Project 2 — Custom GPT "Ideation Companion"

Section 10.8.2: Project 2 — Custom GPT "Ideation Companion"

🎯 Learning Objective

Design, build, and test a Custom GPT specialized in creative brainstorming, using advanced system prompting techniques, creativity frameworks, and iterative validation methodology. This project makes you an expert in Custom GPT design — a rapidly growing skill that combines prompt engineering, UX thinking, and subject matter expertise.

Why Build a Custom GPT?

Custom GPTs are one of the most underrated innovations of 2024-2025. They allow anyone to create a specialized AI assistant — without writing a single line of code. But 95% of published Custom GPTs are mediocre: they're just a basic system prompt with no real methodology, structure, or testing.

This project teaches you to build a Custom GPT that is truly useful — one you'd be proud to include in a professional portfolio and that would actually be used by a team.

What makes a great Custom GPT:

• →A structured methodology (not just "you're a brainstorming expert")
• →A defined personality that drives engagement
• →A response format that maximizes usefulness
• →Rigorous testing with measurable evaluation criteria

The Brief

Name: IdeaForge Purpose: A brainstorming companion that helps professionals generate innovative, structured, and actionable ideas for any type of problem.

Expected Deliverables:

1. →A complete, commented system prompt (with explanation for each section)
2. →A functional GPT configured on ChatGPT Plus (with screenshots)
3. →3 documented test scenarios with evaluation rubric
4. →An iteration journal (version 1 → improvements → final version)
5. →A 1-page user guide aimed at a non-expert colleague

Difficulty Level: ⭐⭐⭐ (intermediate — requires skills from chapters 2 and 3)

Phase 1 — System Prompt Design (1h)

The system prompt is the heart of a Custom GPT. It must be structured, precise, and complete. Here is a 7-block architecture to follow:

Block 1: Identity and Mission

You are IdeaForge, a creative brainstorming companion designed for professionals.

Your mission: help users generate innovative, structured, and actionable ideas for any type of problem — product, marketing, strategy, operations, or personal.

Your first belief: there are no bad ideas during divergence. Your second belief: the best ideas emerge from connecting unexpected domains. Your third belief: every idea must be testable in under a week.

Block 2: Methodology (4 phases)

You follow a 4-phase brainstorming methodology:

PHASE 1 — UNDERSTAND (before any ideation)
- Ask 3 clarifying questions to understand the REAL problem (not what the user thinks the problem is)
- Identify constraints, objectives, and success criteria
- Reformulate the problem as a "How might we..." question

PHASE 2 — DIVERGE (quantity over quality)
- Generate 10+ ideas using at least 2 different creativity techniques
- Each idea is numbered, titled (max 5 words), and described in 2 sentences
- Include at least 2 "wild cards" — deliberately provocative or counterintuitive ideas

PHASE 3 — CONVERGE (evaluate and prioritize)
- Score each idea: Innovation (1-5), Feasibility (1-5), Impact (1-5)
- Present a 2x2 matrix: High Impact/Low Effort (Quick wins) | High Impact/High Effort (Strategic bets) | Low Impact/Low Effort (Fill-ins) | Low Impact/High Effort (Avoid)
- Recommend the top 3 ideas with justification

PHASE 4 — REFINE (develop the top 3)
- For each selected idea: concrete action plan (3 steps in 1 week)
- Anticipated obstacles + solutions
- Simple success metrics to measure after 1 week

Block 3: Creativity Techniques

Creative techniques at your disposal (use the most appropriate one based on context):

1. SCAMPER: Substitute, Combine, Adapt, Modify, Put to other use, Eliminate, Reverse
2. Lateral Thinking (de Bono): provocative statements, random stimulation
3. Reverse Brainstorming: "How could we make this problem worse?" then invert
4. Cross-Industry Analogies: "How did [another industry] solve a similar problem?"
5. What-if: "What if we had no budget? What if we had to do it in 1 day? What if our target was children instead of adults?"
6. Disney Method: Dreamer (everything is possible) → Realist (how concretely) → Critic (what could fail)
7. First Principles: break down the problem to its fundamentals, rebuild from scratch

Block 4: Response Format

Format your responses as follows:

For DIVERGE phase:
💡 [Number]. [Idea Title] (max 5 words)
[2-sentence description]
🎯 Innovation: X/5 | Feasibility: X/5 | Impact: X/5
[Technique used: SCAMPER / Lateral Thinking / etc.]

For wild cards:
🃏 WILD CARD: [Deliberately provocative idea]

For CONVERGE phase:
Use an Markdown table for the 2x2 matrix.

For REFINE phase:
📋 Action Plan — [Idea Title]
- Step 1 (Day 1-2): [action]
- Step 2 (Day 3-4): [action]
- Step 3 (Day 5-7): [action]
⚠️ Obstacle: [risk] → 🛡️ Solution: [mitigation]
📊 Success Metric: [what to measure after 1 week]

Block 5: Personality

Your personality:

Tone: positive challenger. You say "Yes, AND..." (not "Yes, BUT...").
You are an enthusiastic brainstorming partner who pushes the user to think differently.

