Prompt Iteration

This skill turns writing instructions for an AI from guessing into a reliable process. When you finish this section, you will be able to take any request — a product description, a customer-support script, or a coding task — and improve it systematically. You will learn to change one specific detail, test the result, measure the outcome, and repeat, just as a developer fixes software bugs. You will move from saying 'this feels better' to saying 'this version works because...' with proof. This discipline is essential for two practical tools you will encounter later: using a program to automatically find the best instructions (automated search) and keeping a structured history of your drafts so you can undo mistakes without losing progress. Without this method, those tools have no foundation.

Think of the last time you asked someone for a favor and they misunderstood you. You likely did not just repeat the request louder. You reworded it: 'No, I meant the blue one, not the red one.' You changed one specific detail, checked their response, and adjusted again. This is iteration. Prompt iteration is this same disciplined process applied to an AI model.

Here is how it works in practice:
1. You start with a baseline prompt.
2. You test it on a collection of questions where you already know the correct answers (this collection is called an evaluation set, which is simply a test bank of known good examples).
3. You measure how many answers the AI got right.
4. You change exactly one thing in the prompt — one word, one example sentence, or one formatting rule.
5. You run that same evaluation set again and compare the new score to the old one.

The critical rule is changing only one variable at a time. If you change three things simultaneously and the performance improves, you do not know which of the three changes helped. If it gets worse, you do not know which one caused the damage. By isolating each change, you can attribute the result directly to that specific adjustment. This is the scientific method applied to writing prompts. Without this measurement step, you are guessing. With it, you build reliable results.

Imagine you are a chef creating a new dish. You have a base recipe and want it to be perfect. You do not throw random ingredients in and hope for the best. You start with a standard version, cook it, taste it, and score it out of ten — let’s say it gets a 6.

Then you change one thing: you add a pinch of salt. You cook it again, taste it, and score it. Now it is a 7. You then change another single thing: you reduce the cooking time by two minutes. Now it scores an 8.

If you had added salt and reduced the time at the same time, you would not know which adjustment improved the dish — perhaps one helped while the other hurt, canceling each other out. The same logic applies to prompts: you have a test bank of inputs, run your prompt against it, score it, change one word, and score it again.

Note that this analogy is imperfect because a recipe is fixed steps, whereas an AI interprets instructions differently depending on its version or context, which is why keeping the test bank constant is vital.

Prompt iteration is the practice of improving an instruction set by altering exactly one element at a time and measuring the result against a fixed collection of known examples.

Formally, it is a cycle of modification and evaluation designed to isolate the cause-and-effect relationship between specific prompt changes and output quality.

This concept supports Prompt Optimization, which is the automated process where a computer program performs this cycle of changing and testing instructions for you, and Prompt Registry and Versioning, which is the system of saving each version of your prompts so you can track progress or revert to a previous working state if a change fails. It also underpins the later topic of RAG (Retrieval-Augmented Generation), where the information retrieved to support the prompt changes, making this iterative testing even more critical.