpose.

So, if a manager has been promoted from the team and is struggling in the role, the answer isn’t another course on delegation. It’s a conversation about what they now believe their job is, what they value about it, who they need to be in this role, and why it matters. The skills follow once those higher levels are in alignment. They struggle to have impact when they aren’t.

This is why the directors who say “I pay them as a manager, they had better start acting like one” are usually disappointed. Salary is only at the environment level. It doesn’t reliably move anything above it.

What this means now

The arrival of AI has not made the inner work less important. It has made it more important, because the outer work has been so spectacularly automated.

A leader can now generate, in an afternoon, a values statement, a strategic plan, a feedback framework, a development plan for each of their direct reports, and a coaching question bank. None of it requires them to have shifted at the level of belief, value, or identity. None of it requires them to have done the inner work of becoming the kind of leader who would have written those things from lived experience.

These days their teams can tell. Not always consciously, but they can tell. The words and the person saying them are not aligned.

The development that matters now is the development AI cannot do for you. Examining the beliefs you hold about your team that are quietly limiting them. Noticing the values you defend in meetings that are actually outdated and serving an older version of you. Asking who you need to be in your current role, rather than who you were in the last one. Reconnecting with the purpose that made the work meaningful in the first place.

These are not new questions. They are the questions the Logical Levels model has always pointed to. What’s new is the urgency. The capability layer is now so cheap and abundant that the levels above it are exposed in a way they weren’t before. Leaders who haven’t done the inner work will find themselves with more polished outputs and less authority than ever.

Two questions to sit with

If there’s a behaviour you’ve been trying to change, and the change isn’t sticking, the model suggests the answer is one level above where you’ve been working. So:

• What do you believe about the situation, or the people in it, that might be quietly making the new behaviour impossible?
• And who do you need to be, not what do you need to do, for the change to take hold?

So, when it comes to leadership . . . stay curious.