Iterative prompting treats model interaction as a conversation: send a prompt, inspect the result, revise the request, and repeat until the output better fits the goal. It is especially useful when a first answer is confidently wrong or incomplete. section 9.4.5
Improving a multiplication answer
For 234 × 432, the bare prompt produces 98,136, and “What is 234 multiplied by 432?” produces 96,072. A stronger revision asks for three calculation methods, selection of the most accurate method, common long-multiplication steps, and shown work. The prompts become progressively more specific instead of treating the first response as reliable. section 9.4.5 section 9.4.5
A practical iteration loop is:
- Start with the simplest prompt that expresses the task. section 9.4.5
- Inspect the response for accuracy and identify what additional guidance it needs. section 9.4.5
- Add explicit instructions, alternative methods, and requested work when these may improve reliability. section 9.4.5
