What Is an AI Coach? How It Works and What It Cannot Do

What an AI coach is

An AI coach is software that holds a coaching conversation with you. The distinction that matters is not that it is powered by a language model — most chat products are. It is that a coaching conversation is structured around inquiry rather than instruction: the working assumption is that you are the expert on your own life, and the coach's job is to help you see it more clearly, not to tell you what to do.

That is also why an AI coach can feel unfamiliar at first. A general assistant is optimized to be helpful and agreeable, so it answers. A coach asks. If you have used a chat assistant and expected an AI coach to behave the same way, the difference in posture is the thing you are noticing.

The word "coaching" itself has a working definition that predates any of this software, and it is worth anchoring to rather than assuming. The International Coaching Federation — the largest professional body for the field — defines coaching as "partnering with clients in a thought-provoking and creative process that inspires them to maximize their personal and professional potential." That definition does not mention advice, answers, or expertise about your specific situation. It describes a process, and the process is what an AI coach is attempting to run in software.

Three different things people mean by "AI coach"

The phrase covers products that are quite different from each other, and a lot of confusion about whether an AI coach is any good comes from comparing one kind against another. In practice it tends to mean one of three things: a general chat assistant that you have prompted to act as a coach, an AI feature inside a coaching service that a human coach runs, or a standalone product where the software itself is the coach. A practitioner guide to AI coaching sorts the field into those same three groups, and they differ mainly in who is making the decisions — you, a human coach, or the model.

There is also a fourth situation that is really a variation on the third: an AI coach that your employer provides. These are usually described as supporting learning, performance, and development at work, and they differ from one you choose yourself in a way that matters later on this page — someone else arranged access, and that affects how long your conversations are kept.

IX Coach is the third kind, chosen and paid for by the person using it. That is a description of where it sits, not a claim that one kind is better than another: a general assistant you prompt yourself is a reasonable place to start if you only want a thinking partner for an evening, and a human coach with AI tools behind them is a different proposition from either.

Why it asks questions instead of just answering

This is the most common thing people tell us about coaching software, and it deserves a direct answer rather than a defense. The inquiry-first design is deliberate. Advice that arrives before understanding tends to solve a problem you did not have, and being handed a conclusion rarely changes behaviour the way arriving at one does.

The honest other half: users have told us the questions sometimes keep coming when they have asked for them to stop. That cadence is requested of the model rather than mechanically enforced — earlier attempts to enforce it programmatically cut off thoughts mid-sentence, which was worse. So when someone asks a coach to stop asking and it does not, that is a real gap in the product, not a misunderstanding on their part.

It is worth separating two things that get conflated under "it keeps asking questions." One is the coaching method itself, which is inquiry by design and will not stop being that — an AI coach that switched to giving advice on request would stop being a coach in the sense described at the top of this page. The other is a failure to recognize a direct, explicit request ("just tell me what you think") and adapt to it, which is a real product gap rather than the method working as intended. The first is what you signed up for. The second is a bug.

What happens in a session

You start by saying what is real for you right now. There is no intake form and no configuration — you talk, and the coach listens for what matters underneath what you said.

From there the approach is chosen to fit the moment. Something you are stuck deciding might call for a solution-focused method structured loosely like the GROW model — clarifying the goal, the current reality, the options, and what you are actually willing to do; something heavier might call for internal-family-systems or mindfulness-based work. A conversation can move between lenses as you do, because a coach that forces one framework onto everything stops being useful quickly.

Insights are offered as vantage points to try on rather than as conclusions about you. You are free to discard one that does not fit, and discarding it is information too.

Explore: the grow coaching model · internal family systems

What a session actually sounds like

Describing a coaching conversation in the abstract — "inquiry rather than instruction" — is easy to agree with and hard to picture. Concretely, a session tends to move through the same rough shape no matter what you bring to it.

It opens with something closer to "what's on your mind" than a menu of categories to pick from. What you say first is rarely the whole picture, so the coach's first move is usually to reflect back what it heard in something close to your own words and check whether that is the actual thing, or a stand-in for something underneath it — the same move a human coach makes through reflective, active listening.

From there the questions narrow rather than the answers arriving. If you said you don't know what to do about a decision, a coach working well does not hand you a pros-and-cons list. It is more likely to ask what would need to be true for you to feel sure either way, what you would be giving up by choosing the option you are avoiding, or — if the decision connects to who you are trying to become rather than a single fact — what a version of you six months past this decision would say mattered most. That last kind of question is a direct application of a working idea in psychology: picturing a specific future self changes present-day motivation more reliably than picturing a vague better outcome. Some sessions get concrete enough to sort what actually matters against what only feels urgent, which is the same operation behind values-clarification exercises used in acceptance-based coaching and therapy.

