You might know by now that I’m pretty good at spotting details in my work. And as someone who is currently making a SUPER specific GPT and automation flow for a female coach who does social media content, I’m seeing just how specific we need to be in our Custom GPT instructions.
Let me be clear..
🫧 If we write in the instructions for our GPT that are basic:
💛 “You must follow the uploaded documents exactly”
and
💛 “If something is not covered in the documents, ask me before answering”
You’d think that’s enough for it to do that?
No! Go one step further… also say
✅ “Always provide helpful, accurate answers based on these files”
✅ “If a question is outside of the knowledge base, respond that you don’t have that information and tell us which document needs to be updated to get the best possible answer”
YOU’D THINK THAT WOULD BE ENOUGH? But no!!
We need all of the above AND
❤️🔥Guardrails and examples, named and referencing the knowledge by name
❤️🔥 Instructions that use positive language like ALWAYS USE [YOU] INSTEAD OF [SOMEONE] or [FRAMEWORK REFERENCE] NOT A RESEARCHED OR HALLUCINATED RESPONSE.
❤️🔥 Tell it what its purpose is and how it serves people and what it is not for. “YOU ARE THE ALIGNED AI CUSTOM GPT ASSISTANT TRAINED TO ANSWER COACH’S QUESTIONS ON HOW TO INCREASE REVENUE WITH CUSTOM AI AND AUTOMATIONS.
So, your might say “YOU ARE THE [YOUR BUSINESS NAME] CUSTOM GPT TRAINED TO [ANSWER YOUR CLIENTS SPECIFIC QUESTIONS] ON HOW TO [POSITIVE OUTCOME] [
USING THE [NAMED FRAMEWORK YOU’VE ADDED YOUR KNOWLEDGE DOCS FOR]
In addition to the other points. Phew! 😮💨
I hope this helps and if you didn’t already know,
longer instructions can use more credits if very long, and should always be accompanied by smaller, easier to read documents in either .md (Markdown) or .txt (Plain text) format.
Happy Instructing!! 😂