Follow-Up Prompts and Iterating in Conversation

A useful conversation does not require a perfect first prompt. After reviewing the answer, ask for a targeted change: add a missing source, shorten one section, compare another option, or explain a point in simpler language. OpenAI’s current prompting guidance says to review the response and use follow-up messages to shape the result without starting over.

Specific follow-ups work better than “try again” because they identify what should change and what should stay. For example: “Keep the evidence and conclusion, but move the recommendation to the top and cut the background to three sentences.” If the answer made a factual error, provide the correction or a source and ask the assistant to revise the affected part. If the original goal changes, state the new goal explicitly.

Iteration is not automatic verification. The model may preserve an error, introduce a new one, or forget an earlier constraint as the conversation grows. For an important response, check claims and output requirements after revisions. When a thread becomes long or goals shift, summarize the current task, active constraints, decisions, and unresolved questions, or start a fresh conversation with that summary.

Use follow-ups as a way to collaborate and refine, while remaining responsible for the final result. Save versions for high-stakes work and verify citations, calculations, and actions independently. A better conversation can improve fit to the request, but it does not guarantee correctness or replace domain expertise.