Advanced Prompt Engineering Patterns for Repeatable Output
Once you know the basics, the next gains come from patterns: chain of thought, self-critique, and structured output. Here is how I use each. Basic prompt engineering gets you from vague to specific. Advanced patterns get you from specific to repeatable. The goal at this level is not a clever prompt, it is a prompt structure that produces consistent output across many runs and across different models. Here are the patterns I use most, with examples of when each one earns its complexity and when it does not. Pattern One: Explicit Chain of Thought For tasks that require reasoning, I ask the model to work through the steps before giving the answer. This is not the same as letting the model think out loud. I structure it. The prompt says: first, list the assumptions you are making. Second, identify the steps needed. Third, work through each step. Fourth, state the answer. Finally, note any assumptions that turned out to be wrong. The structure matters because it forces the model to expose its reasoning where I can check it. When the answer is wrong, I can usually find the step where the reasoning went off. When the model just gives the answer, I have no way to debug. Here is a simplified example of the structure: Task: Estimate the monthly cost of running a small API. Step 1: List assumptions (traffic, response size, region).