At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Hi, and welcome back to the Prompt Basics session. We can already build prompts and give examples. But what if the result still isn't right? In this episode we learn iteration: treating the first answer as a draft, giving specific feedback, choosing the right way to revise, and knowing when to stop.
Step 1: The first answer is a draft
Many people are disappointed when Claude's first answer isn't quite right, and they give up. Yet even professional writers are rarely satisfied with their first draft. Change how you see it: the first answer is a draft. The best results usually come after two or three rounds of revision, and each round takes only a few seconds. So from now on, don't judge Claude by its first answer. Judge it by the result after you've guided it, just like coaching a new employee. Iteration follows a simple cycle. First, ask with a structured prompt, as we've already learned. Second, read the result carefully. Mark the parts that already fit and the parts that don't yet. Third, give specific feedback. This is the part that matters most, and we'll go deeper into it. Fourth, repeat the cycle until the result meets your criteria. The advantage of iterating in the same chat: Claude remembers all your earlier requests. So with every round, it understands your taste and needs better.
Step 2: Specific feedback
Let's compare. Make it better sounds reasonable, but Claude has to guess what you consider better. The specific version: the opening line is weak, replace it with a question. Too long is vague too. How long is right? Better to write: shorten it to three sentences max. Not casual enough can mean many things. Clearer: say you instead of sir or madam, and add one emoji. Something's missing is a riddle. Just name what's missing: add the opening hours at the end. Notice the pattern. Specific feedback always names which part and what to do. Claude doesn't have to guess, so one revision is enough. To make it easy, use the four-part feedback formula. First, say what's already good, for example the casual tone is just right. This tells Claude what to keep. Second, say what needs changing by pointing to the part: the second paragraph is too long. Third, give a clear direction of change: turn it into three short points. Fourth, set limits: don't change the title or the prices. Without limits, Claude sometimes changes parts that were already correct. Not every piece of feedback needs all four parts. But for important revisions, this formula saves many rounds.
Step 3: Iteration example: menu brochure copy
Let's see iteration in practice. Laras asks for Sunset Coffee menu brochure copy to hand out on campus, complete with three product prices. Round one: the result is too formal and long. Lines like we proudly present don't suit students. Laras doesn't start over. She uses the four-part formula: the warm tone is good, but it's too formal, make it casual and friendly, four lines max, and don't change the prices. Round two fits much better. The headline is catchy, the language is casual, and there's a student touch: stocking up for study night. The prices stay the same. Round three is just one detail: adding how to order and the free shipping info. Done. The brochure copy is ready to print. Laras just needs to check the prices once more before sending it to the printer. Three rounds, under five minutes. Notice that the feedback gets smaller each time. That's a sign the iteration is heading in the right direction. Imagine if Laras had only written make it better in round two. Claude might have added fancier words, when what she needed was shorter and more casual.
Step 4: Choosing the right way to revise
Besides typing feedback, there are four ways to revise. First, a follow-up message, like Laras did. It's good for gradual adjustments. Second, edit your message. Use it when the original request turns out to be off track. For example, Mr. Hendra realizes his summary is for students' parents, not the students. It's better to edit his first message than to pile up corrections. Third, retry. Use it when the request is right, but you want another version, for example to find alternative headlines. Fourth, a new chat. When a chat has become very long and the instructions contradict each other, ask Claude to summarize the latest decisions, then start a new chat with that summary. Signs a chat needs a fresh start: Claude begins forgetting the original instructions, or repeats mistakes that were already fixed. Iteration also works for results shown as an Artifact. Dewi asks for a notice to customers: in July there will be no Saturday deliveries. Claude creates the draft in the right panel as an Artifact. Dewi gives specific feedback: add a table of deliveries per day, say Saturday orders ship on Monday, and don't change the title. Claude updates the notice. Look at the version label in the right panel: it's now version two. The previous version can still be opened and compared. This version feature lets you experiment with confidence. If a revision makes things worse, just go back to the previous version. We'll cover it fully in the Artifacts session. Dewi also set a limit: don't change the title. As a result, the title stays the same and only the requested parts change. The four-part formula works just as well with Artifacts.
Step 5: Ask Claude to critique, then know when to stop
There's one technique that saves many rounds: ask Claude to critique its own draft. The task part: review your latest draft, name its weaknesses honestly, then write an improved version. The limits part holds your criteria, for example four lines max, student-friendly language, and every price included. Also name the parts that must not change. The format is a short table with the criterion, a score from one to five, and the reason, followed by the improved version below the table. For example, Laras uses this prompt on the brochure copy. Claude gives a score of four for length, then notices on its own that the opening hours are missing and adds them. With this technique, Claude finds weaknesses you might have missed. Save it as Go-To Prompt number three. You still decide whether to accept the improvements. Finally, when should you stop? First, stop when all your original criteria are met. That's why you set the criteria at the start. Second, make sure facts, prices, and names have been checked. That's your responsibility, not Claude's. Third, if the only changes left are small matters of taste, like one word, it's faster to edit it yourself after copying. Watch out for two traps. First, five rounds in, and the result keeps drifting further off. That's a sign your original prompt needs fixing, or it's time to open a new chat. Second, chasing a perfect result when it's already good enough. Your time is worth more than tiny differences readers won't even notice. Remember, the goal of iteration is a useful result, not a perfect one. Two or three rounds are usually enough for most daily work.
Done!
We've learned to treat the first answer as a draft, give feedback with the four-part formula, choose the right way to revise, ask Claude to critique its own draft, and know when to stop. In the next episode we learn to ask for tidy output: tables, lists, and summaries ready to paste into a document or spreadsheet. See you there!
Practice exercises
Try these on your own after watching to lock in what you learned.
Revise one piece of writing for up to three rounds with specific feedback
Ask Claude to critique its own draft using your three criteria
Save the self-critique prompt as Go-To Prompt number 3
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,286 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
