AI for Coaches: A Human-First Guide

AI for coaches starts with the coach, not the software. Decide what your coaching is made of, decide which parts of it a machine can carry, and every tool decision after that takes about a minute. Go the other way round, tool first, and you end up with four subscriptions, a practice that sounds like everyone else's, and a nagging sense that the thing you're good at is getting less of your time, which is the opposite of what you wanted AI for.

I've taught AI to hundreds of coaches over the last four years, through my own programs and the AI for Coaches Mentorship with Coaching.com, and I coached coaches for eleven years before a machine could write a sentence. This is the page I'd send you if you asked me where to begin. It's long because the question is bigger than "which app," and it's in one place because I got tired of answering it in pieces.

What the machine is actually good at

Start with a clear look at what AI does well, because most of the fear and most of the hype come from getting this wrong.

AI is good at the average. It has read nearly everything, so it produces the middle of everything: the competent, the expected, the already-said. Ask it for a LinkedIn post about resilience for coaches and you get something that reads like the forty other posts about resilience that trained it. Ask it for a coaching framework and you get a plausible five-step model with an acronym in under a minute.

That used to be a large part of a coach's visible work. Producing on-topic, competent material was itself labor, and plenty of practices were built on top of it, mine included for a stretch. The machine now produces that layer for free, and free things stop proving anything about the work behind them.

So AI won't replace you. It will reveal you. If a machine can replicate what you do, that was never the work, it was the packaging around the work. What's left is the part that doesn't survive being averaged: your specific judgment, the question only you'd think to ask, the read on a client that comes from having sat with two hundred of them. The full argument is in AI Won't Replace You. It Will Reveal You., and it's the belief everything else on this page rests on.

Decide what stays human before you open a tool

Here is the first thing I'd have you do, and it involves no software at all.

Write down three things in your coaching that never get decided without you. The coaching conversation itself. The moment you say the thing nobody else will say to a client. The read on whether someone is ready. Whatever it is for you, write it down, on paper, before you deploy anything.

I call this your sacred territory. It's the part of your work that stays human, named deliberately, in advance.

The important part is what defines it. Sacred territory isn't a list of things AI can't do. That list shrinks every quarter, and a boundary built on it needs rewriting every time a lab ships something. It's the list of what you choose to keep human, because of who you are and how you want to work. Values don't get made obsolete by a model release.

"But Colin, isn't this the human-in-the-loop thing everyone talks about?" No, and the difference matters. Human in the loop means you check the machine's output, every time, which at any real volume puts you in the one place you do least good. Sacred territory is a decision you make once, calmly, and it holds whether or not anyone is watching. I've taken the loop question apart in The Human in the Loop Is the Bottleneck. The short version: decide what never goes to a machine, then stop pretending you were reading all of the rest anyway.

Then decide what to hand over

Once the line exists, everything on the other side of it is a candidate, and two questions sort the candidates.

Does it need you? Not could you do it. Does it need your judgment, your relationship with the person on the other end, your read on what's actually going on.

Does it drain you? Does it cost you more energy than it gives back.

Four combinations, and every task in your week lands in one of them.

Needs you and doesn't drain you: that's the coaching. Protect it. Everything else on this page exists to buy you more hours there.

Needs you and drains you: preparing for a session by rereading six months of notes, or writing a proposal for a complicated client. Don't hand these over. Point the AI at the draining part and stay in the deciding part. It pulls the history together, you say what it means.

Doesn't need you and doesn't drain you: scheduling, filing, chasing an invoice. Automate it when it's convenient. This is where most coaches start, and it's the least valuable place to start.

Doesn't need you and drains you: transcribing and tagging your own session notes, first-pass research on a new client's industry, the monthly summary nobody reads closely. Start here. This is where your hours are going, and it's where AI pays for itself inside a fortnight.

Run it on last week, not on next week. Open your calendar and your sent folder, take the actual tasks from the last seven days, and place them. Then pick one item from the bottom right and hand it over this week. One. The grid itself, with the diagram and the mistake that costs coaches the most, is in Does It Need You? Does It Drain You?

The tool is not where the value is

This is the part the tool lists leave out, and it's the part that decides whether any of this works for you.

There's no magic in the tool. The AI companies would rather you didn't hear that, because the story sells better when the intelligence lives in the product. The power is in what you bring: your material, your constraints, your judgment. When an AI is answering from your material, it has far less room to answer from its own. When you've said what counts as evidence, the invented citation has nowhere to sit. When you've told it whose judgment is running the exchange, it stops leading.

Whatever you leave out, the AI fills in with the average, and the average is nobody's. That's the mechanism behind almost every bad experience a coach has told me about. I've laid it out in Whatever You Leave Out, the AI Will Fill In.

So the real work of AI for coaches is getting what's in your head into files a machine can read. Who you serve. How you work. What you believe that your industry doesn't. What's live this week. I call this written base the Human Stack: five layers, from who you are down to what you're thinking about today. The first three files (one page about you, one index of how you work, one note on what's current) take about thirty minutes, and the walkthrough is here: Set Up Your Human Stack.

The format matters more than people expect. Plain text files, in folders you own, are what every model reads best, and they're the only format that survives the next five years of tool churn. That's why my whole practice lives in a plain-text second brain, and why I send coaches to Obsidian when they ask what to use. Whichever AI happens to be best this month inherits the same context in about ninety seconds.

