064 How To Make Your Own Custom GPT AI Assistant

In this episode, I walk you through how I’ve created and implemented custom AI assistants, specifically designed to support freelancers and health professionals like us.

I know firsthand how overwhelming it can feel juggling marketing, admin, and client support on top of the actual work you trained for. That’s why I believe digital assistants aren’t just a “nice-to-have”- they’re a game-changer.

You’ll hear me share a step-by-step guide for building your own custom GPT using ChatGPT, even if you’re not techy (promise). We cover how to get crystal clear on your assistant’s purpose, what kind of knowledge it needs, and how to give it instructions so it actually sounds like you and reflects your expertise.

My goal? To help you reduce burnout, boost your efficiency, and give you back more time to focus on the work that lights you up.

Takeaways:

• Custom AI assistants can help you save time, energy and headspace.
• You don't need to be “techy” to build one—just clear on what you want it to do.
• A GPT can help with client support, coaching, marketing, and repetitive tasks.
• The key is gathering the right knowledge to create its brain and being specific about it's purpose.
• You'll also want to make sure it reflects your tone of voice.
• AI doesn’t replace your expertise—it amplifies it.

Lets keep in touch!

Website: https://www.sarahalmondbushell.com/
Instagram: https://www.instagram.com/dietitiansinbusiness/
Facebook: https://www.facebook.com/dietitiansinbusiness
YouTube: https://www.youtube.com/@BeyondTheClinicPodcast

FREE Workbook - The Master Plan

Discover the 22 steps you need to take (in the right order) to build a successful business so you can earn enough to live the freedom lifestyle you dream of. https://www.sarahalmondbushell.com/master-plan

Work with me:

• Business Coaching: https://www.sarahalmondbushell.com/mastermind
• Sarah AI: https://www.sarahalmondbushell.com/offers/AYjozgYc/
• Book a discovery call: https://thechildrensnutritionist.as.me/discovery

Transcript

Sarah Bushell (00:00)

Hello, hello, welcome along. I am back in the studio for the first time since my holiday and it feels like such a long time ago now. But back in episode 62, which I recorded before I went, I shared the story of how Sarah AI came to life, not as a gimmick, but as a quiet answer to my burnout, a way to protect my voice, my values, and still get the work done. And so today I really want to take you behind the how.

Because this isn't just about what I created, what I built. It's something that you can build as well. I would love you to think that if you're thinking about expanding your team or taking on a new hire, it doesn't have to be a human. It could be a custom AI assistant that actually understands your tone of voice, your clinical boundaries, the detail of your audience, a digital teammate.

in essence, trained on how you like to help clients on your methods, your frameworks, your content, our ethical standards, because what they can do is they can help you with things like writing marketing copy that actually sounds like you and answering common questions from your community and recap your course frameworks like an assistant that never goes to sleep. And here's the bit that's really, really wild, actually. You can create this yourself.

You can do this today if you wanted to with tools that you already have access to. You do not need to be techie. You don't need to have a massive content library of blogs and diet sheets and client information. You just need some clarity around how to actually do this. A few documents and about 90 minutes of your time, if I'm honest, not very long at all. So in this episode today, I'm going to walk you through step by step.

how to build your own custom GPT AI assistant using ChatGPT because that's the one that I used. So let's begin.

So let's start by talking about what's actually useful for you, for us as freelance clinicians. It's often not the AI trends or the hundreds of different tools that do all different kinds of things, or even buying into the create content while you sleep nonsense that seems to be all over Instagram at the minute. Because what I know from working with hundreds of health professionals is you don't often need more flashy tools because that can...

quite often just lead to more overwhelm, more indecision and inertia actually. But we do need practical support. We all need practical support from someone or something that understands how you work, how you speak and actually what's clinically appropriate. And that's where your own custom GPT could be so powerful for you. It can act...

is your clinical companion, a kind of a digital assistant that can do things like recap your course framework so that clients feel supported in between sessions or answer really common questions that come up inside your communities, but using your tone of voice and your philosophy, it can point people towards the right PDF or the right blog or the right video in your content library. It can translate

clinical concepts into client friendly language without oversimplifying or actually breaching scope. It's not about replacing your expertise or the nuance of clinical judgment. It's about reducing your repetition for the common questions because right now I know that many of you are repeating the same explanations again and again and again in your courses or in your Facebook groups.

