TL;DR - Treat AI like a conversation, not a vending machine. Get a rough draft, then refine with small, specific nudges. Iterating beats trying to write one perfect prompt.
Why it matters
The best AI users operate in a loop - ask, react, redirect - rather than expecting perfection in one shot. Each small correction steers the result, and you stay in control of the outcome.
Worked example - the refine loop
You: Draft a follow-up email to a lead who went quiet.
AI: [draft]
You: Warmer, and shorter - 4 sentences max.
AI: [tighter draft]
You: Add one specific reason to reconnect this week.
AI: [final - ready to send]Three small turns beat one giant prompt, and each step is easy to judge.
Steal this - refinement nudges
Length: "cut by half" · "one paragraph"
Tone: "warmer" · "more direct" · "less salesy"
Focus: "lead with the ask" · "drop the second point"
Form: "make it a bulleted list" · "add a subject line"Common mistakes (and the fix)
- Restarting from scratch when a draft misses. Fix: reply with the single change you want.
- Vague feedback ("make it better"). Fix: name the specific change.
- Losing good earlier versions. Fix: "keep the last version but only change X."
Good to know
This is why chat history matters - the model uses the whole conversation as context, so each turn builds on the last. In Claude Projects and ChatGPT threads, you can keep iterating on the same material for a long session. Start a fresh chat only when you want a clean slate.