AI Augmentation vs. Substitution: Where the Line Is in Your Coaching

ICF Core Competency 5 requires the coach to “be fully conscious and present with the client, maintaining a mindset that is open, curious, flexible and client-centered.” Nothing in that wording prohibits an AI tool. Nothing in it permits one either. What it does is hand you a test: whatever you are considering - an AI note-taker, a prep tool, a post-session summariser - ask what it does to your presence in the room. That question is this article’s spine, and it sits inside the full AI in coaching guide as the decision every other AI choice routes through.

The Question Beneath the Question

The AI-in-coaching debate usually arrives in one of two shapes: replacement (will AI replace coaches?) or adoption (should coaches use AI at all?). Both feel urgent, and both miss the question a working coach actually faces on a Tuesday afternoon.

That working question is narrower and more useful. When a specific AI tool is in or near a specific part of your practice, is it serving your competency or standing in for it? Not “is AI good or bad,” but “in this use, with this tool, right now - which is happening?” That distinction - augmentation versus substitution - is not a values judgment. It is a functional test, not one imported from the technology world. It is encoded in the framework you trained against. The ICF Core Competencies were written as principles a coach applies with judgment; read closely, they tell you which acts are yours.

The replacement question and the adoption question are both about AI. The working question is about you - which acts of coaching are yours to perform, and which a tool has quietly taken over.

This article does not enforce that line for you. No senior coach can hand you a list of approved tools and call the job done. What follows is a diagnostic instrument - a way to run the augmentation-versus-substitution test on your own practice decisions.

What “Substitution” Actually Means in This Context

Substitution, in this article, has a precise meaning: the AI performs a cognitive or relational act that, under the ICF Core Competencies, is the coach’s act. Your competency is supposed to do the thing. A tool that approximates it instead has substituted for the competency. That is the whole definition - functional, not moral.

Substitution is not the same as risk, and conflating them is the most common error here. A tool can carry serious data risk without substituting for any competency: an AI note-taker running on your own system, never disclosed to the client, is a confidentiality and consent problem, but it does not stand in for Listens Actively. The competency still happened. The failure is elsewhere.

The reverse is also true. A tool can substitute for a competency while carrying no obvious risk. An AI that generates client questions for you mid-session might produce genuinely sharp questions - no data breach, no consent gap, nothing that trips an ethics alarm. And yet the curiosity that produced the question was the tool’s, not yours. The substitution is invisible precisely because the output is plausible.

A tool that produces good questions has not passed the test. It has only made the test harder to run, because the output looks like something you would have said.

It also matters what substitution is not. Using AI to transcribe a session after it ends is not substituting for Listens Actively - the listening already happened, in real time, performed by you. The line is more specific than “AI in the session, bad; AI outside it, fine.” Substitution is a property of how a tool is used, not of the tool itself. The same note-taker can augment one coach’s practice and substitute in another’s, depending on which act it performs.

That gives you the gating question for everything below: name the act the tool performs, then ask whether the Core Competencies assign that act to the coach.

The Three Competencies Where the Line Falls

The line is sharpest against three ICF Core Competencies, because each names an act only the coach can perform: Maintains Presence, Listens Actively, Evokes Awareness. Quoting each verbatim and mapping common AI uses against it shows where augmentation ends and substitution begins. For the broader regulatory picture, what ICF’s AI Coaching Framework says about presence sits underneath this reading - the framework names the domains; the analysis here applies them.

Competency 5: Maintains Presence

The ICF Core Competencies define this competency as being “fully conscious and present with the client, maintaining a mindset that is open, curious, flexible and client-centered.” Read the verbs. Conscious. Present. These are not states a tool produces on your behalf. They are what the coach brings into the room and sustains, second by second, for the length of the session.

An AI note-taker running quietly in the background does not, by itself, substitute for Maintains Presence. If your full attention stays on the client - the tool silent, off to the side, capturing a record for later - the competency is intact. But the same tool can crowd presence the moment your attention shifts toward it: glancing at the live transcript, editing as it scrolls, relying on it to do noticing you would otherwise do yourself. The question to run is direct: is the tool freeing your attention for the client, or redirecting it toward the tool’s output?

