Academy
5. Iterate in Place — Don't Start Over
Use targeted feedback within the same conversation to refine results without losing context, and know when to restate the goal.
When the model gives you a result that's close but not right, your instinct may be to start a brand-new message from scratch. Resist that. The conversation itself is context: the model remembers your goal, your constraints, and every refinement you've already made. Starting over throws all of that away and forces you to rebuild from zero. Instead, stay in the same conversation and give targeted feedback that names exactly what to keep and what to change.
The key word is "targeted." Saying "make it better" gives the model no direction — "better" is not a dimension it can optimize. Instead, name the specific change: "shorten section 2 to three sentences," "replace the hypothetical example with a real customer case," "make the tone more formal for a board audience." Each instruction should point at one thing to fix while implicitly confirming that everything else is fine. The model can only improve what you explicitly name.
Use a simple pattern for feedback: (1) say what to keep, (2) say what to change, (3) say how. Example: "Keep the structure and the intro. Shorten section 2 by half. Replace the generic example in section 3 with a concrete SaaS onboarding scenario." This protects the good parts while steering the rest.
After several rounds of refinement, the conversation can drift — the model starts losing track of the original goal or mixes up earlier instructions. When you notice the output wandering, don't abandon the chat. Instead, restate your goal in one clear sentence: "Reminder: we are writing a 500-word internal FAQ for new hires about expense-report rules." This re-anchors the model without losing the work already done. Think of it as a mid-course correction, not a restart.
When should you actually start a new conversation? When the task itself changes — not just the output. If you were drafting a blog post and now need a slide deck on a different topic, a fresh chat is appropriate because the old context would confuse more than help. But if you're still refining the same deliverable, stay in place and iterate. The rule of thumb: same goal → same conversation; new goal → new conversation.
Open a ChatPro conversation. Ask the model to draft a short professional email (pick any topic). Then, without starting over, send three successive refinements: one about tone, one about length, and one about a specific detail. Notice how each reply preserves your earlier corrections automatically.
If a refined result is excellent, don't let it stay buried in a chat. Turn the prompt and instructions into a reusable skill or agent so you — and your team — can apply them automatically next time.
Iterating in place is faster and smarter than starting over. Stay in the same conversation to preserve context. Give targeted feedback that names the exact dimension to change — never just "make it better." When the chat drifts, restate your goal in one sentence to re-anchor the model. Only open a new conversation when the goal itself changes. And when you land on a great result, save the prompt as a reusable skill.