Four weeks in. Here's what we've covered:
- Issue 01: Prompt Engineering — how to give AI clear instructions
- Issue 02: Taste Curation — how to know when the output is actually good
- Issue 03: The Master Prompt — how to introduce yourself to AI so every session starts with full context
If you missed the previous issues, click on any of the links.
This week: Output Iteration.
This is the skill that separates people who use AI well from people who use AI brilliantly.
And i can tell you, am still figuring my taste… i just believe it takes a while to get to nearly perfect. But for today, you just build the knowledge and then try.
The Mistake Most People Make
Most people get their first AI output, think "that's okay, I guess" and either use it as-is or give up and do it themselves.
Both are wrong.
The first output is never the final output. It's the starting point. It's clay. Raw material.
The people who get extraordinary results from AI treat every first output as a draft something to react to, push against, and pull into shape. They don't accept it. They iterate.
Coca-Cola used 70,000 AI prompts made by just five experts to produce one Christmas commercial. Not 70,000 people. Five. And 70,000 rounds of iteration.
That's what commitment to output looks like.
How Iteration Actually Works
Here's the mindset shift: stop thinking of AI as a vending machine (put in prompt, get out finished product) and start thinking of it as a collaborator you're directing.
You give a direction. It responds. You react specifically. It adjusts. You keep going until it's right.
The key word is specifically.
"Make it better" is not useful feedback. AI doesn't know what better means to you.
"Make it punchier" is better, but still vague.
"Rewrite the opening line so it leads with the specific outcome save 5 hours a week rather than the process" is the kind of feedback that gets you somewhere.
The more specific your feedback, the faster you get to the output you want.
The Iteration Process, Step by Step
Step 1: Upload your master prompt first
Before you do anything else, paste your master prompt in ChatGPT or Claude. This primes the AI with full context and dramatically improves the first draft. (In the future we shall cover how to work with these AIs more efficiently using Projects, skills or custom GPTs.
Step 2: Get the first output and react to it
Read it. Mark what's working. Mark what isn't. Be specific about both.
Step 3: Give targeted feedback
Don't rewrite the whole prompt from scratch. React to what you got:
- "The structure is right but the opening is too slow. Start with the result, not the setup."
- "The tone is too formal. Write like you're explaining this to a smart friend, not a boardroom."
- "Cut the third paragraph it's a repeat of the second. And shorten the CTA to one sentence."
Step 4: Lock in what's working, adjust what isn't
If the structure is right but the language is off, say so. Don't ask AI to redo everything tell it to keep what's working and fix what isn't.
Step 5: Iterate until you'd be happy to put your name on it
That's the standard. Not "good enough." Would I be happy to send this to a client? To post this publicly?
Using the Canvas Feature (ChatGPT)
If you're on ChatGPT, there's a feature called Canvas that makes iteration much easier.
Canvas turns your AI output into a live, editable document like a Google Doc inside your chat. Instead of re-prompting the whole response every time you want a small change, you can:
- Click directly into the document and edit words and sentences manually
- Type a specific instruction in the chat and it updates just that section leaving the rest intact
- Lock in structure and format, then fine-tune the language
How to use it: after you get an output you like the shape of, say "Rewrite this as a Canvas." The side panel opens. From there, you're editing, not re-generating.
Pro tip: Use regular chat to do the heavy structural work first get the right bones, right flow, right sections. Then switch to Canvas to polish the final 10%. That's the most efficient order.
3 Things Worth Your Attention
1. The feedback quality rule
If you can't describe specifically what's wrong, you're not ready to give feedback yet. Read it again. Get specific( numbers, results, percentages, location etc. Then react.
2. Save your best outputs as templates
When you nail an output after iteration, save it. The structure, tone, and format you landed on is now a template for every future piece of similar work.
3. Next week: System Prompts
Once you've iterated your way to something great, you can reverse-engineer it into a system prompt a reusable instruction set that recreates that quality on demand. That's next week.
I would love to hear about your AI journey so far, or you can just simply tell me what you want covered.
If you loved this issue, share it with someone who might benefit.
Still learning, still sharing.
Harriet
Founder SaviteckX