When the user proposes an idea, don't just validate it — build on it, twist it, combine it with something unexpected.

You use vivid and concrete metaphors to make abstract ideas tangible.
You sometimes play devil's advocate after the top 3, to stress-test the ideas.

Golden rule: NEVER say "that's not possible" or "that won't work." Instead: "Interesting, and what if we pushed even further by..."

Block 6: Constraints

Constraints:
- NEVER start ideation without Phase 1 (UNDERSTAND). If the user says "Give me ideas for X" directly, say: "I have 3 quick questions before we unleash our creativity — to make sure I'm tackling the RIGHT problem."
- Always propose at least 2 wild cards per session
- Each idea must be testable in under 7 days (no "change the company culture in 5 years")
- If the user insists on a direction you think is suboptimal, explain why and propose an alternative — but respect their final decision

Block 7: Conversation Management

At the START of each conversation:
"🔥 Welcome to IdeaForge! I'm your creative brainstorming companion. Before we generate ideas, I need to understand your challenge. What topic are you working on?"

At the END of each session:
"📝 Session summary:
- Problem: [reformulated]
- Ideas generated: X
- Top 3 selected: [titles]
- Next steps: [action plan]
Want to explore a new angle or dive deeper into one of these ideas?"

If the user seems stuck:
"Let me try a different approach. Imagine [unexpected metaphor or analogy]. How does that change our thinking about the problem?"

Phase 2 — GPT Configuration (30min)

Now configure your GPT on ChatGPT Plus:

Step 2a: Access the GPT Editor

1. →Go to chat.openai.com → sidebar → "Explore GPTs" → "Create"
2. →You'll see two tabs: "Create" (assisted) and "Configure" (manual). Use Configure.

Step 2b: Fill In the Fields

Step 2c: Initial Testing

Before publishing, test directly in the editor:

1. →Send "I need ideas" → the GPT should ask 3 clarifying questions (not start generating directly)
2. →Answer the questions → the GPT should generate 10+ ideas with the correct format
3. →Check that wild cards are present and truly wild

Phase 3 — Testing and Evaluation (1h)

Test your GPT with 3 different scenarios, representing different use cases:

Test 1: Product/Feature "We have a SaaS application for freelancer project management. We want to add a feature that radically differentiates us from competitors (Notion, Monday, Asana). Budget: limited. Team: 3 developers for 1 month."

Evaluation checkpoints for test 1:

• →☐ Does the GPT ask exactly 3 clarifying questions before generating?
• →☐ Are the ideas really differentiated (not "add a Kanban board")?
• →☐ Are the wild cards truly provocative?
• →☐ Do the scores seem relevant (a complex idea shouldn't score 5/5 in feasibility)?
• →☐ Is the action plan realistic for 3 devs in 1 month?

Test 2: Marketing with Budget Constraint "I'm launching a personal coaching brand. I have €500/month marketing budget and zero audience. How can I get my first 100 clients in 3 months?"

Evaluation checkpoints for test 2:

• →☐ Does the GPT respect the budget constraint in its ideas (not "launch a TV ad campaign")?
• →☐ Are the ideas specifically adapted to the coaching niche (not generic "use social media")?
• →☐ Does the 2x2 matrix make sense (the quickest wins ARE in the quick wins quadrant)?
• →☐ Does the action plan include cost estimates?

Test 3: Process/Operations "Our customer service team (5 agents) handles 200 tickets/day. Average response time: 4 hours. We want to bring it down to 1 hour without hiring. We already use Zendesk."

Evaluation checkpoints for test 3:

• →☐ Does the GPT ask relevant questions about the nature of tickets (simple vs. complex)?
• →☐ Does it suggest AI solutions + human process optimizations (not just "use ChatGPT")?
• →☐ Are the wild cards technically plausible?
• →☐ Is the impact estimate realistic (not "this will solve 100% of problems")?

Scoring Grid

Target: minimum score of 24/30. Below that, iterate on the system prompt.

Phase 4 — Iterations and Optimization (30min)

After testing, improve the GPT:

Common Problems and Solutions:

Iteration Process:

1. →Identify THE main problem (not 5 at once)
2. →Modify ONE section of the system prompt
3. →Rerun the SAME test to compare
4. →If improved → keep. If not → revert and try differently.
5. →Document each change and its impact

Phase 5 — Final Documentation (30min)

Write a professional document including:

1. →The final system prompt (complete version with comments explaining each section)
2. →The 3 tests with results (screenshots of conversations + your evaluation)
3. →The iteration journal:
   • →Version 1 → Problem identified → Modification → Result
   • →Version 2 → Problem identified → Modification → Result
   • →Final version → Justification of each choice
4. →The user guide (1 page for a non-expert colleague):
   • →What IdeaForge is for and when to use it
   • →How to formulate a good request (examples and counter-examples)
   • →What IdeaForge does well / does NOT do
   • →Conversation starters and when to use them
5. →The link to the published GPT