I’ve been experimenting with ways to reduce hallucinations and clarify the reasoning process in ChatGPT responses—especially for users who value transparency, consistency, and safe AI interaction.
This led to a reusable prompt template that structures answers in four parts:
Reinterpretation of the question
Source of information (Web or model memory)
Reasoning and judgement (how the answer was formed)
Final answer (proposal or conclusion)
The difference in output—both in quality and structure—was quite significant.
In my tests, using this format increased factual clarity, made it easier to separate speculation from fact, and provided more confidence when sharing AI-generated answers with others.
The guide includes:
- Comparison examples (template vs. no-template)
- Tips for reintroducing or skipping the structure
- Advanced prompts for analysing reasoning or prompting personality
- Japanese and English versions of the full markdown guide
GitHub: https://github.com/kou-saki/Tips-A-Template-for-Safe-and-Reliable-Conversations-with-ChatGPT
I’d love to hear if others have tried similar approaches—or if this helps you improve your own workflow with ChatGPT.
Feedback, improvements, and adaptations are more than welcome!
（参考 / For Japanese readers）
ChatGPTの出力における幻覚（hallucination）を抑え、事実と推論、判断を分けて表示させるための4構造テンプレートを作成しました。
特に「AIとの対話の信頼性」を気にする方や、他人にAIの回答をシェアする必要がある場面に向いています。
- 質問の再解釈
- 情報ソースの明示（Webか記憶ベースか）
- 推論と判断の過程
- 結論としての提案または回答
GitHub上には、日本語版も含めたテンプレート全文、応用例、再提示タイミング、スキップ方法などを掲載しています。
ぜひご活用ください：
https://github.com/kou-saki/Tips-A-Template-for-Safe-and-Reliable-Conversations-with-ChatGPT