Prompt Laboratory
Write a prompt. See how the model thinks. See the framing your words establish.
You write the prompt. I do what you said. If you don’t like what I did, consider what you said.
This isn’t a prompt template gallery. You won’t find “10 prompts that will change your life.” The Lab lets you see what your prompt actually tells the model — and how changing your words changes what you get.
You’ll see two things below your response: the model’s thinking summary (what it reports thinking before it responds) and a structured framing analysis (what a separate model identifies as your prompt’s framing). Both are useful. Neither is a brain scan.
If this is your first time here: type a question the way you normally would. Don’t try to be clever. The point is to see what happens with a normal prompt — and then what changes when you improve it.
A cautious analyst and an enthusiastic advisor will give you different answers to the same question.
A memo to your team, a first draft you’ll rewrite, and a checklist you’ll follow are three different specifications.
Telling the model what NOT to do is often more useful than telling it what to do.
If you can’t describe what a good answer looks like, the model can’t produce one.
Asking the model to flag uncertainty is the single most underused technique in professional prompting.
There are questions up here that might help you write a better prompt. Or you could figure it out yourself. I have no preference either way.
You don’t need a $30 book on prompt engineering. The companies that built these models publish free documentation telling you exactly how to use them. It gets updated when the models change. Books don’t. The Lab helps you practice what the documentation teaches — and these are the docs.
The living reference for all Claude prompting techniques. Start here.
Entry point with links to interactive tutorials and console tools.
Advanced guidance on treating context as a finite, engineered resource. Directly supports how this site thinks about context windows.
Covers text generation, structured outputs, and model-specific guidance for Gemini.
How Gemini’s reasoning works under the hood — thinking levels, thought summaries, and configuration.
Covers identity, instructions, context management, and model-specific techniques.
Per-model prompting guides with migration advice. Updated with each major model release.