The three levels of AI partnership

Most coaches using AI today are at level one and don't know there's a level three.

Level one is the mirror. The AI reflects and organizes what you give it. Drafts the email, tidies the notes, summarizes the call. Useful, and small. Most people live here, and it's where the "it just gives generic answers" complaint comes from, because a mirror shows you what you brought.

Level two is the thinking partner. It starts to see patterns you'd miss, challenges an assumption, asks the second question. You get here by giving it something real to think about: the actual client situation, the actual pricing, the decision you've been half avoiding.

Level three is the wisdom amplifier. The AI works inside your world, applying your frameworks, surfacing what you can't see alone, extending your methodology instead of substituting its own. This is what the Human Stack produces, and it's the level where AI stops being a faster typist and starts being the best-briefed colleague you've ever had. I've written about what changed when mine reached it.

What moves you up the levels is context, never a better model. The same tool that hands a stranger general principles hands me counsel aimed at this week's decision, because it has read my written business first. The clearest example I can give you is my AI board of advisors: five world-class minds, cast fresh for one question, each writing an independent brief on my situation. It's a level-three tool, and it only works because of what it can see.

Which AI tools do coaches actually need?

Fewer than the lists say. Here's what I'd give a coach starting today, and it hasn't changed much in two years.

One capable general model, on a paid plan. Claude or ChatGPT, at the tier that gives you enough capacity to work in it all day and lets you save instructions and projects. Whichever you pick, find the data settings and decide what you're comfortable with before a client's name goes anywhere near it. More on that below.

One note-taker, with consent. If you coach on video, an AI note-taker that transcribes and summarizes is the single highest-return tool you can adopt, and it's the one that needs your client's informed consent every time.

One place your context lives. A folder of plain-text files, ideally in a tool like Obsidian, holding your Human Stack. This is the piece that turns the first two from gadgets into a practice.

That's the stack. Three things. Everything else, the scheduling assistant, the design tool, the clip-maker, is optional and can wait until the three are working.

Two things I'd hold off on. An AI trained on your content to stand in for you with clients can wait until your methodology is proven with real people, because scale multiplies whatever it's given, and multiplying an unproven method multiplies the unproven part. And any tool that makes coaching decisions, or can't be overridden, is a hard line for me rather than a preference.

Ethics, consent, and what to tell clients

I'm careful here, so read this as one practitioner's reading of the rules and check it against your own body's code.

The current ICF Code of Ethics, which took effect on April 1, 2025, names artificial intelligence directly. Standard 2.5 has the coach fulfilling their obligations "directly and through any technology systems I may utilize," and lists AI among those systems. The companion guidance, revised in February 2026, says it more plainly: the coach is responsible for informed consent when any technology is used, AI included, and for researching the tool and its risks before proposing it. Using a tool doesn't move the responsibility onto the tool. The confidentiality standards still cover the transcript sitting in your note-taker and the client detail you pasted into a chat.

What that means in practice, for me and for the coaches I teach.

Consent before recording, every time, including when an AI device is doing the recording. Tell the client what the tool does, where the data goes, how long it's kept, and that they can say no and still be coached.

Anonymize by default. Client names rarely add anything the AI needs. "A senior leader in a family business, six months into a succession" gives the model everything it needs to be useful and gives away nothing. Written consent moves "may not" to "may, carefully." Anonymizing means you rarely have to ask.

Know your tool's data settings. Where it's stored, whether it trains on what you type, how to delete it. If you can't find out, that's your answer.

Then five questions I'd put to any tool before it touches client work. Can I explain this tool to my professional body? Does it serve my client's interests above my convenience? Do I understand it well enough to use it responsibly? Have I done my homework on it? Would I want it used on me if I were the client? Five yeses, or it waits.

None of this is legal advice. It's what I do, and what I'd want a coach working with me to do.

Where to start this week

Four moves, in order, one week.

1. Write your three lines. The three things in your coaching that never get decided without you. Ten minutes, on paper.
2. Run the two questions on last week. Take the real tasks from your calendar and sent folder, place them, pick one from the bottom right, and hand it over.
3. Write one page about yourself and paste it in. Who you serve, how you work, what you believe, what you're working on. Put it at the top of your next AI conversation and ask the same question you asked yesterday. Compare the two answers. That gap is the whole lesson.
4. Tell your clients. Add a line to your agreement about the tools you use and how their information is handled. Consent is cheaper to ask for than to apologize for.

After that, the next layer is letting the work that repeats run without you: the weekly prep, the research pass, the summary. That's a loop, and loops amplify whatever you hand them, which is why the three lines come first. A Loop Amplifies Whatever You Hand It is where I take that on.

Going further

If you'd rather do this with me than from a page, the AI for Coaches Mentorship is where I teach the whole of it live with Coaching.com, over twelve months: sacred territory, the two questions, the Human Stack, the three levels, the ethics, and then the building.

I'm also writing the book version, The Amplified Coach. If you'd rather read than enrol, the waitlist is here.

Whichever route, the order doesn't change. You first. Then what stays yours. Then what the machine carries. The coaches doing well with AI right now knew what they were for before they opened a tool. I still catch myself on the wrong side of my own lines most weeks. We're all figuring this out as it arrives, and I don't think that changes.