← Back to Blog
March 15, 2025·3 min read

10 Prompt Engineering Techniques That Actually Work

Practical, battle-tested techniques for getting dramatically better results from ChatGPT, Claude, and Gemini.


Beyond the Basics

Once you understand the fundamentals of prompt engineering, the next step is building a toolkit of specific techniques. These are patterns that work reliably across different models and tasks.

Here are ten techniques that consistently produce better outputs.

1. Chain-of-Thought Prompting

Ask the model to "think step by step" before giving an answer. This simple instruction dramatically improves performance on reasoning tasks, math problems, and multi-step analysis.

Example: "Solve this problem step by step: [problem]"

2. Role Prompting

Tell the model to act as a specific expert. This shapes not just the tone but the depth and angle of the response.

Example: "You are a senior UX designer with 10 years of experience. Review my app wireframes and give feedback."

3. Few-Shot Examples

Show the model what you want by including two or three examples before your actual request. The model will pattern-match and apply the same structure to your input.

4. Structured Output Requests

Ask for a specific format explicitly. JSON, markdown tables, numbered lists, or headers with subpoints — whatever makes the output most useful to you.

Example: "Return your response as a JSON object with these keys: summary, pros, cons, recommendation."

5. Constraint Setting

Give the model boundaries. Word count, reading level, what to exclude, what tone to avoid. Constraints actually help — they narrow the solution space to something more useful.

6. Ask for Alternatives

Don't settle for the first version. Ask for three different versions, or ask the model to generate options from different angles.

Example: "Give me three alternative headlines for this article, each targeting a different angle."

7. Clarification First

For complex tasks, ask the model to identify any ambiguities before proceeding. This surfaces hidden assumptions before they become problems.

Example: "Before you write the email, list any questions or clarifications you'd need to write the best possible version."

8. Persona Specification

Define not just the expert role but the full context: who is the reader, what do they already know, what action do you want them to take? A complete persona specification dramatically improves targeted content.

9. Iterative Refinement

Treat each prompt as a starting point. After the first response, ask for specific improvements: "Make this 20% shorter and more direct," or "Add two more concrete examples."

10. Meta-Prompting

Ask the model to help you write a better prompt. Describe your goal and ask it to generate the ideal prompt for achieving it. Then use that prompt.


These techniques compound. The more you use them together, the better your results. Start with one or two, master them, then layer in the others.

Put these techniques to work

Alchemist helps you write better prompts in one click — right inside ChatGPT, Claude, and Gemini.

Get Started Free