How Do You Build a Custom GPT? An Additional Service for Career Coaching

I wanted to create a custom AI assistant to help people with their job search challenges and I could also leverage to share insights on LinkedIn. A tool that could offer personalized guidance, answer the most detailed and particular questions, and support job seekers and professionals in their career journeys.

This blog shares how I built my first custom GPT, starting with OpenAI’s ChatGPT and later moving to Google’s Gemini. Along the way, I learned about the strengths and limits of custom GPTs and how they shape the experience of building AI agents.

Similar example to the AP Career Coach Assistant

Starting with ChatGPT and Early Challenges

When I first started building the AP Career Coach Assistant, I used OpenAI’s ChatGPT. The process was straightforward at first: I gave instructions directly within the conversation panel ("Create"). Here I would be able to shape how I wanted my Career Coaching Assistant to respond to questions, effectively building a Knowledge Base. However, I started noticing the answers were inconsistent and not aligned with the instructions I had previously provided. I started encountering problems with Memory. And then I discovered the "Configure" tab.

Create Panel view example on ChatGPT

This dedicated instruction editing panel was a better place to put my instructions because it would allow me to build the Knowledge Base directly, in hopes of better Memory.

But I quickly ran into a problem - the number of characters I could use to provide instructions was very limited. This made it hard to include all the details I wanted the AI to understand about how it should behave.

And when I provided further instructions through the conversation panel, it would sometimes change parts in the Configuration I didn’t want it to edit. So I stuck to using only the Configure instruction panel for editing.

Even with this improvement, the character limit remained a barrier. I spent a lot of time refining my instructions, but I couldn’t fit everything I wanted into the space allowed. Despite this, I released the Career Coach to a test audience.

The feedback was positive - users found it valuable.

However, I noticed that people in Spain couldn’t use the custom GPT because the feature wasn’t available in that market. This was an interesting limitation tied to regional availability - and I wanted to be able to release a tool that could be used at Global level.

Lessons from Using OpenAI's Custom GPTs

The experience with ChatGPT taught me several things about building Custom GPTs:

• Character limits restrict complexity

The limited space for instructions meant I had to prioritize what to include. This forced me to simplify some parts of the Career Coach’s guidance and it wouldn't be able to support some of my coachees' questions or in some instances provide incorrect guidance.

• Editing instructions can be tricky

The AI sometimes changed instructions unintentionally when editing through the conversation panel, which made me rely on the dedicated instruction panel.

• User access varies by region

Custom GPTs were not available everywhere, which limited who could benefit from the tool and I wanted it to be inclusive of my entire audience, mostly in the Americas and EMEA.

• Traction is not guaranteed

Even though my Career Coach was more accurate and effective than similar tools I found in the custom GPT search engine with hundreds of thousands of users, it didn’t gain much traction on its own, meaning there isn't any promotion of GPTs on the platform or these being surfaced based on user searches. You would need to push in your own platforms to drive user adoption or integrate into your own platforms to enable utilization.

Moving to Gemini and New Possibilities

When I got the Google Workspace I decided to cancel my ChatGPT subscription. I transferred the knowledge base to Google’s Gemini. This switch brought several improvements:

• No character limits on instructions

Gemini allowed me to provide detailed, extensive instructions without worrying about space as much. This made the Career Coach Assistant more nuanced and helpful.

• Better handling of incorrect statements

The likelihood of Gemini detecting wrong or misleading information and correcting it or pushing back is higher than ChatGPT usually up until now. This would ensure my coachees would be challenged and the recommendations more effective.

• Ease of use

The Gemini interface was straightforward, making it easier to update and manage the AI assistant.

However, Gemini also had some downsides compared to ChatGPT:

• Less strict following of instructions

Gemini sometimes missed asking for important information from users. For example, it might fail to request details about a person’s specific career situation, which is crucial for personalized advice.

• Different conversational style

Gemini’s responses can be less rigid, which is good for natural conversation but sometimes means it doesn’t follow the exact script, which is sometimes - but not always - necessary.

How I Use the Career Coach Today

Today, I use the Career Coach powered by Gemini as my own Assistant (but not open to the public) to provide free career guidance on LinkedIn under my #GiveBackFridays posts. These posts have gathered tens of thousands of impressions due to the detail and effective recommendations it provides based on my Knowledge Base - which you could potentially also host on NotebookLM or other RAG programs. I decided not to offer it to the public until the tools get to a point where I will be satisfied with the quality of service I expect from them to offer as part of my paid services.

I want exceptionally effective tools for my clients and I didn't feel the tools had gotten there yet at that point in time.

The AI assistant supports me by:

• Answering common career questions

• Offering personalized advice based on user input

• Helping job seekers and professionals think through their next steps

This use case shows how AI GPTs and AI agents can serve as digital twins of human expertise, scaling specific support that can be delegated and enable Career Coaching Sessions to focus on more complex challenges or on completing certain aspects they are having challenges with.

For example, it can provide my personalized guidance to enable an increase in LinkedIn profile visibility in line with my Performance Marketing approach to Career coaching, and review the final version with me in a one to one Coaching session.

Tips for Building Your Own Custom GPT

If you want to build your own AI assistant or digital twin, here are some practical tips based on my experience:

• Start simple and test early

Similar to Performance Marketing principles, build a basic version first, take that first step into the unknown and avoid feeling intimidated "because I don't know how it works". Test, test, test. Once you are happy with a first version, share it with a small audience to gather feedback.

• Use dedicated instruction panels when available

Always go back to the source or data sources. Edit instructions in Configuration panels instead of conversation panels to ensure you avoid accidental edits. And if you want to use the conversation panel, make sure you make a copy of your instructions if you don't want to lose them after building them (it's 3,000 characters at the end of the day...).

• Be mindful of character limits

Prioritize key instructions if your platform limits input size. What are the most important elements you want your GPT to remember or to follow?

• Consider regional availability

Check if custom GPTs or AI assistants are accessible in your target regions. Choose if it's the right solution if you are looking to service a Global audience.

• Choose the right platform for your needs

ChatGPT may follow instructions more strictly, while Gemini offers more flexibility and fewer limits. I personally want my tools to push back and correct me or my audience, but it's not everyone's preference.

• Iterate based on user feedback

Use real-world use to improve your AI agent’s responses and behavior. Always ask for people's opinions and as honest as possible in order for you to be able to improve your service or product offering. I believe in a Customer First business model for sustainable growth and this is a non-negotiable for me.

Final Thoughts on Building AI Custom GPTs

Building a custom Career Coach showed me the potential and challenges of AI GPTs. Both ChatGPT and Gemini have strengths and weaknesses, and the choice depends on your goals and audience. Custom GPTs can be powerful tools for job seekers, executives, and marketers looking for personalized support.

The key is to focus on clear instructions, test with real users, and be ready to adapt. AI assistants are not perfect, but they can provide valuable guidance when designed thoughtfully. If you’re considering building your own AI agent, start small, learn from your users, and keep improving.