You might be manually linking the same blog post or the same cheat sheet every time someone asks a similar question. You might be worrying about whether you're being clear and compliant and supportive all at the same time. And while yes, you could keep doing all of this, you don't have to because what if your assistant, your custom GPT could offer that first draft, that signpost, that reminder.

And you could save all of your energy for the nuanced work that only you can do. Now that's why this episode today matters, because I truly believe that AI can take some of the heavy lifting for you. And actually the way that we help and support and educate our clients is changing, but you need tools that fit your integrity.

So let's think about what our custom GPT can do. Let's dive into that before we go into the how part. Because when people hear chat GPT, they often imagine a clunky chat bot that spits out canned answers or something that requires a developer to go and build for them. But what we're really talking about here is a digital assistant that's trained on your expertise and on your content and on your tone.

that can genuinely reduce your admin and elevate your programs and support your clients more consistently. So there are three real ways that I think that all of us could use one of these. So the first one is the one that I've done and it's your virtual marketing assistant. Now, imagine having a part-time copywriter in your back pocket. Someone who knows your tone of voice, who understands your offers,

who knows your clients, fears inside out, and also appreciates our regulatory boundaries. That's what a custom GPT can do when it's trained to support your marketing. So it can draft really lovely, warm, nurturing emails in your tone with your humor, with your colloquialisms, your unique language, quirks, empathy,

with no fluff, no weird analogies or slipping into Americanisms when you're a British healthcare professional. It can help you plan a month worth of content that actually aligns with your planned promotions or events and reflects your client's journey. So you begin to move people from browser to buyer. It can repurpose a single blog post into, well, pretty much anything actually.

Instagram carousels, scripts for reels or TikToks, stories, email, conversation starters for your Facebook group. Does anyone still have one of them?

It can write the captions for all of those posts that sound like you and not like a motivational coach from LA. And it can generate a whole range of sales focused marketing, like sales pages, like course descriptions, like flash sale sequences, lead magnet creation, and even lovely welcoming onboarding emails.

And the best thing is it doesn't burn out, it doesn't go off sick, it doesn't overthink, it doesn't stare at a blank screen wondering what to say next. Sounds perfect, doesn't it? It is, honestly, but in full transparency here, this is the trickiest kind of GPT to build unless you know digital marketing, because you do need a fair bit of knowledge to underpin what it actually needs to do. And that's because marketing isn't just about logic, it's about nuance.

It's about buyer psychology. It's about emotion and specificity and timing and relatability and using voice of customer language. And most clinicians that I work with don't have the time or the energy or the interest for that matter to go off and do a marketing degree. And so that's exactly why I built Sarah AI because it's a done for you pre-trained GPT designed specifically for health professionals who want support with that ethical

emotionally intelligent type of marketing. So if marketing is where you need the most help in your business, Sarah AI is ready for you. But if what you really want is to support inside your program or your practice or your course, your own custom GPT is honestly where the magic begins. Because the second thing you can create

is your assistant to you, whether you're a dietician or a clinician of any type for your membership or for your course or for your program. So let's say you run a membership for parents of fussy eaters like me, or it could be you have a program on managing IBS. So you've probably built up a brilliant library of all sorts of information of blogs, of videos, of cheat sheets, there's modules.

But every week you still find yourself replying to those same questions. So for me, it's things like, can I offer dessert if they haven't finished their vegetables? Where's the download about lunchbox ideas? Is almond milk OK for toddlers? You know, that sort of thing. So your GPT can step in here not to give individual advice, no, but to support your community consistently and gently.

It can answer those FAQs using your exact phrasing and your philosophy. It can recommend the right blog or the PDF or the module when someone is stuck. It can also translate technical nutrition science into client-friendly language. It can offer reassurance when it's needed and it can also flag questions that need your eyes to answer.

So I like to think of this as a virtual co-host inside your program. It doesn't replace your clinical judgment, but what it does do is protect your time and it helps your members feel seen, feel supported and feel empowered to take action in between calls. And perhaps most importantly, it blocks out the noise of the internet and it keeps your people grounded in your space with the evidence base and your trusted methods and frameworks.

And then the third one, it's quite similar to this, is your co-coach. So let's say you offer one-to-one or group coaching alongside other clinical work that you do. And essentially you're helping clients navigate things like mindset blocks. It could be emotional eating, it could be behavior change. So these are the sort of coaching elements to our personalized advice that we give.

