Lessons0 of 20 done

Module 3: Using AI well. Lesson 11 of 20, 8 minutes.

The conversation is the tool.

How to turn a so-so first reply into a good one: options, self-critique, edited prompts, fresh chats and a library of what worked.

You ask for a short note to your landlord about a leaking tap. The reply arrives in two seconds. It is fine. It is also stiff, twice as long as it needs to be, and it apologizes for something that is not your fault.

Most people do one of two things here. They paste it anyway, or they sigh and write the note themselves. There is a third option, and it is where almost all the value in these tools lives: keep talking.

The first answer is a draft

Think of the first reply the way an editor thinks of a first draft: raw material, not a verdict. The model made dozens of small decisions on your behalf (how formal, how long, which point goes first, whether to add a sign-off), and it made them without knowing your taste. Some will be wrong. That is not a failure of the tool. It is the normal starting point.

What changes when you treat the reply as a draft is that you now have something concrete to react to. Reacting to a real sentence (too formal, cut the second paragraph) is far easier than describing a perfect sentence from nothing. The conversation is where the good answer gets made.

Ask for three and pick one

If you are not sure what you want, do not make the model guess once. Make it guess three times, in three different directions, and choose.

Give me three versions of the opening line: one plain and factual, one warm, one short enough for a text message.

Picking is faster than describing, and seeing the options side by side often tells you what you wanted all along. You can also mix: Use the tone of the second one and the length of the third. Three is a good number. Two feels like a coin toss and six is a chore to read.

Make it critique its own draft

Models are better at finding problems in text than at avoiding them on the first pass, so use that. The key is to give it a goal to measure against. Make it better gives it nothing to check. This does:

The goal: a parent skimming this on a phone should know the date, the place and what to bring within ten seconds. Critique your draft against that goal. List the three biggest problems, then rewrite it.

If the critique comes back vague (could be more engaging), push once: Be specific. Quote the sentence and say what is wrong with it. You will usually get a sharper rewrite than any amount of general encouragement produces.

Fix the prompt, not the reply

Here is a habit that quietly ruins long chats. The reply is off, so you type shorter. Still off: no, keep the greeting. Then actually, make it formal. Each correction fixes one thing and leaves the thread longer.

Remember that the model re-reads the whole conversation on every turn. By the fourth correction it is reading your original prompt, three rejected drafts and three contradictory instructions, and trying to honor all of them. Old instructions linger. Things you rejected creep back in.

The cleaner move: most chat apps let you edit an earlier message. Go back to your original prompt, add what you have learned, and resend. The rejected drafts drop out of view and the model starts from a clear brief.

Before

Write a note to my landlord about the leaking tap. (Then: shorter. Then: no, keep the greeting. Then: make it formal.)

After

Write a formal note to my landlord about the leaking tap in the kitchen. Under 80 words. Open with a polite greeting. Ask for a repair date this week. Do not apologize.

One quick follow-up is fine. If you are on your third correction, stop and edit the original instead.

When to start a fresh chat

Sometimes editing is not enough and the thread itself has become the problem. Signs it is time to start over:

  • You have changed topic. The old material is now noise the model keeps reading.
  • The model keeps reverting to something you rejected three replies ago.
  • Replies are getting slower, longer or oddly repetitive.
  • You can no longer remember what it knows and what it has lost track of.

Starting fresh throws away the context, so carry over what matters. Paste the current draft, or write three lines: Here is where we got to. The goal is X. The draft so far is below. The one thing still wrong is Y. A chat is a workspace for one job. When the job changes, open a new one.

Let it interview you, and keep what works

For anything with more than one moving part, flip the order. Instead of writing a long brief and hoping you covered everything, ask the model to find the gaps first:

Before you write anything, ask me the five questions you need answered to do this well. Then wait.

It will surface decisions you had not thought about: who else reads this, what happens after they read it, when it is due. Answer all five in one message and let it go. This takes a minute and saves three rounds of corrections.

Finally, when a prompt works, save it. A plain notes file is enough: the prompt, what it was for, and one line on what made it work (naming the reader fixed the tone). After a few weeks you have a set of starting points you trust, and each new task begins from the best version of the last one rather than from a blank box. A prompt helper such as Alide can sharpen a prompt as you type; the record of what actually worked for you is something only you can keep.

Try it

Open any chatbot and type: Write a two-sentence message asking my neighbor to move their car from in front of my driveway. Read the reply. Now, instead of replying to it, find the edit control on your own message (usually a pencil icon that appears when you hover or tap) and change it to: Write a two-sentence message asking my neighbor to move their car from in front of my driveway. Friendly, not apologetic. Mention that it blocks me from leaving for work at 8. Resend and compare. Then try one more: Give me three versions: one warm, one neutral, one that is a single line. Notice how much faster it is to pick than to describe.

Key terms

Draft
A first version of a reply that you expect to change, rather than a finished answer.
Edit and resend
Changing an earlier message in the chat and sending it again, so the model starts from a clean brief instead of a pile of corrections.
Self-critique
Asking the model to judge its own reply against a stated goal before rewriting it.
Fresh chat
A new conversation with no history, used when the old thread has become noise.
Prompt library
Your own saved collection of prompts that worked, with a note on why each one did.

Check yourself

1.You have sent three corrections in a row and the model keeps bringing back something you rejected. What is the best next move?
2.Why ask for three versions instead of one?
3.What makes a self-critique prompt useful?

The first reply is raw material; the good answer comes from what you do next, and the cleanest move is usually to edit the prompt rather than pile on corrections.