How to Optimize Prompts for Claude 3.5 Sonnet and GPT-4o

The Evolution of Prompt Engineering in 2026

As large language models like Anthropic's Claude 3.5 Sonnet and OpenAI's GPT-4o become more advanced, the way we communicate with them determines the quality of their output. Simple one-sentence prompts often yield generic results. To unlock their full potential, developers and writers must use structured prompt engineering techniques. This guide covers how to design prompts that yield professional, production-ready code and high-quality content.

1. Use XML Tags for Structure (Claude's Specialty)

Claude models are trained to respond exceptionally well to XML tags. Using tags like <context>, <instructions>, and <examples> helps the model distinguish between different parts of your prompt, preventing instruction drift.

<context>
You are a senior React developer refactoring a legacy state management system.
</context>

<instructions>
Refactor the following component to use React Context. Keep all styling intact.
</instructions>

This clear separation prevents the model from confusing your instructions with the source code you want it to process.

2. Specify the Output Format Explicitly

Both GPT-4o and Claude 3.5 Sonnet perform better when you define the exact output schema. If you need JSON, provide a sample template. If you want code, specify the folder structure and language version. For text writing, define the tone, perspective, and reading level.

3. Leverage Local-First Utilities for Secure Prompts

Before submitting text or data payloads to AI models, it is essential to ensure you are not leaking API keys, personal credentials, or customer data. You can prepare and sanitize your data inputs locally in your browser. For example, use our Remove Duplicates tool to clean up raw lists, or format text with the Case Converter before sending it as a prompt variable. This ensures your data remains secure and private on your machine.

Conclusion

Prompt engineering is about reducing ambiguity. By structuring your prompts, separating code blocks with tags, and sanitizing raw data inputs locally, you can achieve predictable, high-value AI generations. Check out our full suite of Developer Tools to optimize your day-to-day workflow.