Run the model as an iterative dialogue, using follow-up moves like 'critique your own answer', 'redo it shorter', and 'give me three options' instead of accepting the first reply.
Prompting is a conversation, not a vending machine
Prompting is a conversation, not a vending machine
Most people treat an LLM like a search engine: type a question, read the answer, done. This one-shot mindset leaves most of the model's value on the table. A better mental model is a dialogue with a knowledgeable but fallible collaborator — you guide it through multiple turns until the output is actually good.
The first reply is a draft, not a deliverable. It tells you what the model understood, where it drifted, and what it missed. Your next move is to reshape the conversation based on that signal. Common follow-up moves include: «critique your own answer» (makes the model spot its own weaknesses), «redo it shorter / longer» (controls length), «change the tone to be more formal», «give me three alternative versions», and «what information do you need from me to improve this?».
For example, a first prompt — «Write an email to my team about the project delay» — might produce a vague, overly apologetic draft. A follow-up — «Now critique that draft: what is weak about it, and then rewrite it to be direct and action-oriented» — will almost always produce a sharper result in one extra turn.
Iteration is not a sign that you phrased the first prompt badly. It is the normal, professional way to use these tools. Experienced users rarely expect perfection on turn one.
Lesson notes
Prompting is a conversation, not a vending machine
Most people treat an LLM like a search engine: type a question, read the answer, done. This one-shot mindset leaves most of the model's value on the table. A better mental model is a dialogue with a knowledgeable but fallible collaborator — you guide it through multiple turns until the output is actually good.
The first reply is a draft, not a deliverable. It tells you what the model understood, where it drifted, and what it missed. Your next move is to reshape the conversation based on that signal. Common follow-up moves include: «critique your own answer» (makes the model spot its own weaknesses), «redo it shorter / longer» (controls length), «change the tone to be more formal», «give me three alternative versions», and «what information do you need from me to improve this?».
For example, a first prompt — «Write an email to my team about the project delay» — might produce a vague, overly apologetic draft. A follow-up — «Now critique that draft: what is weak about it, and then rewrite it to be direct and action-oriented» — will almost always produce a sharper result in one extra turn.
Iteration is not a sign that you phrased the first prompt badly. It is the normal, professional way to use these tools. Experienced users rarely expect perfection on turn one.