Lesson Content

The cost of wrong-direction output.

When Claude misunderstands your intent and produces a long, detailed, well-written response to the wrong question – that is still wasted time. You have to re-read it, figure out what went wrong, and restart. For short tasks, this is minor. For long documents, complex analysis, or multi-step plans, a wrong-direction output is a significant setback.

The solution is not to write longer prompts every time. The solution is to use two powerful techniques that catch misalignments before Claude commits to an answer.

Technique 1: Ask Claude to state its assumptions.

Add this to any complex prompt: "Before you answer, list the key assumptions you are making about my request."

Claude will surface the interpretations it chose when your prompt was ambiguous – and you can correct them before it proceeds. This is especially valuable when:

• Your prompt could be interpreted in multiple ways
• The technical domain is one where wrong assumptions produce wildly different outputs
• You are asking about strategy, planning, or decisions where framing matters enormously

Example addition to any prompt: "Before answering, briefly state what you are assuming about [my role / the audience / the scope / the technical context]. I will confirm or correct before you proceed."

Technique 2: Ask Claude to request clarification first.

For complex, high-stakes, or genuinely ambiguous requests, you can explicitly tell Claude: "Before you begin, ask me any clarifying questions that would help you give a better answer. Wait for my response before proceeding."

This inverts the usual pattern. Instead of Claude guessing and you correcting, Claude asks and you inform. The result is almost always a more targeted first output.

Use this technique when:

• You are working on something with significant real-world consequences
• The task has many possible valid approaches and you want Claude to use yours
• You are not sure you have given enough context but are not sure what is missing

Technique 3: Review Claude's reasoning, not just its conclusion.

Ask Claude to "think through this step by step" or "show your reasoning." This lets you spot where Claude went wrong before accepting the output. A wrong conclusion from correct reasoning tells you something different from a wrong conclusion from flawed reasoning – and you cannot tell the difference if you only see the output.

When not to use these techniques.

For simple, low-stakes tasks – "summarize this paragraph," "rephrase this sentence," "suggest five names for this product" – assumption-surfacing and clarification requests add friction without benefit. Use them for complex, high-stakes, or ambiguous prompts where a wrong direction is costly.