Iterative Prompting

Why First Drafts Aren't Final

Expecting a single prompt to produce a perfect result is like expecting a first draft to be publication-ready. The model doesn't have access to the full context of your taste, your constraints, or the unstated criteria you'll only know when you see the output and think "that's not quite right." Iterative prompting treats the conversation as a collaborative drafting process, where each round narrows the gap between what the model produces and what you actually want.

How to Critique Output Effectively

When the output isn't right, don't just regenerate — diagnose what's wrong. Is it the tone? The structure? A specific factual error? The vocabulary level? Being specific in your critique produces targeted improvements: "This is too formal — rewrite in a conversational, first-person tone" works much better than "make it better." You can also ask the model to reflect: "What assumptions did you make about the audience in this draft?"

When to Refine vs Start Fresh

Refinement works best when the output is directionally correct but needs polish, tone adjustment, or minor restructuring. Start fresh when the output misunderstands the task at a fundamental level — if the first response heads off in the wrong direction, no amount of refinement will fix the underlying wrong interpretation. In that case, rewrite the initial prompt with more specificity rather than trying to steer a mis-aimed response.

Building Complexity Iteratively

For complex deliverables, iteration isn't just about refinement — it's a deliberate construction strategy. Start with a rough outline, confirm it's the right structure, then expand each section, then add supporting detail, then polish the language. This keeps each individual prompt focused and prevents the model from having to juggle too many simultaneous constraints at once.