A phrase has been spreading through the AI-for-coaches market over the last few months, and it sounds so reasonable that almost nobody has examined it. The phrase is: train AI on your intellectual property.
I think it’s important to examine it, because the definition of that middle term determines everything about what such a service can and cannot do, and the definition is not ambiguous.
What the term actually means
Intellectual property is a legal category. In every jurisdiction that recognises copyright, the qualifying condition is fixation in a tangible medium of expression. An idea in your head is not intellectual property. The same idea written down, recorded, or otherwise fixed becomes protectable.
So intellectual property is, by definition, the portion of your expertise that has already been made explicit. Your book. Your recorded programme. Your named framework with its defined stages. Your workbooks, your slide decks, your course modules.
That is a real and valuable body of material. It is also, for anyone with twenty years of practice behind them, the smaller half of what they know.
The two methods every experienced practitioner has
Every established practitioner I have worked with has two methods, and only one of them has been written down.
The written method is what appears on the programme page. Five stages, or seven principles, or three phases. It is honest work and it was assembled retrospectively, by looking at the practice and describing it in language. Assembling it was a genuine act of clarification and most practitioners feel better for having done it.
The unwritten method is the set of real-time judgements that determine what actually happens with a client. When to interrupt and when to wait. Which of three apparently identical presentations needs a completely different approach and why. The moment you recognise that the stated problem is functioning as cover for a different problem, and the specific cue that told you.
That second method was never assembled. It accumulated, session by session, over thousands of hours, and it operates below the level at which it narrates itself. Ask a highly experienced practitioner why they did what they did in a particular moment and you will often get an honest and unsatisfying answer: it seemed like the thing to do.
Which is why they are not lying when they say the framework describes their work, and also why the framework does not contain the thing that makes the work effective.
Why this matters commercially rather than philosophically
For most of the history of this profession, the gap between the two methods did not cost anyone anything. Practitioners delivered personally. Their competence travelled with them into the room. Nobody needed a specification.
Two things have changed that.
The first is the emergence of tools that can, in principle, carry a method. A model can execute a described process with consistency. What it cannot do is infer a process that has never been described. Fed the articulated framework, it produces a competent version of the articulated framework, and the practitioner reads the output and thinks: this sounds like my marketing rather than my work. They are correct.
The second is that the articulated layer is precisely the layer that is being commoditised fastest. Frameworks that have been published are now trivially available to any client with a subscription. The 2025 ICF Global Coaching Study records 122,974 practitioners globally, a 15 per cent increase since 2023, in a profession generating around $5.34 billion. In a market that crowded, differentiation cannot rest on the part of your expertise that has already been written down and can be read by anyone.
Extraction is a different discipline from training
Here is the distinction I think the market has not yet made, and it matters more than any technical question.
Training means taking existing material and using it to condition a system’s outputs. It is well understood, increasingly cheap, and it works.
Extraction means surfacing decision logic that has never existed in described form, from a practitioner who cannot narrate it on request, through structured interrogation designed to reach material below conscious access. It is slower, it is harder, and there is currently very little of it happening in this market.
You cannot train a model on a document that does not exist. So extraction is not an optional refinement that comes after the technical work. It is the precondition for the technical work meaning anything.
This is the reason Magic to Method exists as a framework within the AI Augmented Practitioner programme. It starts before the folder of existing content, because the folder is the part you were already conscious of.
The test you can run this week
If you want to know which half of your expertise you are working with, try this.
Take your most recent difficult session. Write down what you did at the three points where the session could have gone in different directions. Now write down why you chose as you did, in terms specific enough that another practitioner could apply the same reasoning to a different client.
Most people find the first part straightforward and the second part surprisingly difficult. That difficulty is not a deficiency. It is the signature of genuine expertise, which becomes automatic precisely because it has been practised past the point of conscious attention.
It is also the exact material that no upload will ever capture, and the exact material worth the most.
