TL:DR: If you’re using AI tools like ChatGPT or Claude to create content and still struggle to get coaching clients, your prompts are the problem. This guide teaches the CRAFTY method (Context, Role, Action, Format, Target, Your turn) so that coaches can write prompts that yield accurate and actionable results.
If you want help with tools, check out The Best AI Tools for Coaches.
I vividly remember, as a teenager, watching with astonishment as my sister’s boyfriend missed the entrance to our driveway by several feet and slammed her car into my dad’s prize privet hedge.
He got out totally unfazed, walked up to the house, opened the door, and confidently proclaimed, ‘There’s something wrong with the gearbox.’
The fact that, at almost 30 years of age, he’d decided to drive a car (and my sister let him) without a licence or ever having had a lesson for the first time, seemed irrelevant to him.
No gearbox issue was ever found and, to the best of my knowledge, Mick never drove a car again.
There are a lot of people as confident there is a problem with AI models like ChatGPT and Claude, as Mick was with the car.
They hop aboard, ask it to perform something akin to witchcraft, and when it fails, assume all the hype is just that—hype.
It isn’t.
AI is all that, and then some.
But just as Mick ought to have mastered the rudimentary basics of braking and turning before getting behind the wheel of a car, you need to know how to write prompts that AI can work with.
Think of AI as a highly intelligent virtual assistant.
One who will never take a day off with a hangover, get bored, or demand a pay raise.
With any such assistant, you would take the time to train them or, at the very least, clearly explain your standard operating procedures (SOPs).
You would also provide ongoing support and offer help and feedback.
AI is the same.
Saying, ‘Write me a blog post on how to avoid procrastination and make it good,‘ is about as likely to yield great results as me telling my dogs to stop fucking about when I’m trying to put their collars on to go for a walk.
However, unlike my dogs, who will ignore me and continue to fuck about, AI will give you what it thinks you want.
And it will probably be shit because it’s pulling answers out of its digital arse.
Hopefully, you will know it’s shit and not publish it.
But some many coaches don’t know, and moments later, it’s out in the public domain representing their brand.
That’s not good
So, to help you avoid that, here is a quick guide to writing brilliant prompts using the easy-to-remember acronym CRAFTY.
Stick to the end, where I will give you some real-world examples.
C – context (this is about you)
Give the AI as much background and context as you can.
You wouldn’t hire a Michelin-starred chef, and when asked what you would like cooking, respond, food!.
Maybe you can share the copy from your About page so it knows more about you.
Or perhaps you have been smart enough to figure out your client avatar and can include that as part of your prompt (more at the end).
R – role (this is about the AI)
Who or what do you want the AI to be?
If you don’t give it a role, it will take the afternoon off and serve up the kind of generic crap that we all immediately know is AI.
It has almost the entire fucking internet to choose from, so make it work for its living.
This may be as simple as asking it to be yourself, in which case, the context and format elements become even more crucial.
But you may want it to get a lot more creative than that.
I may say I want it to be the brilliant (dead) advertiser and direct response marketer, David Ogilvy, or (not dead) marketing maven, Seth Godin.
Both of those have a huge body of work from which the AI can draw.
You don’t have to give it a specific person or people by name. You can be more generic and tell it to assume a role as one of the world’s leading <insert whatever topic you are writing about>.
A – actions (this is what you want it to do)
What do you want the AI to do?
Are you writing an email or a sequence of emails for a launch to your newsletter list?
Perhaps you want it to generate 365 Facebook posts, allowing you to schedule one a day for the next year.
Or perhaps you want to rewrite the copy for your entire website because you’ve just gone nuclear because some bastard stole your niche, and you feel like you need to start again.
F – format (this is how they present it)
This is similar to context, but it literally means how you want it to be written for you.
You may have noticed that I tend to drop the F-bomb occasionally.
That may be necessary in longer posts if it’s to sound authentic, even when I’m asking it to write in the style of another person.
Because style is the operative word, I’m not copying or mimicking them.
Similarly, I make sure my posts are easy to scan by not having paragraphs longer than a sentence or two, so I tell it.
You may have an irrational and slightly concerning love of bullet points, or you may want a blog post with several subheadings, or you may need a comparison table created.
Whatever it is, spell it out.
T – target (this is about your ICA)
A chef needs to know if their audience is vegan or has allergies.
A clothes designer needs to know if the attire will be worn for work, or an S&M party.
AI needs to know who your ideal client is so that it can provide you with relevant and worthwhile content.
It also needs to know what you want people reading, listening, watching the content to do next.
On copy or sales pages, this is often the call to action (CTA), which may involve clicking a link, purchasing a product, or sending an email, among other actions.
Knowing this in advance enables it to produce content with a clear purpose.
Y – your turn
AI is designed to fill in any blanks you don’t fill in yourself.
This is one of the reasons it can give shit or flat-out wrong answers.
