A slightly wrong response often triggers a chain of follow-ups: make it shorter, keep the original tone, add the missing context, change the conclusion. Soon you are eight messages into a task that should have been one.
That is not an efficient AI workflow. It is an unreproducible conversation with instructions scattered like breadcrumbs.
Why follow-ups compound badly
Each new instruction sits on top of the previous ones. Contradictions accumulate, and the final result cannot reliably be recreated or handed to a colleague.
When a response is wrong because the original brief was incomplete, go back and improve the prompt. Add the missing constraint, clarify the ambiguity, remove the instruction that caused the issue, or include an example. Then run the improved prompt as a fresh conversation.
What improving a prompt means
It is not adding more words. It is replacing vague instructions with precise ones: audience context, a maximum word count, an explicit format, a worked example, or a definition of “professional” that names the actual constraints.
Follow-ups are appropriate when you are refining one dimension of an otherwise correct output. The signal to fix the prompt is that you are adding a constraint the original brief should have included.
A good prompt is not written once. It is revised. Every edit makes the next first output closer to what you need.
This is part of the Parenting Your AI series, a practitioner's guide to building AI skills that are safe, effective, and worth trusting. Written from inside enterprise AI systems by someone who has spent years diagnosing what goes wrong when AI meets real work at scale.
Read the full series at KnowledgeManagement.ie