Insight
Coaching and AI
When the answers are free
Two years ago the conversation about AI and coaching had a settled, comforting shape. Artificial intelligence would help around the edges - the scheduling, the note-taking, the chance to rehearse a difficult conversation - but the real work was safe. Coaching was a human relationship, and a machine could not do relationship. The only open question, asked with some confidence that the answer was no, was whether anyone would really want to be coached by software at all.
I want to describe what has happened since, because the ground has moved further and faster than almost anyone expected, and in a stranger direction.
Author: Simon Coops
The research that unsettled everyone
Start with the evidence, because it is more interesting than the reassuring story allowed. A few years ago, Erik de Haan, Nicky Terblanche and colleagues - among the most respected researchers in the field - ran a controlled trial comparing an AI coaching chatbot with human coaches. Over an extended engagement, with a sample of students, they found something many had not expected: the AI coach was as effective as the human coaches at helping people reach their goals. The finding travelled quickly, and for good reason. It suggested that the part of coaching built around setting a goal and working steadily towards it - structured, methodical, model-driven - was something software could do perfectly well.
What is less often reported is what happened when the same researchers tested the finding again, this time with senior leaders rather than students. In a randomised controlled trial published in 2026, they put a human coach, an AI chatbot coach and a do-nothing control group side by side inside a global company. This time the result reversed. The leaders coached by humans improved across almost every measure, with substantial effect sizes. The leaders given the AI coach barely moved at all - on most measures they were indistinguishable from the group that received no coaching. People abandoned the chatbot roughly ten times more often than they dropped the human coaching, and several who walked away from the machine asked, specifically, for a person instead.
So the picture is not “AI works” or “AI fails.” It is more precise, and more useful, than that. Where the task is genuinely just a task - a clear goal, a defined process, someone motivated to follow it - a machine can carry it. Where the work depends on the relationship itself, on presence and judgement and being met by another human being who is paying real attention, the machine has, so far, not been able to stand in. The first study found the floor of what coaching can be reduced to. The second found the ceiling it cannot reach.
So the consensus did not fall apart because the technology failed, nor because it triumphed. It fell apart because the truth turned out to be conditional. And in becoming conditional, it exposed something the profession had been reluctant to say aloud: that a good deal of what is sold as coaching was always fairly mechanical, and the mechanical part is precisely the part a machine does well. The coaching that follows a model and marches a client through a fixed process is the part now under real pressure. What survives is harder to automate and harder to fake - presence, judgement, and the ability to sit with someone in difficulty and help them think.
The coach that became something else
And then the genuinely surprising part. Just as the AI coach proved it could hold its own on the narrow ground where it works, it began to stop being a coach at all.
The clearest example is the market leader. Its AI coach is now deployed across close to a hundred of the world’s largest companies, with millions of conversations behind it - and the company has been quietly redrawing what the product actually is. It now describes it less as a coach and more as a talent platform: a system that sits in the flow of work, holds the organisation’s development data, and gives HR a view across the entire workforce of what people are working on and where they are stuck. The coach has become a repository. I have heard it said - and while I cannot stand the claim up word for word, it rings true - that the company no longer regards other coaching firms as its real competition, but something closer to LinkedIn. A place you live in, rather than a person you go to see.
Nobody set out to build that. It is mutating in real time, from a chatbot that might help you rehearse a tricky conversation into something that wants to be the professional infrastructure of an entire organisation. Which is the honest reason I will not offer you a confident prediction about where it ends. Anyone who does is guessing.
The one thing I am sure of
Here is what I think the research is really pointing at, underneath the effect sizes and the platform strategy. The more universal AI becomes, the more it pushes all of us towards acting on instant answers. Ask, receive, move on. The answers are free now, and they arrive before you have finished forming the question. That is genuinely useful, and it is also quietly corrosive, because the one thing it removes is the pause - the slow, awkward, productive business of working out what you actually think.
Coaching, at its best, exists to protect that pause. It is the place where a person is not handed an answer but helped to find the one they are ready for. As the answers get cheaper, that becomes rarer, and rarer things become more valuable, not less. So I do not know what the platforms will turn into, or how good the machines will get. But I am confident there will remain a deep human want for the real thing: another person, fully present, helping you think. That want is not a market that AI is poised to take. It is the part of us the free answers cannot reach.