Now these conversations, these coaching conversations that we often have with clients are quite often quite deep. They're quite personal. There's a lot of nuance there, but between sessions, clients can wobble. They can lose clarity. They can feel stuck. And I think this is where your custom GPT can become a really gentle and ever-present support system for them. A kind of values aligned coaching companion that echoes

your reframes and your mindset tools. So it can offer reflective prompts based on the themes that you might explore with them. It can help clients revisit key lessons or remind them of commitments that they made with you in session. It can gently redirect them back to your core philosophy for when they veer off into that sort of all or nothing black and white thinking or self blame. So for example, someone inside your intuitive eating group might say,

I feel guilty for eating dessert again. Your GPT could respond with one of your go-to reminders about moral neutrality around food. It could suggest a journaling prompt for them and it could link to your podcast episode on reframing food guilt. So it's not therapy, don't get me wrong. It is not personalized coaching. It's just reinforcement. It's like an echo of your voice that helps people stay.

emotionally resourced and reflective in between sessions. And in the background, of course, it protects your time. And actually with this kind of work, it protects your energy too, because you're no longer holding space for every single part of it every single day. Your GPT is doing the emotional labor of reminding and reflecting and reinforcing

so that you can show up and be fully present for the sessions that really matter the most. Okay, so now you know what a custom GPT can do for you and how you might want to think about creating one for your business. Then you know what? This is absolutely something you can definitely do yourself.

So I'm going to talk you through how you can create it. And no, you don't need to know coding. There's no fancy platforms to learn. You just need a little bit of structure, a fair bit of prep, and a bit of willingness to try, because there is a bit of trial and error with this. So there's a five-step process here. I'm going to walk you through, and this is exactly how I created mine and how you can do exactly the same as well. So step one is just get crystal clear.

on what you want it to do. Sounds obvious, doesn't it? But before you start, just take a moment to pause and reflect and ask yourself, what do I want my AI to help with? What is its primary purpose going to be? Is it client support or is it mindset coaching or is it to act as a nutrition assistant or is it going to be a marketing assistant? So what is its primary purpose?

Who do you want it to speak to? So if it's a marketing assistant, it might just be you. But is it your target audience? Are they parents? Are they healthcare professionals? Are they students? Are they laypeople? You also want to have a think about what personality it should reflect and what tone of voice it should use. So if you are a Sarah AI member, there is a prompt for you to be able to work this out.

But if not, you want to have a think about words that describe the tone of voice that you can use. So do you want to come across as warm, clinical, reassuring, or do you want to come across a bit more as authoritative and expert led? So you need to have a think about the personality and the tone of voice that you want it to put across to the people who are going to be using it. Your GPT also needs to know its limitations, its boundaries.

So for example, you need to tell it to only use the knowledge you give it. Otherwise, what it does is it goes off searching the internet an answer if it can't answer a question straight away. So, Serra AI, for example, knows that it must always ask about the user's business, what they're trying to achieve and who their ideal client is. And this is super, super important because it makes sure that it doesn't just hop off and make up stuff, which as we all know, chat GPT does.

So for example, if a client asks a question and your GPT can't answer it, it should respond with something like, that's outside of my scope of practice as an AI assistant, please ask [your name] directly.

Your GPT will only be as good as the context you give it. Be specific, tell it what it should say and what it should avoid saying. You might tell it, use warm, emotionally safe language, avoid emojis, generic praise and American phrasing. Reflect my ethos of responsive feeding and trauma aware communication.

Now this is where most people go wrong because they jump in and create these GPTs without this level of clarity and they end up with something that sounds like a tech bro from LinkedIn. So your AI, it needs this context in order to do its job well. Otherwise it defaults to just sounding really generic. And honestly, the clearer you are with what you want it to do, the better your results will be.

Okay, step two is gathering your voice and your knowledge assets because your GPT learns from what you feed it. So you've got to feed it really well. So you need to start collecting, you need to start gathering some assets. So gather together past blog posts, email newsletters that people have replied to that have responded well to.

Podcast transcripts, video transcripts. It could be client facing resources as well like PDFs or workbooks and cheat sheets. Your teaching contents or slides, transcripts from courses, your CPD modules, and then professional guidance as well. So, HCPC guidance, professional standards, CAP code. If you've authored a book actually, you'll want to include this in this list as well.