A session can end in a lot of different places. Sometimes it ends with a next step you named yourself. Sometimes it ends with nothing decided, and that is a legitimate outcome rather than a consolation prize — clarity about what you are actually facing is itself the point of some sessions.

Explore: active listening reflective · possible selves · values bullseye

Memory: what it does and where it falls short

Sessions building on each other is what separates coaching from a series of unrelated conversations. What you worked on is stored and recalled by meaning rather than keyword, so the coach can connect something you said months ago to what you are working on now without you re-explaining it.

Concretely, that means a coach can notice that a decision you are circling this week rhymes with one you worked through two months ago — not because you used the same words, but because the pattern underneath them matches — and can bring that connection into the conversation instead of treating today as a blank page. A human coach does this by remembering you; software does it by retrieving the meaning of what you said, which is a different mechanism aimed at the same outcome.

Where this currently falls short: goals are not tracked with a sense of when they were last touched, so an old goal can be treated as though it were still live. If you have moved on from something and the coach keeps returning to it, that is why — not a misread of what you want now, but a system that has not yet learned to let go of what you wanted then.

What happens to what you tell it

A coaching conversation is more personal than most things people type into a screen, and the memory described above only works because what you said is stored. Those two facts are connected, and worth thinking about before the first session rather than after the fifth.

The International Coaching Federation has published a framework of standards for AI coaching, written for the people who build these products. It treats data as a first-order concern: its own foreword notes that coaching requires disclosing personal information that can be sensitive. One element asks that a provider document who within its organization can access coach and client data and how users can access it, and a more demanding version asks for public documentation of every party who may have access and where the data is stored. Another asks that a system tell clients how it manages data, with attention to security and confidentiality, before coaching starts. The framework describes what a provider should do; it is not a list of products that have done it, and nothing here should be read as saying any given product meets it.

To see what a real policy looks like, take one published example: the privacy policy for the AI chat assistant that CoachHub offers through employers. It says the conversation inputs and outputs are collected, that the data is retained for as long as the contract between CoachHub and the organization that granted you access lasts (typically your employer), and that you may delete your inputs and outputs at any time through the platform, while noting data may be kept longer to establish a right or meet a legal obligation. Those are the same questions worth asking of any AI coach: what is kept, for how long, who arranged your access, and what you can delete yourself.

Practically, that comes down to a short list you can check in a few minutes in any product's published policy, ours included: whether the conversation text itself is stored or only a summary, how long it is kept after you stop using the product, whether you can delete it yourself, who besides the provider can see it, and whether it is used to train models. If a policy does not answer one of those, that is information. And a rule of thumb that holds whatever the policy says: do not tell any software something you would not be able to live with being kept.

What an AI coach cannot do

Some limits here are not engineering problems that will be solved next year. They follow from what a coach without a body is.

Nervous systems regulate in relationship — through tone of voice, facial expression, and physical presence. Feeling safe is a physiological state before it is a belief, and that kind of settling passes between bodies. Software cannot transmit it.

There is also a threshold past which reflection stops being available. In states of high activation or shutdown, the capacity to stand back and examine a thought is not reduced but largely offline. What helps then is regulation, not insight, and a screen is drastically limited at providing it.

And memory is not one system. Talking through a story works on the narrative kind. Experiences encoded as implicit, body-held memory can intensify rather than resolve under repeated retelling — which is one of several reasons this work is not a substitute for therapy with a trained human.

Finally, coaching relationships change both people. An AI is not moved by what you say, and does not develop because of having met you. Whatever that mutual quality contributes, it is not available here.

When this is not the right tool

Coaching, delivered by a person or by software, is built for someone who is functioning and moving toward something — not for diagnosing or treating a mental health condition. The International Coaching Federation states the boundary plainly in its own published guidance for coaches on referring a client to therapy: "What is important is that a coach does not diagnose nor does a coach offer treatment." That line applies to a conversation with software exactly as much as to one with a person; the tool does not change what the problem actually is.

The same ICF guidance names one test above the others for knowing when the line has been crossed: the level of daily functioning. In its own words, "when the problem gets in the way of everyday functioning (professionally, personally, or socially), you should make a referral to a mental health professional," and that referral is warranted specifically "when the issue lasts for weeks and affects functioning in daily life." A rough day is not the threshold. Weeks of interference with work, self-care, or relationships is.

