The Scratchpad Framework is a structured prompting methodology that makes AI reasoning completely transparent. Instead of just getting an answer, or even a summary of reasoning like you see in the newer thinking/reasoning models, you see the complete analytical process: how the AI interpreted your question, what it considered, what it ruled out, what assumptions it made, and how it arrived at its conclusion.
Think of it like this: Traditional AI prompting is like getting a consultant's final report with no supporting analysis. Standard reasoning models show you a summary of their thinking. The Scratchpad Framework shows you their complete research notes, their reasoning process across structured analytical sections, and their quality checks before they give you the recommendation.
The result: You can trust outputs because you can evaluate the thinking behind them. And when AI has to show its complete work, the quality of that work goes up significantly.
Recommended models: I get the best results with Perplexity using this framework. Claude is a strong second option. Both handle the structured reasoning sections well.
There are two ways to use this framework:
Copy the prompt below and paste it directly into your chat, then add your question underneath it. Good for one-off queries.
Add the prompt to your AI tool's custom instructions so it automatically applies to every conversation:
Once set up, every chat in that Space or Project will automatically use the scratchpad framework. This is the best way to use it long-term.