Prompt Engineering Guide for Claude

Claude is Different (In a Good Way)

If you've been using ChatGPT and just switched to Claude, you might notice something: the same prompts don't always work the same way.

Claude has its own personality. It's more... thoughtful? Less eager to please? It'll actually push back if you ask for something sketchy, and it tends to give more nuanced answers.

Here's how to work with that, not against it.

Claude Loves Structure

Seriously, this AI is a sucker for well-organized prompts. XML tags work surprisingly well:

<role>You're a senior product manager at a SaaS startup.</role>

<task>Write a PRD for a new feature</task>

<constraints>

</constraints>

Is it weird to write prompts in XML? Maybe. Does it work? Absolutely.

Ask It to Think Out Loud

Claude is great at reasoning, but sometimes it jumps to conclusions. Add "Think through this step by step" or "Explain your reasoning" and watch the magic happen.

This is especially useful for:

The "Constitutional AI" Thing

Claude was trained differently than other AIs — it has built-in principles about being helpful, harmless, and honest. This means:

Work with this. Ask for balanced analysis. Request counterarguments. Claude excels at this stuff.

Make It Even Better

If you're using Claude regularly, check out Aetherify's Architect mode. It asks you 3 quick questions about your prompt and builds something optimized for Claude's quirks.

Way faster than learning all this yourself. Just saying.

Also — if you want pre-built Claude prompts, we've got a bunch in the Discover section.