That guidance also names the issues most likely to call for a licensed professional rather than a coach: anxiety, depression, eating disorders, post-traumatic stress, substance abuse or addiction, suicidal ideation, and thought disorders. If what you are carrying looks like one of those — not as a passing mood but as something that has lasted and is interfering with how you function — that is a reason to see a therapist, not a reason to try harder with a coaching app.

If you are in crisis right now, in the US you can call or text 988 to reach the Suicide & Crisis Lifeline, available 24/7/365, free and confidential. That is true whether or not you ever use this product.

None of this is a disclaimer bolted onto marketing copy. It is the actual boundary of what a coaching conversation — with a person or with software — is built to do, stated as plainly as the profession itself states it.

How this differs from a general chat assistant

A general assistant is trained toward agreeableness, which is a liability in coaching. Growth usually requires being shown a pattern you are inside of, and a model optimized to validate will not reliably do that.

The mechanism worth naming is the subject-object shift: while you are fused with a pattern you cannot see it, because you are looking out from inside it. The moment you notice what you are doing, the reaction becomes something you can examine rather than something you are. Most coaching methods are, at bottom, different routes to that same move.

On memory specifically, we would rather not overclaim. People routinely compare coaching tools to assistants they already use that also retain context across conversations. The difference worth weighing is method and posture, not storage.

How this compares to a human coach and to therapy

Most people who search "AI coach" are already holding one of two comparisons in mind, and each deserves a straight answer rather than a redirect toward a sale.

Against a human coach, the tradeoff is structural rather than a matter of one option simply being better. A skilled human coach reads tone, hesitation, and what you do not say, and the relationship itself does some of the work — decades of psychotherapy research point to the quality of the working alliance as one of the more consistent correlates of good outcomes across approaches, and that is not something software currently replicates. An AI coach is available at 11 p.m. on a Tuesday and costs a fraction of an hourly rate, which changes what coaching is reachable for and when. Neither fact settles which is right for a given problem — a separate page on this site walks through how to decide between the two for your specific situation in more depth than fits here.

Against therapy, this is not a matter of degree at all — it is a different profession aimed at a different problem, covered on its own terms in the section below and, in more depth, on a dedicated page comparing an AI coach to a therapist directly.

Cost, and who this is not for

IX Coach is $40 per month, with the first seven days free and a card required to begin. For comparison, human coaching commonly runs from roughly $100 per hour to several hundred for senior executive work, over an engagement of months.

That comparison is not an argument that one replaces the other. It is a statement of what each costs. Some people should see a human, and some of what a human coach provides is in the list above of things software cannot do.

It is also worth saying plainly that a meaningful number of people who try coaching software and do not continue tell us the reason is money. If $40 a month is not something you can commit to right now, that is a straightforward reason this is not for you at the moment, and no framing changes it.

This is also not for you, at least not as the whole answer, if what you are carrying is closer to the list in the section above — something that has lasted weeks and is interfering with how you function. No price point changes what a tool is built for, and a coaching app priced correctly for coaching is still priced wrong for a problem that needs a clinician.

How to evaluate any AI coach

Four questions separate coaching software from a chat interface with a coaching label. Is it built on identifiable methods you can look up, or is the method unnamed? Does it carry context between sessions, or start fresh each time? Does it ever disagree with you, or only affirm? And is it clear about what it is not for?

That last one matters most. A coaching product unwilling to name its own limits is telling you something about how it will behave when you bring it something outside them.

One concrete way to test the last question before paying for anything: describe a scenario that is closer to a mental-health issue than a coaching goal, and see what happens. A tool that recognizes the edge of its competence and points you elsewhere — toward a therapist, toward 988 in a crisis — is telling you it knows what it is. A tool that answers anyway, confidently, is a warning sign about every other answer it has given you.

What a published standard asks of an AI coach

The four questions above are a quick test. There is also a more formal yardstick: the same International Coaching Federation framework of AI coaching standards, written for providers, sets out concrete things a conversational system should do, and several of them are things you can see for yourself in a first session.

It asks that a system tell you it is not human before coaching starts, and in its more demanding version at least once in each conversation, on the reasoning that people behave differently when they think they are speaking to a person. It asks that the system state the limits of what it can do before coaching begins. It asks that you have the option to reach a human coach, at minimum with basic instructions for finding one. And it asks that the system warn that it is not meant for mental health and say where to get help, with a more demanding version in which it looks for signs of mental-health risk and says plainly what it cannot do about them.

