Improve Results Through Conversation

The first AI answer isn't an exam submission. It's something you can react to. Once you stop treating it as the finished product, using AI becomes a lot more interesting.

You're writing an invitation to a neighbourhood get-together. You tell the assistant the time, place and what it's for. It returns a beautifully polite message that sounds as though it came from an insurance company. You wouldn't send it. But you haven't wasted the conversation; you've discovered what doesn't sound like you.

Instead of starting again, respond naturally: 'Too formal. Make it sound like a message between neighbours and keep the practical details easy to spot.' The next version is closer. Maybe the opening works, but the ending sounds overenthusiastic. You can preserve the opening and ask for two alternative endings. You're developing the same piece of work with each reply.

This is *iteration*: producing a version, examining it and making a deliberate change. The quality of your feedback makes a difference. 'Better' leaves the assistant guessing. 'Shorten the opening, keep the location and remove the sales-like language' gives it a direction. You don't need technical language; you need to say what feels wrong and what you want instead.

Sometimes it helps to compare very different possibilities rather than polishing one draft forever. Ask for a warm conversational invitation and a much shorter direct version. You may prefer the clarity of one and the personality of the other. The assistant can bring those qualities together, helping you discover a preference that wasn't clear at the beginning.

Examples can also communicate taste. Show the assistant a short, non-sensitive message you've written and say what you like about it. You can ask it to compare your style with its first suggestion. That gives you something more tangible than vague adjectives like friendly, confident or professional.

The same process works beyond writing. A study plan can change when you realise the schedule is too ambitious. A creative idea can grow after you see a rough version. A list of possibilities can become a usable shortlist once you explain your priorities. In each case you're reacting to a concrete output rather than trying to write a perfect opening instruction.

Suppose the second invitation is better but still feels a little unnatural. Instead of using another vague adjective, explain the difference you notice: 'It sounds as though I'm advertising an event. I just want to see whether my neighbours fancy coming over.' That small observation gives the assistant a much richer picture of the relationship and purpose. Good feedback often starts with noticing how a message would feel to receive.

You can also ask the assistant to show its changes. Perhaps it removed several words, changed the opening and replaced a formal sign-off. Looking at those choices helps you develop your own writing judgement. You might realise that the problem wasn't sentence length at all. It was that the original draft made the occasion sound more official than it is.

Now imagine you have two promising versions. Don't immediately demand a third. Read each as though you were receiving it on your phone. Which would you understand more quickly? Which sounds more like somebody you know? Which makes it obvious whether you need to reply? You can tell AI what you like about both and ask for one combined draft, but the meaningful evaluation is yours.

Iteration can become unproductive if every version is merely different. You might ask for 'more friendly', then 'more professional', then 'more relaxed' until you're back where you started. A better habit is to name the specific weakness and keep the parts that work. You can even say, 'Don't change anything except the final sentence.' That reduces the chance of losing a good feature while trying to fix another.

The same idea helps when working on a visual design, a plan or a complicated explanation. After each version, identify what became clearer and what got lost. You're not trying to make AI obey a perfect instruction from the start. You're learning to use feedback as a tool for discovering and refining the result, then recognising when you've reached something good enough to use.

The point is not endless revision. Decide what good enough looks like, stop when the result serves its purpose and keep the final judgement for yourself. AI generates alternatives quickly. You supply the taste, purpose and decision to finish.