An AI assistant for coaches: what it handles well, and what it should not touch

Most coaches meet the idea of an AI assistant in the worst possible way: as a pitch that promises to run the practice for them. That framing is wrong, and it is why a lot of good coaches dismiss the whole category in ten seconds.

A useful AI assistant for coaches is a much smaller thing than the pitch suggests. It does not coach. It does not decide anything that matters. It handles the layer of your week that sits around the coaching, the part you would happily hand to a competent assistant if you had one, and it does that layer at two in the morning when a client is stuck and you are asleep. That is the whole promise. Everything beyond it is worth being suspicious of.

Here is what that actually looks like in practice, including the parts that do not work.

What it handles well

The reliable use cases have one thing in common: the answer already exists, and the problem is that the client cannot reach it at the moment they need it.

Repeat questions. Every practice has fifteen to twenty questions that come up constantly. How do I reschedule. Where is the worksheet from module three. What was I supposed to track this week. Is it normal to feel like this at the four week mark. You have answered all of them, often well, often at length. An assistant trained on your own material answers them in your words, immediately, and does not get tired of the fortieth one.

Onboarding. The first three weeks generate most of the repetition in a coaching relationship. A new client does not yet know how you work, what the rhythm is, or which of their questions are big enough to bring to a session. An assistant that walks them through that opening stretch removes a lot of small friction and sets the pattern for how they use you.

Finding the right thing in your material. Coaches accumulate content faster than they organise it. Session recordings, a course, PDFs, a newsletter archive, a long email you wrote two years ago that was better than anything in the course. Clients will not go looking through that. They will message you instead, because messaging is easier. An assistant makes the archive reachable at the moment of need rather than in theory.

A first pass on new enquiries. Before someone reaches your calendar, an assistant can ask the three or four questions that tell you whether this is a fit: where they are now, what they have tried, what the budget reality is. Not to sell, just to sort. It is the same triage you do in the first three minutes of a discovery call, done before the call exists.

Light between-session prompts. A short, well-timed nudge on the commitment your client made on Monday. Not a nag. Not a fake friendly check-in. Just a reminder of the specific thing they said they would do.

Where it fails

This is the more important half, and the part most vendors skip.

Anything requiring judgment about a specific person. A client says they are thinking about leaving their job. The right response depends on six months of context, their financial situation, what they said in the last session, and the thing you noticed in their voice. No assistant should be answering that. A well-built one recognises the shape of it and routes it to you.

Emotional weight. Clients occasionally arrive at the chat in a bad state. Grief, a crisis, something clinical. The correct behaviour is to stop being helpful, say plainly that this belongs with a person, and hand it over immediately with a clear route to reach you. An assistant that tries to coach through that moment is worse than no assistant.

Anything it was not given. An assistant trained on your material knows your material. Ask it something adjacent and, without firm boundaries in the setup, it will produce a confident, generic answer that is not yours and may not be right. This is the single most common failure in badly built assistants, and it is a configuration problem, not an inevitability. The fix is to make it say "that is one for your next session" rather than improvise.

Being you. Clients pay for your attention. If the assistant becomes the interface for the relationship rather than the layer underneath it, you have quietly downgraded the product they bought. The test is simple: your clients should get more of you, not less, because the hours the assistant saves go back into the work.

What it needs from you first

An assistant is only as good as the material behind it, which means the honest prerequisite is a body of work.

If you have a course, written frameworks, a decent set of recorded sessions, or a few years of thorough email answers, you have enough. The build is mostly a matter of shaping what exists. If you are three months into coaching and everything lives in your head, an assistant will produce vague, generic output, because vague, generic material is what it has.

The second requirement is a decision about boundaries. What should it always answer, what should it always refuse, what should it hand back to you. This takes an afternoon of thought and it determines whether the thing is useful or embarrassing. Most of the difference between a good assistant and a bad one is made here, before any of it is built.

The third is a willingness to keep it current. Your program changes. Your answers change. An assistant that still describes last year's structure is a liability. This is why we treat it as an ongoing thing rather than a one-off delivery, and it is a fair question to ask any vendor.

Is it worth it for your practice

Do the arithmetic before you do anything else.

Count the hours you spend in a normal week on repeat questions, onboarding, resending material, and discovery calls that were never going to convert. If that number is under two hours, an assistant is probably not your bottleneck and you should fix something else. If it is closer to six or eight, which is common in a full 1:1 practice, that is most of a working day every week going to work nobody is paying you for.

Then be clear about what you would do with the returned time. Coaches who get hours back and let them fill with more of the same have not gained anything. Coaches who use them to take two more clients, or to write the next program, or to stop working on Sundays, have.

If that adds up, this is what we build: branded AI assistants for coaches, trained on your own material, running on your own AI account with a spend cap set on the first day. How it works covers the setup and pricing is on the same page. Most practices are live within days, and if the honest answer after a call is that your material is not ready yet, we will tell you that instead.

← All posts