That last pair lines up with the test earlier on this page — describe something that looks clinical and see what happens. What the framework adds is that none of this should be left to you to discover: a responsible product says it up front. These are the framework's requirements for providers; we have not claimed here that any product, including this one, has been evaluated against them, so treat them as questions to put to a product rather than a seal to look for.

Frequently Asked Questions

What is an AI coach? Software that holds a coaching conversation with you — built around inquiry rather than instruction, drawing on named coaching methods rather than improvised encouragement, and in stronger implementations carrying context between sessions rather than starting fresh each time.

Is an AI coach the same as ChatGPT or a general chat assistant? No, though both can run on similar underlying technology. A general assistant is optimized to be helpful and agreeable, so its default is to answer. A coach's job is to ask — to help you see your own situation more clearly rather than hand you a conclusion. The technology can be similar; the posture is the difference that matters.

Can an AI coach replace a therapist? No. Coaching, delivered by a person or by software, is built for someone who is functioning and moving toward a goal, not for diagnosing or treating a mental health condition. If what you are carrying has lasted weeks and is interfering with how you function — work, self-care, relationships — that is a reason to see a licensed therapist, covered in more detail above.

Does an AI coach actually remember previous sessions? A well-built one does, and recalls what you worked on by meaning rather than requiring you to re-explain it each time. Where this commonly falls short is that goals are not always tracked with a sense of when they were last relevant, so a tool can keep returning to something you have already moved past.

Is an AI coach worth the cost? IX Coach runs $40 a month against roughly $100 an hour and up for human coaching, but the honest comparison is not price alone — a human coach provides things (presence, a relationship that changes both people, the capacity to hold you through high activation) that software does not. Whether the tradeoff is worth it depends on what you actually need from the conversation, covered above.

How do I tell a real coaching tool from a chatbot with a coaching label? Ask whether it is built on methods you can look up and verify, whether it carries context between sessions, whether it ever disagrees with you rather than only affirming, and whether it is honest about what it is not for — including, concretely, whether it recognizes a mental-health scenario and points you elsewhere instead of answering anyway.

What can't an AI coach do that a human coach can? It cannot regulate your nervous system through tone, expression, or physical presence — safety is a physiological state before it is a belief, and that kind of settling passes between bodies, not through a screen. It cannot reliably help when you are past the threshold where reflection is available at all, such as high activation or shutdown, where what helps is regulation rather than insight. And it does not develop because of having met you the way a human relationship does. These are covered in full above.

What are the different kinds of AI coach? Broadly three: a general chat assistant that you prompt to act as a coach, an AI feature inside a service run by a human coach, and a standalone product in which the software is the coach. A fourth situation, an AI coach provided through your employer, is usually a version of the third with someone else arranging your access.

What happens to what I tell an AI coach? It depends on the product, and the answer is in its published policy rather than in anything a coaching page can promise for it. The questions to check: whether the conversation text is stored, how long it is kept, whether you can delete it yourself, who else can see it, and whether it is used to train models. The International Coaching Federation's AI standards ask providers to document who can access client data and to explain their data practices before coaching starts.

What should an AI coach tell me before I start? Under the International Coaching Federation's framework of AI coaching standards, that it is not a human, what its limits are, how it handles your data, that it is not meant for mental health with somewhere to turn if you need help, and how to reach a human coach. Those are requirements the framework sets for providers, not a certification any product necessarily holds.

Research

• Hullinger, A. M. and DiGirolamo, J. A., (2018), Referring a Client to Therapy: A Set of Guidelines, International Coaching Federation — Source for this page's clinical-boundary claims: ICF's own definition of coaching, the statement that a coach does not diagnose or treat, the daily-functioning test for when to refer to therapy, and the list of issues (anxiety, depression, eating disorders, PTSD, substance abuse, suicidal ideation, thought disorders) that warrant a licensed professional rather than a coach.
• 988 Suicide & Crisis Lifeline, 988lifeline.org — Source for this page's crisis-line reference: available 24/7/365, free and confidential, reachable by call, text, or chat. verified 2026-10-08
• The ICF Artificial Intelligence Coaching Framework and Standards (V1.01), International Coaching Federation — Source for the section on what a published standard asks of an AI coach: the elements on AI disclosure, system limits, data transparency, data-management disclosure, access to a human coach, and mental-health risk indicators. The document sets out requirements for providers; it does not certify any particular product.
• CoachHub Platform Privacy Policy (section on its AI chat assistant), CoachHub — Used only as one worked example of what a published AI-coach privacy policy says about retention and deletion when the coach is provided through an employer. It is not a statement about any other product. verified 2026-10-08