An AI generating client questions in real time is a different matter. “Maintaining a mindset that is open, curious, flexible” - the curiosity named there is the coach’s. An AI that reads the live transcript and proposes the next question is generating the curiosity. The test is unforgiving: if you used that question, who was curious? An AI summarising the session afterward raises no presence question - the session is over, presence is complete. That use belongs to a later section.

Competency 6: Listens Actively

The Core Competencies describe this one as focusing “on what the client is and is not saying to fully understand what is being communicated in the context of the client’s systems and to support client self-expression.” The interpretive weight sits in five words: what the client is not saying. Listening, in the ICF sense, is the coach’s noticing - of the hesitation, the change in register, the thing carefully not mentioned, the silence that carried more than the sentence.

AI transcription records what was said. It cannot record what was not said - the pause, the swallowed word, the shift from confident to uncertain in the space of a clause. As a review tool after the session, transcription serves Listens Actively well. As a stand-in for it during the session, it cannot: the thing the competency names is not in the transcript at all.

An AI-generated session summary with “key themes” sits right on the edge. Used to supplement your own noticing - checking the themes you caught against ones you missed - it augments. If it replaces your theme-formation, if you skip “what did I notice?” because the AI already answered it, it substitutes. One question settles which: did you form your own reading before you read the AI’s? An AI trained on the transcript to surface “emotional patterns” or “resistance signals” carries the highest substitution risk here. However sophisticated the pattern detection, an AI doing that focusing for you is standing in for Listens Actively.

A transcript holds every word the client said. It holds none of the silence. The silence is where the listening competency actually lives.

Competency 7: Evokes Awareness

The Core Competencies frame this competency as facilitating “client insight and learning by using tools and techniques such as powerful questioning, silence, metaphor or analogy.” The wording explicitly includes “tools and techniques” - and an AI could be one of them. The competency does not forbid tools. It rests on a different act: the coach’s judgment about which question or intervention, at this moment, with this client, will create the conditions for insight.

AI-assisted pre-session prep - generating possible questions from previous session notes - augments the competency when the coach reviews, selects, adapts, and discards from the list. Your judgment does the evocation work; the AI supplies raw material. Substitution risk rises sharply the moment the coach delivers AI-generated questions without that judgment step.

AI suggesting questions during the session, in real time, is the highest-risk use here. The judgment about “what question, at this moment” is the evocation act, and a tool that performs that judgment - even performs it well - is substituting for the competency. AI-generated metaphors follow the same rule: hear the AI’s suggestion, set it aside, generate your own from the coaching, and that is augmentation. Deliver the AI’s metaphor, and the AI evoked the awareness while you carried the package.

Across all three competencies, the through-line is the same. The test is not “where is the AI in the session timeline?” It is “which cognitive and relational acts belong to the coach, and is this tool performing them?” Timing is a heuristic, not the test.

The AI-Safe Zone: What Augmentation Looks Like in Practice

The caution above holds in full. An AI performing a Core Competency act during a session is, at minimum, an unresolved question. Granted. Which is exactly why it is worth being precise about where AI clearly does earn its place - a blanket “no” is as unexamined as a blanket “yes.” Here is the part of your practice where the answer is a confident yes, each use paired with the condition that keeps it augmentation.

Pre-session prep. Reviewing previous session notes, generating possible thematic threads, researching context - a client’s industry, a framework they mentioned in passing. This augments because the coach’s preparation sharpens their readiness while the session-time work remains the coach’s. The condition: prep produces readiness, not a script.

Post-session reflection prompting. Using AI as a reflection partner - “what did I notice about my own patterns in that session?” - on the understanding that its responses are prompts for your thinking, never answers to it. AI produces fluent, agreeable, slightly flattering reflection output that tends to confirm rather than challenge. Your job is to hold your own observation as primary and treat the AI’s as a provocation to test against.

Administrative and back-office AI. Scheduling, invoicing, email drafting, marketing copy. These touch the operations layer, not the competency layer - no Core Competency is at stake. The relevant tests here are disclosure and data handling, and the vetting test applies every time a tool touches client data, even when no competency is in question.

Transcript review after the session. Reading back what was said, noting what you missed, identifying patterns across several sessions. The session is complete; listening and presence were performed in real time, by you. Reviewing the record afterward is reflective learning.

