The 2025 ICF Global Coaching Study contains two numbers that should be read side by side, because separately they are unremarkable and together they describe something structural.
The first: 54 per cent of coaches globally name improved platforms and technology-driven solutions as a priority for meeting future client demand.
The second: 6 per cent report currently using AI-powered coaching tools or chatbots.
That is a gap of roughly 48 points between what practitioners say matters and what practitioners are doing. In a profession of 122,974 practitioners worldwide, generating around $5.34 billion and growing 15 per cent in practitioner numbers since 2023, that gap represents an enormous quantity of stated intent producing no action.
The usual explanations do not survive contact with the numbers
Ask why the gap exists and you will get four answers: no time, no budget, not technical enough, ethical concerns.
Each is real for some practitioners and none of them accounts for a gap this size.
This is a profession that adopted video delivery almost universally inside about eighteen months. It absorbed online scheduling, digital assessment tools, payment platforms, membership sites, email automation. Practitioners who taught themselves three different course platforms are not being defeated by an interface.
Nor is it cost. Entry-level AI tooling now costs less per month than most practitioners spend on their scheduling software.
Ethical concern is the most substantial of the four and it is genuine, particularly among therapeutic practitioners. But ethical concern produces careful, bounded adoption. It does not produce a 48-point gap between priority and practice.
Something else is happening, and I think it is structural rather than attitudinal.
Two entirely different things called AI adoption
A Forbes piece in early July reported research finding roughly a 130 per cent valuation gap between businesses that embed AI into the products they sell and businesses that use AI only for internal productivity. The finding was aimed at technology companies and it circulated widely as general business advice: stop using AI for admin, start building it into what you sell.
For practitioners, that advice is close to useless, and understanding why explains the whole adoption gap.
Internal use is straightforward. Drafting emails. Summarising session notes. Producing marketing copy. Reorganising a workshop outline. Practitioners do all of this now, and they do it easily, because every one of those tasks was already explicit. You know what a good client email looks like, so you can brief a system to produce one.
Embedded use means something categorically different. It means the AI carries part of your method and delivers value directly to a client. Between a client session, or inside a programme, or as a diagnostic that reflects your specific way of reading a situation.
To do that, the method has to exist in describable form. Steps. Decision points. The conditions under which you do one thing rather than another. Not a philosophy, not a set of values, not five stages retrospectively named for a sales page. An operative description.
Most established practitioners have never produced one.
The adoption gap is a specification gap
That, I think, is the actual explanation. The 48-point gap is not a technology adoption problem. It is a documentation problem, and it has gone unnamed because nothing before now required practitioners to solve it.
For the entire history of the profession, competence has been enough. You developed judgement over years, you carried it into the room, and you delivered personally. Nobody ever asked you to specify how you decide what to do, because there was no downstream system that needed the specification.
Now there is, and the profession has discovered that its core asset was never written down.
The research literature supports this reading. Work published on arXiv in 2026 on tacit knowledge extraction identifies the same structure across professional domains: expert execution depends on implicit assumptions, contextual constraints and experience-based judgements that experts rely on but rarely document. Related work on AI-enabled tacit knowledge frameworks makes the point that this material does not surface through simple questioning, because the expert genuinely cannot access it on request. It requires structured interrogation designed for the purpose.
So the practitioners best positioned to build something genuinely differentiated, the ones with twenty years of accumulated pattern recognition, are also the ones furthest from being able to, because their expertise is the most deeply automated and therefore the least accessible.
What changes if you accept this
Three things follow, and they are practical rather than theoretical.
First, stop treating your current AI use as a measure of your progress. Using AI for admin tells you nothing about whether you could embed it in your work, because those are different problems with different bottlenecks.
Second, stop collecting prompts. A prompt library is a substitute for a specification and it does not become one however large it grows. If the underlying method is undescribed, more prompts produce more variations of generic output.
Third, treat extraction as its own piece of work with its own timeline. Surfacing how you actually decide, in a form precise enough that a system could execute it, takes structured effort over weeks. It is genuinely difficult work and it is also the only work that unlocks everything after it.
The honest position
I do not think practitioners are behind on AI. I think most of them have been handed step three and told it was step one.
The instruction has been: gather your content, upload it, build your tool. And the practitioners who followed it produced something competent and generic, felt disappointed, and quietly concluded the technology was overhyped for depth work.
The technology was fine. The sequence was wrong. Extraction comes first, then structure, then anything technical. Reverse those and the result is predictable, which is roughly what one in sixteen adoption looks like across an entire profession.