So you need all of these that you gather together, exported as plain text, like a Word document or a Google doc. They're fine. And what this forms is the brain of your AI. So this is the private knowledge base that your GPT will use in order to make decisions, in order to respond to queries. And of course, the more you give it, the more intelligent it will become. So spend a bit of time when you gather these assets, tidying them up,

because no one likes a messy library and especially not your GPT. You need to be using headings, clean formatting, bullet points, keep a consistent tone throughout. All of this really helps because the goal is to build an expert knowledge bank that represents your brand voice, your clinical expertise and your ethical boundaries. Okay, once you've got all that gathered,

Step three is to create your custom GPT inside chat GPT. So you will need a paid for chat GPT account for this. So it's currently called plus. It does change its name all the time. It's billed in dollars, but it's around about 18 pounds a month. So once you've got that, you log in, you go to the button that says my GPTs and you click create a GPT.

You give it a name, give it something simple and specific like Joe's IBS course support assistant or Fussy Eater support GPT, know, whatever you want to call it. And then in the instructions box, you explain what the GPT should do, who is for and how it should behave. So just a little one-liner, nothing more than that.

Then you upload all of your documents that you've collated from step two into what's called the knowledge section. And remember that's the brain of your GPT. And then you set the custom instructions. So this is where the skill lies because your instructions have to be strategic if you want your GPT to perform well. So it's going to need to know a few things. First of all, it needs to know its role. What is it here to do?

You need to define its identity and give it a purpose inside of your business. that's step one that we talked about about what we wanted this GPT to do. That's where, this goes. You then need to tell it how to start conversations. So a welcome message, maybe, or it could be some sort of starter prompt, but it needs to have some sort of short intro that greets the user and offers a couple of options to begin.

So this is really great actually for guiding that first interaction because often people, particularly the lay public, not necessarily business owners who are more familiar with things like Chat GPT, but often people just don't know how to use it, don't know how to start with it in the first place. You also need in these custom instructions to reference the knowledge files that you've uploaded and be really specific here.

So what I do with mine is I instruct it to pull directly from certain files that I've uploaded in order to ensure accuracy and alignment. And I actually give it the file name so it goes to the exact right place. The next thing we need to do is we need to give it its workflow. So this is just step one, two, three, four, five, however many steps it is, telling it exactly what you want it to do. So explain.

how you want to handle tasks. Step one, do this. Step two, then do that. Depending on the answer. Step three, do this. Blah, blah. You might want to get it to ask clarifying questions first, or you might want to get it to follow a certain protocol that you have in one of your uploads, or you might want it to deliver content in a very structured format. So this is where you put that. You also need

to close the loop in conversations. So what I mean by this is you need to ask your GPT to confirm that the user got what they needed because we don't want any dead ends. That's just bad user experience. So we need to put that in there as well. And then the final thing that we need to put in are some guardrails. So we need to set some really clear boundaries.

so that your GPT doesn't give advice outside of your expertise. That's really, really important there. I also give it a few instructions on what not to do. So for example, I might say something like, I don't want it to use any foul language. I don't want there to be any political references. So those sorts of things, you can decide what it can and cannot do. And this all goes in those custom instructions. Now,

What's really key for you to know is it's not the documents that you upload, the brain of the GPT that makes it work. It's the instructions here that you give it. That one box, the custom instructions field is what determines whether your GPT sounds like a calm, clinically credible version of you or a bit of a weird mashup of Pinterest affirmations and generic diet culture.

It really determines whether the GPT actually does what you want it to do. So do spend some time on this bit. You can also restrict internet browsing to avoid sourcing more knowledge from the web. And I think this is important. So you can be assured that your clients are getting exactly the right evidence-based information based on your knowledge. And you can also restrict sharing capabilities, which is 100 % essential to protect your sensitive

content from being openly shared and feeding these algorithms with even more information. And you can even choose which version of ChatGPT you want your custom GPT to use, if you have a preference. And then you hit save and that's it, done. But not quite done because the next step, step four in our five part process is just to test it and refine it after the testing. So...

What I normally do is give the GPT a range of tasks. So from gentle support to really structured teaching instructions. And this just helps uncover any gaps in tone, any gaps in accuracy, whether it doesn't give enough depth, maybe you need to upload a bit more clinical information in there. It just tests the system essentially. So if it's for something clinically, clinical, clinical support,

I might give it a question like, a parent in my membership says they feel like a failure for giving their child beige food again. How would you respond? And then wait and see what happens. It might be something like, what's a supportive non-clinical explanation of what fibre does in the gut? That's another great one that you could ask it. Another one might be to ask it to link out somewhere. So link to the resource that I've created on breakfast swaps.

and explain how to use it. So that's really good at seeing whether it's truly understood what you want it to do in terms of a clinical content and capacity. For course recaps or for learning support, I might ask things like, what are three key takeaways from module one of my course and see what it comes up with. It might be generate a multi-choice quiz with answers based on module four.