Consider two coaches making opposite choices about pre-session AI prep. One asks an AI to generate possible questions from the last four sessions’ notes, then reads, discards most of it, and selects two threads with judgment. One writes prep questions by hand, because the act of writing is how they metabolise the previous session. Neither is wrong. The tool is the same; the practice it serves is different. Both kept Competency 7 in their own hands - by different routes.

Where the Test Gets Hard: The Ambiguous Zone

Three AI uses do not sort cleanly. An honest framework names them rather than pretending its categories cover everything - so here they are, addressed directly, and left unresolved where they should be.

AI for between-session client support. Picture a client who uses an AI app between sessions to process what came up in your last conversation. You neither deployed the tool nor control it. Does this touch Competency 6 in the next session? Not directly - but the coaching conversation now has a third voice in its history. The competency response is not a policy about the client’s app; it is the coach’s attunement to the fact that the client arrives having already done some of the work, somewhere else.

AI that generates session summaries from audio. Vendors position this as “freeing the coach’s attention for the client.” The test cuts through the marketing: if the AI is summarising, are you still doing the Listens Actively work? It depends on sequence. Did you form your own reading - what was said, what was not, what landed - before relying on the AI summary? If yes, the AI is a cross-check and the use augments. If no, the summarising act has been substituted, and the polish of the summary does not change that.

AI role-play for coach development. A coach practising their skills against an AI between supervision sessions. No client session, no real-client data, no competency at risk. But the feedback loop has a flaw: an AI conversation partner is obliging, and obliging feedback can produce fluency without accuracy - a coach who sounds smoother because the AI rolled with it, not because the competency deepened. AI role-play is useful and bounded; it should not replace supervision, where another trained human notices what an agreeable model will not.

The framework handles all three cases. It just does not produce “yes” or “no” - it produces “investigate further, and here is what to watch.” A framework that admits its edge is more trustworthy than one claiming to resolve everything.

Running the Test: A Question Sequence for Any AI Decision

The three-competency analysis distils into a sequence you can run on any AI decision, in any context. Each question either ends the evaluation or passes it forward. Call it the competency line test.

1. Which layer does this touch? If the use is purely operational - scheduling, invoicing, email, file management - the three competencies are not in play. Move to the data and disclosure tests and stop here. If the use touches the session, the prep that shapes it, or the delivery of interventions, continue.
2. What act does this tool perform? Name it specifically. “It summarises” is not specific enough; “it forms the reading of the session that I would otherwise form myself” is. If the named act is one the Core Competencies assign to the coach - noticing, curiosity, judgment about what question to ask, presence - go to the next question.
3. Who is performing that act? If your competency does the work and the tool serves it, that is augmentation. If the tool does the work and you deliver it, that is substitution. The output looks similar from the outside; be honest about the source.
4. If there is genuine ambiguity, what would I need to observe to know? Name the observable thing in your next session that would tell you which is happening. A test with no observable is a test you cannot apply.

This sequence does not resolve the hard cases - it hands you the method, not the verdicts. Run it before the tool is used, not after you have admired the output. The fluency of a good result conceals the source of the act, which is why the timing of the test matters as much as the test.

A coach who asks an AI to run their session knows they crossed a line. A coach whose AI-generated prep is so good they never write their own - who would notice?

The Competency Line, Restated

The competency line is not a wall you build once and then stand behind for the rest of your career. It is a question you keep running: in this use, with this tool, right now - is the AI serving the coaching, or standing in for it?

You will not always like what that question surfaces. Sometimes the honest answer is “I should not be using this tool the way I have been.” That inconvenience is not a flaw in the test - it is the test working. Asking the question anyway, especially when the answer costs you a convenient habit, is precisely the judgment the Core Competencies were written to ask of you.

One context tightens the line further. When sessions are submitted as evidence for an ICF credential, the question is no longer only about your practice - it is about the record an assessor relies on, and credentialing submissions have a separate rule worth knowing first.

That is the whole test. Take it into your next decision.

This article quotes the ICF Core Competencies and references the International Coaching Federation’s AI Coaching Framework (November 2024). It is professional education, not legal advice. The full ICF Core Competencies document is the authoritative source for the wording quoted here.