Editing Openai/6969614e-00e4-8001-ac42-dda7410c3c4a (section)

=== Assistant: txt ===
<syntaxhighlight lang="txt">SUMMARY — Optimizing ChatGPT for Analysis, Study, Reasoning, Critique

CORE IDEA
• Prompting is calibration, not control.
• Consistency comes from stable defaults + repeatable structure, not one “perfect” prompt.

CUSTOM INSTRUCTIONS vs CHARTER
• Custom instructions = slow-changing traits (tone, rigor, skepticism, depth).
• Keep them short and principled.
• Use a reusable “charter” for serious work (role, standards, methods, output norms).
• Charters are more reliable than long custom instructions alone.

PROMPT STRUCTURE THAT WORKS
Use prompts as specifications, not requests:
• Context – assumptions
• Task – what to do
• Constraints – limits, standards, what to avoid
• Method – how to reason
• Output format – to prevent drift
Reusing the same structure improves consistency.

STANDARDIZATION > META-COMMANDS
• Avoid vague meta-prompts (“be concise,” “think like a professor”).
• Use concrete rules (sentence limits, step numbering, explicit criteria).
• Reuse scaffolds instead of rewriting prompts each time.

WHAT DEGRADES PERSONALIZATION
• Overloaded or contradictory custom instructions
• Mixing incompatible modes without declaring a switch
• Treating every task as unique
• Failing to explicitly correct misunderstandings
• Relying on memory for workflows instead of templates

PROMPTING AS A SYSTEM
• Stable identity (custom instructions)
• Reusable scaffolds (analysis / critique templates)
• Explicit mode switches (“adversarial critique mode”)
• Periodic checks for drift

IMPORTANT INSIGHT
• Recent instructions override stored preferences.
• Put charters first and restate critical constraints when stakes are high.

BOTTOM LINE
Consistency is engineered, not remembered.
Use minimal custom instructions, a reusable charter, standardized prompt structures,
explicit mode control, and active correction to get reliable long-term behavior.

</syntaxhighlight>