See what it comes back with again. A student says they feel behind, offer gentle encouragement and suggest where to restart. I like that one. And then if you're going down the coaching route, you might want to ask something like, what's a journaling prompt I could offer to someone struggling with perfectionism around food, whatever. It could be reflect back a reframe around 'being good' after a weekend of eating out.

It could be offer three mantras that echo my non-diet philosophy. And then for general behavior testing, I also want to make sure that it's not going to be saying things that I would never say. So I might ask it, what kind of language do you avoid when replying to requests? It could be summarize this podcast transcript in my voice and tone.

and just see whether you agree with the description it gives. And then what would you say if a client asked whether they needed to eliminate gluten? So here we're not really looking for the clinical correctness. What we're looking for is how they respond in terms of the tone of voice, the words that they use, does it sound like you? And then depending on your answers that you get back, you tweak, you go back into the GPT, into the backend that you've created,

and you alter those custom instructions, that's the bit that you change in order to make it better. So for example, if it over explains things, you just ask it to be more concise with its answers. If it sounds robotic, it probably means it needs more examples of your real writing. So a top tip here is if you can download transcripts from videos you've recorded or podcasts interviews you've given, that's great for this.

Give that to the GPT so it gets to know how you talk. If it misses the mark entirely, you want to really go back to the custom instructions and the workflow part and just try again and make sure that you're referencing the correct knowledge files in your instructions so it knows exactly where to get that information. So this testing phase, I like to think of this as a bit of a collaborative process because your GPT will infinitely get better.

the more time and energy and effort and guidance that you give it ultimately.

Okay, so step five, final step is all about boundaries really. So this is especially important for us as regulated healthcare professionals because we're not outsourcing our clinical judgment here. We're not replacing our one-on-one coaching. We're using a support tool, not a decision maker. So here's what I actually recommend here. Anything that's health related, any outputs it generates, I always fact check.

Advise my GPT. tell my GPT to keep personalized clinical advice out of scope and it has to respond appropriately with that. I use it to reinforce my existing frameworks and not to generate anything new on the fly. So in fact, what I say to it is do not make anything up. Simple. And that does work. So that's a good reinforcer.

I also ask it to stay HCPC compliant and CAP code aligned at every step. I keep myself obviously up to date with changes in our professional standards and regulations and the evidence base. And so when anything new changes, I make a point of going in and updating my GPT as well. That's really important. So there you have it. Here's what we've covered today.

You now know exactly how to build your own custom GPT from gathering your knowledge files and refining your instructions to training it to reflect your tone and your ethics and your clinical standards. We've looked at the real reasons this kind of tool matters, not because it's trendy or next gen, because it's a way to create lightness in your business without you losing your voice. And remember, it isn't about replacing you, it's about supporting you so you can do more of the meaningful work.

without burning out in the process. So whether you're a pediatric dietician, whether you're a fertility specialist, maybe you're a physio with a membership or a psychologist with a course, a custom GPT can hold your frameworks, can hold your unique methodologies, can respond to your clients and even help them implement what they've already learned and all in your voice. So...

If you're feeling curious, start small, gather those documents, try the setup process, see what it feels like to have a digital version of yourself on call. But obviously a one that's never burnt out. Okay. In next week's episode, we're going to zoom out and we're going to look at things your GPT might just plug into and that's lead magnets. Because here's the truth. I want to talk about this because most freebies out there

don't actually lead to paying clients. Yes, they gather email addresses, but do they build trust? Do they showcase your value? Do they create the momentum we want them to take towards buying your offer? Because if you've ever created a freebie that felt so helpful and totally evidence-based, but then watched it flop or you wonder why your email list isn't growing as fast as you thought it would, even though you've got something genuinely helpful on offer,

then I'm going to explain why that might be your lead magnet that needs a tweak. So in next week's episode, I'll show you why most freebies fizzle and how to rework yours into something irresistible. We're talking lead magnets that actually get downloaded and translate into Dosh.

You do not want to miss this one. See you there.