45.1 The model ignores instructions

Symptom

You said "only return JSON" but it adds explanation. You said "be concise" but it's verbose. The model does its own thing.

Common Causes

• Lost in long context: Rules at the top get forgotten
• No examples: Model doesn't understand what you want
• Conflicting instructions: "Be helpful" overrides "be brief"
• Weak wording: "Try to" vs "You must"

Fixes

// Fix 1: Put critical rules at the END
const prompt = `
${contextAndExamples}

CRITICAL RULES (MUST FOLLOW):
1. Output ONLY valid JSON
2. No explanations before or after
3. If unsure, return {"error": "reason"}
`;

// Fix 2: Show, don't tell
const prompt = `
Return JSON like this example:
{"name": "Alice", "score": 95}

NOT like this (no explanations):
The answer is: {"name": "Alice", "score": 95}

User input: ${input}
`;

// Fix 3: Use response_mime_type (Gemini)
const response = await model.generateContent({
  contents: [{ role: 'user', parts: [{ text: prompt }] }],
  generationConfig: {
    responseMimeType: 'application/json'  // Forces JSON
  }
});

// Fix 4: Stronger wording
// Weak:
"Try to be concise if possible"
// Strong:
"Maximum 50 words. No exceptions."

Where to go next

• 45.2 Outputs are correct but messy