To help minimise the possibility of this, you need to ask it what else it needs to know, but saying something like..
‘Is there anything else you need from me to improve this? Anything I’ve missed or could expand on to help you give a better response?.
Real-world example for a career coach
Context
I am a career coach who helps experienced professionals in their 30s and 40s escape the ‘golden handcuffs’ of a stable but unfulfilling job. My brand voice is empathetic but direct, like a firm friend who wants the best for you. I’ve created a free checklist titled ’10 Signs It’s Time To Leave Your Job.’
Role
Please act as a world-class direct response copywriter, specialising in creating high-engagement social media content.
Action
Write three distinct LinkedIn posts announcing my snazzy new lead magnet PDF.
Format
- Post 1 should use the AIDA (Attention, Interest, Desire, Action) framework.
- Post 2 should use a storytelling approach. Start with a short, relatable anecdote about feeling stuck in a career before revealing the solution.
- Post 3 should offer a ‘Contrarian Take’ or ‘Myth-Busting’ perspective. Start by challenging a common piece of career advice and then introduce the checklist as a better way forward.
- All posts must be under 200 words, use short paragraphs.
- Incorporate a psychological trigger of ‘urgency’ by mentioning the PDF is only available for a limited time.
Target
My target audience is mid-career professionals (ages 35-50) who are successful on paper but feel a deep sense of dissatisfaction.
They are active on LinkedIn, value professional development, but are risk-averse and fear making a career change
Your turn
Based on this, do you need any clarification on the specific pain points of my audience to make the posts more resonant?”
Making all of the above easy
You may be thinking this all sounds like hard work.
It is.
But so is explaining to your partner that you have no clients when they are browsing vacations in Bali and contemplating early retirement.
However, the real heavy lifting is done upfront, so you don’t need to keep reinventing the wheel.
I have a lot of the information I mentioned above saved in documents that I can upload to ChatGPT as and when I need them.
I have a Word document that explains my writing style and samples of about a dozen emails I have sent that I really like.
It also has a list of words/phrases that I don’t like because they either aren’t things I’d say, or sound too like AI. I update this frequently.
I have another document that contains my client avatar.
Yet another containing all my website copy.
And I have built a custom GPT (a process that is a lot easier than most people think) that I use for brainstorming.
I will mix and match as needed and update them as necessary.
You may be wondering why I don’t just slap everything I’ve got into every chat.
That’s because each model has a finite memory for each conversation/chat.
As of January 2026, depending on the model and plan you have, Claude has a content window of approximately 200,000 tokens (about 160,000 words).
ChatGPT 5.2 is about twice that, but it can still gobble it up quite quickly, especially if you’re using really long prompts and having long, in-depth conversations.
And the problem is that neither ChatGPT nor Claude notify you when you’ve used up all your tokens. All they do is truncate the conversation, often chopping off some of the important information and prompts you used at the beginning.
This is another reason why it’s sometimes best to start a new chat window after you’ve reverse-engineered everything that you’ve done, by asking the LLM explicitly what it needs to start a new chat window.
A simple prompt you are welcome to copy and paste, would be:
“Please review our entire conversation and create a “new chat starter pack” so I can continue this work in a fresh thread without losing quality.
Include:
- Goal: what we’re trying to produce and what “good” looks like
- Audience: who this is for and what they care about
- Key facts: any relevant background details you should remember
- Decisions made: what we agreed, chose, rejected, or changed
- Constraints: tone, formatting rules, length limits, do/don’t lists
- Assets: any drafts, frameworks, examples, links, or deliverables created
- Open loops: what’s unfinished, unclear, or needs a decision next
- Next step: what you should do first in the new chat
Finish with a single paste-ready block titled: “Context to paste into a new chat”.
Don’t trust the memory
Of course, ChatGPT has its own memory now, which is why it can be uncannily accurate when you ask it questions about yourself.
However, it can also be uncannily wrong as it does not have the ability to remember all the conversations you’ve had with it!
Try not to rely too heavily on the memory function; as of January 2026, it’s still hit-or-miss. Infinite memory is not here yet, and it’s easy to run out of memory in the context window.
Writing AI prompts doesn’t end there
As mentioned, this process gives you a starting point for your work with AI.
Besides the occasional very short social media post, doing all of the above will not produce results that don’t need editing or changing.
The first pass is just that, a first pass. Nothing more.
Now, the process of tweaking, refining, and improving begins.
Other than as an experiment, I’m pretty sure I have never sent out an email or published a blog post that wasn’t 70%+ my writing.
I’m sure there are times when I could have sent what I had sooner, and it would have been fine.
However, I have a very distinct and unusual writing style that ChatGPT and Claude find difficult to replicate.
As such, I tend to edit more than I suspect most coaches need to.
Need help using AI to help you get clients?
Click here, and let’s talk.