Prompting Claude Specifically
Most of prompting is universal. But a few techniques are especially effective with Claude — and a couple of habits from older models are now counterproductive. Here's what's actually Claude-specific.
Lean into these with Claude
- XML tags. Claude is well-tuned to structured prompts with named tags for inputs, instructions, and examples. This is the highest-value Claude habit.
- Be clear and direct. State the goal, constraints, and what "done" looks like. Claude follows explicit instructions closely.
- Give it a role via the system prompt.
- Let it think on hard problems — ask for step-by-step reasoning, or use extended thinking/effort on reasoning models (where you often don't need to ask for CoT explicitly).
- Use examples (few-shot) for format and edge cases.
- Prefill the start of Claude's response (on the API) to steer format — e.g. begin the assistant turn with
{to push JSON, or a heading to push structure.
Drop these old habits
:::warning Newer, literal-following models change the rules
- Ease off ALL-CAPS and threat-style prompts ("YOU MUST… OR ELSE"). Newer Claude models follow instructions literally and politely-phrased, specific instructions work better than shouting.
- Don't over-engineer with elaborate "jailbreak-style" scaffolding — clear, honest instructions outperform tricks.
- Less is often more — a tight prompt beats a sprawling one that buries the ask. :::
Claude vs other models
The principles transfer (see Cross-AI Prompt Translation), but the XML-tag affinity and prefilling are the things most worth remembering as Claude-specific. When you switch models, re-check tone intensity and formatting cues.