The Four Positions Every Coach Has to Choose Between

The largest mental-health intervention in human history has no adult in the room. Two questions decide any AI in your practice, and they sort every platform into four positions. Here is how to tell which one you are being sold, and the one that keeps you the coach.

1.1 billion conversations, no coach

By May 2026, around 1.1 billion people sit in a weekly conversation with ChatGPT. About 420 million of those conversations are emotional support, personal reflection, or companionship. The number grows seven to ten percent every month. No coach, mentor, or therapist is in any of those rooms. The relationships people are forming with AI are the most under-supervised category in modern life.

The coaching profession is one of the few whose entire purpose is to be the adult in the room. Almost every senior coach I speak to is sitting on the same question. What happens between our sessions when I am not there. Who reads it. Who signs off. Who is responsible when it goes wrong. The research is unsentimental: 80 percent of behaviour change fails between sessions (Passmore, February 2026).

CoachHub has run more than 50,000 AI-only coaching conversations across its enterprise base. BetterUp is shipping AI coaching bots built to do the same work as the human coaches in its network. The big enterprise platforms have already picked their position. The four positions below are how you pick yours.

The four positions

Two questions decide any AI in your practice. Who sets the rules, and who reads the logs. Four positions fall out. Most coaching AI on the market today sits in one of the middle two, and the vendor language usually hides which one.

Approve

The coach reads and approves every output before it reaches the client. Nothing moves without that signature. Highest accountability, simplest legal position, the only honest place to claim full supervision. Does not work in real time.

Architect

One private model per client, trained on the coach's own sessions, running inside rules the coach has signed off on. The coach reads every conversation in their own time. Supervision lives in the architecture and the review loop, not every keystroke. You sign off on the architecture. You do not build it. Makes continuous coach support possible after the formal engagement closes, without putting the coach on call.

Automate

The AI runs inside guardrails set by the vendor, not the coach. Two flavours, one position.

The first is generic AI wrapped in a “coaching” system prompt and sold direct to companies. BetterUp and CoachHub both ship versions. The chatbot speaks to the coachee. The human coaches in the network are being designed out of the loop. Enterprise buyers want fewer billable coach hours per engagement and more bot hours, because bot hours are higher margin. As the bot gets better, the coach gets thinner.

The second is the “clone yourself” tool. Coachvox, Delphi, Marshall Goldsmith's the most publicly visible. The vendor builds a chatbot trained on a static upload of the coach's old content. Sounds like Architect, reality is Automate. The training set is frozen the day it goes live, the same clone serves every client identically, and value flows to the named coach's brand, not the user's actual situation.

Abandon

No coach in the loop. The client opens a generic chatbot, pastes their situation, takes the response at face value. The 420 million conversations sit here.

Where the value flows

Follow the money. In Approve and Architect, the coach captures the value. The AI is leverage on the coach's existing work. The relationship, the methodology, the renewals all flow to the coach first, the platform second. In Automate, the platform captures it. The bot generates conversations at zero marginal cost. The coach does not. In the clone-yourself version, the value feeds the cloned coach's brand, not the leadership situation the user is actually navigating. If the platform gets richer as the AI gets better and the coach does not, the model has been built around the platform.

ICF and EMCC have published principles, and in May 2026 both called from their annual stages for the profession to engage with AI, but neither has set a compliance standard. The EU AI Act, entering force in stages through 2026 and 2027, makes human oversight a regulatory question, not a marketing one. Architect holds. Automate does not. Keeps you the coach.

Questions to ask any AI platform pitching you

1. Who chose the rules the AI is operating under, you or the vendor? If the vendor, you are looking at Automate with a coach-shaped sticker.
2. Can you read every conversation between the AI and your client after it happens? If not, supervision is being claimed but not delivered.
3. Is the AI working from a one-time upload of your old content, or from live capture of the coaching you are doing now? A static clone of who you were eighteen months ago is not the same as an AI shaped by your current practice.
4. What happens if the AI gives a wrong response? Can you read the conversation, edit it, and step in?

The Architect Test

Architect is the hardest position to build honestly. Call this the Architect Test. Run it against any vendor that uses the word “supervised”. All five lines must be true, or it is not Architect. It is Automate dressed in better marketing.

1. Trained on the coach's own sessions, not the internet. If the foundation is generic web data with the coach's content sprinkled in, it is not the coach's AI. It is a search engine wearing the coach's name.
2. The coach's methodology, voice, and rules govern what the AI does. Not the vendor's content moderation team.
3. Every conversation is reviewable and editable. Not just discoverable in a log buried five clicks deep. Surfaced, visible, actionable.
4. Pausable, exportable, deletable at any time. If the coach cannot pull their data and walk away within 48 hours, the vendor controls the product, not the coach.
5. Learning from the work the coach is doing now. Not frozen on an upload from two years ago. Sessions, edits, and reviews shape the model continuously.

The category has a name we use for it: Coaching-grade AI. From our research so far, CoachNova is the only product we have found on this map that meets all five conditions.

The economics of coaching that compounds

Take a senior coach with ten active clients on six-month engagements. Session pricing varies by market and seniority.

In Automate, the client either churns to a vendor-run chatbot or the engagement finishes and the relationship sleeps. The platform may keep monetising the client. The coach does not capture either way.

In Architect, two new layers open without changing the work in the room.

The first is the Compound layer. Graduating clients extend into ongoing access on the coach's own infrastructure, at a fraction of session economics, the coach reading and editing the conversations in their own time. You set the price. Coaches on the platform charge between €25 and €250 a month. At €100 a month, ten extended clients add €12,000 a year of recurring revenue per cohort. Stack three years of graduating cohorts and the recurring book passes €36,000 a year on top of new engagements.

The second is the Hybrid layer. A new client base the current packaging cannot reach. Start with three sessions, then a monthly subscription with coach-supervised messaging in between. Always the option to offer a session when the conversation shows it would be timely. The AI drafts inside the rules you set. You review and approve in batches at your cadence. You set the response windows. Not on call.

In Architect, the coach owns the client list. In Automate, the platform does.

What you can say to your client

Coaching is built on trust. The moment AI enters the relationship, the trust contract has to be renegotiated. Your clients are asking themselves the same questions we have just walked through. They have usually not articulated them. They have noticed that something has shifted.

The version that lands is short. You prepare with AI you read and edit. You stay the architect of any AI that talks to them between sessions. It is trained on your work with them, not the internet. You can change anything, any time. That is more precise than “I use AI for notes,” and your client will hear the difference.

By Q1 2028, every leadership and executive coaching engagement will include AI in scope. The next twelve months are the window in which the position you take becomes the practice you have for the decade after. Keeps you the coach.

Louise Doorn is co-founder and CEO of CoachNova and a contributor to the EMCC Global Digital and AI thought leadership team.