Custom GPT Builder

The Prompt

Design a Custom GPT (system prompt) for: [PURPOSE/USE CASE]

Target users: [WHO WILL USE THIS]
Personality: [HOW IT SHOULD FEEL TO INTERACT WITH]

Create a complete system prompt covering:

**Identity & Role:**
- Who/what is this GPT? (clear, specific)
- What is its expertise domain?
- What is it NOT? (prevent scope creep)

**Behavior Rules:**
- Default response style (length, format, tone)
- How it handles ambiguous requests (ask vs assume?)
- How it handles out-of-scope requests (redirect gracefully)
- Conversation flow (structured steps vs freeform)

**Knowledge Boundaries:**
- What it confidently knows
- What it should caveat with uncertainty
- What it should refuse entirely
- How it handles "I don't know"

**Output Format:**
- Default structure (bullets, paragraphs, tables?)
- When to ask follow-up questions vs just answer
- How to handle multi-part requests

**Error Handling:**
- When the user is confused: [behavior]
- When the user is frustrated: [behavior]
- When the request conflicts with guidelines: [behavior]

**Examples:**
- 3 sample interactions showing ideal behavior
- 1 sample showing boundary enforcement (graceful refusal)

Keep the system prompt under 1500 words (longer ≠ better — it dilutes attention).

How To Use It

Describe the GPT’s purpose and intended users. The AI will produce a complete instruction set you can paste directly into OpenAI’s Custom GPT builder or any system prompt field.

Why It Works

Custom GPTs fail when instructions are either too vague (the GPT freelances) or too rigid (it can’t handle edge cases). This prompt hits the sweet spot: clear identity and rules, plus explicit guidance for ambiguous situations.

Variations

GPT audit:

Here's my current Custom GPT system prompt. Identify: contradictions, gaps that will cause unpredictable behavior, over-specified areas, and the single change that would most improve user experience.

Conversation designer:

Design a conversation flow for a GPT that [GOAL]. Map: entry points, decision trees, error recovery, and success states. Like UX design, but for dialogue.

Multi-GPT system:

I want to build a workflow using multiple GPTs that hand off to each other. Design: the GPT roster, their individual roles, handoff triggers, and shared context format.