Prompt Engineering: How to Write Good AI Prompts in 2026

The quality of your results with [ChatGPT, Claude](/en/blog/chatgpt-vs-claude-assistant-choisir) or [Gemini](/en/blog/gemini-google-guide-complet) depends as much on your prompt as on the model itself. Prompt engineering is the discipline of formulating effective instructions to get the best possible outputs.

## The Basics: What Makes a Good Prompt

An effective prompt generally contains:

1. **A role**: "You are a GDPR expert" → the model adapts its register and precision
2. **Context**: the situation, constraints, what already exists
3. **A specific task**: clear action verb (write, analyze, compare, list, summarize)
4. **An output format**: "in 3 paragraphs", "as a table", "in JSON"
5. **Constraints**: length, tone, technical level, language

## Technique 1: Chain of Thought

Ask the model to "think step by step" before giving a final answer. Particularly effective for complex problems, calculations, and multi-step reasoning.

Example: "Explain your reasoning step by step before giving the answer."

## Technique 2: Few-Shot (Examples)

Give 2-3 examples of the expected output before posing the real question. The model understands the pattern and reproduces it.

## Technique 3: Maximum Contextualization

LLMs have no context about your situation by default. The more you provide, the better the response.

Bad prompt: "Write a follow-up email"
Good prompt: "Write a B2B follow-up email for a SaaS prospect who downloaded our white paper 7 days ago. Warm but professional tone, 3 paragraphs, call to action for a demo."

## Technique 4: Task Decomposition

For long and complex tasks, break into several sequential prompts rather than one massive prompt. For content production, pair with [SurferSEO](/en/blog/surferseo-optimisation-contenu) for SEO optimization.

## What LLMs Don't Do Well Despite Good Prompts

- Precisely calculate complex numbers → use Code Interpreter / Python
- Access real-time data → use [Perplexity](/en/blog/perplexity-ai-guide-complet) or tools with web access
- Remember previous conversations → all context must be provided in the prompt

## Our Trust-Vault Assessment

Prompt engineering is a cross-cutting skill that improves the use of all tools in our [catalog](/en/). For the full evaluation of AI tool capabilities, see our [methodology](/en/pages/methodology).