Hey,
I once spent an afternoon building a clean landing page. Good layout. Good copy. Took me two sessions with Lovable to get it right.
Then I typed six words into the chat.
"Make the homepage look more professional."
The AI took me seriously. It changed everything. Different fonts, different spacing, a new color scheme, the hero section completely rearranged. Even the copy I had written got rewritten.
I sat there staring at the screen. I had not asked for any of that. I had just wanted the buttons to look a little sharper.
The AI was not wrong. It just had no idea what I actually meant. So it guessed. And it guessed big.
The Real Problem With Most Prompts
Here is the thing nobody tells you when you start using AI tools.
AI has no memory of what you wanted it to be. It does not know your taste. It does not know what you spent time on. It does not care about what is already working.
It only knows what you type.
So when you type "make it look more professional," you are handing it a blank cheque. You are saying: do whatever you think professional means. And AI has a very broad definition of "whatever."
Most people treat prompts like Google searches. Short, vague, one line. They expect the AI to fill in the gaps.
AI does not fill gaps. It invents. And the vaguer your prompt, the more it invents.
The 3-Part Prompt
After the homepage disaster, I started treating every prompt like a small brief. Not a big document. Just three things, always in this order.
1. State
What exists right now. What has already been built. What is working.
This is the most important part and the one people skip most often. AI has no sense of what you have done before. You have to tell it.
2. Change
Exactly what you want different. One thing. Just one.
Not five things. Not "clean it up and also fix the nav and make the mobile view work and change the font." One thing per prompt. This is the same principle as the ROOM Method from Issue 01. Build one room at a time. Prompt one change at a time.
3. Keep
What must not change.
This is the one that saves you. Every time I skip this part, I lose something I wanted to keep. When you tell AI what to leave alone, it stops guessing. It knows the boundaries. It works inside them.
What This Looks Like in Practice
Here is the bad version of the prompt I typed that afternoon.
"Make the homepage look more professional."
Here is what I should have typed.
"The homepage currently has a white background, blue buttons, and a two-column layout with a headline on the left and an image on the right. I want the buttons to look more solid. Make the button corners sharp instead of rounded and increase the font size inside them to 16px. Do not change anything else on the page."
Same goal. Completely different output.
The second prompt has all three parts. State tells AI what exists. Change tells it exactly what to do. Keep tells it what to leave alone.
The AI has no room to guess now. It just executes.
The Same Idea, Different Business
Say you run a home tiffin service. You built a simple website with Lovable. It has a menu page listing your three meal options. A new customer asked if you do breakfast, so you want to add it to the menu.
You open the chat and type:
"Add a breakfast option."
The AI adds it. But it also decides to redesign the whole menu section. New layout, new fonts, the three existing items now look different. You did not ask for any of that.
Here is the same request with all three parts.
"The menu page currently lists three lunch options in a simple vertical list, with the name, description, and price for each item. I want to add a fourth item at the top of the list called Morning Box, with the description 'Poha, chai, and a seasonal fruit' and the price Rs 80. Do not change the layout, fonts, or style of the existing items."
The AI adds exactly one new item. Nothing else changes.
State told it what the menu looks like. Change told it exactly what to add. Keep told it what to leave alone.
The output you get is the output you asked for.
The Shorter Version
You do not need to write a paragraph every time. Once you understand the three parts, you can write fast.
Even something like this works.
"Right now the nav has five links. I want to remove the About link. Keep everything else exactly as it is."
State. Change. Keep. Three sentences. No disaster.
The habit is more important than the length. Train yourself to always include all three parts, even in a short prompt. Especially in a short prompt.
One More Thing
Here is a pattern I see a lot with people who are new to vibe coding.
They write a long prompt at the start when they are setting up the project. Good context, good detail. Then as the build goes on, the prompts get shorter and shorter. By the end they are typing three-word instructions and wondering why the AI is going off track.
The AI's memory does not improve as your session goes on. If anything, it gets more confused as the conversation gets longer. Your prompts need to stay clear throughout the build, not just at the start.
Every prompt is a fresh instruction. Treat it that way.
Your Action for This Week
Go back to the last thing you built or tried to build.
Find a prompt you used that gave you a bad result. A result where AI changed something you did not ask it to change, or missed the point entirely.
Rewrite that prompt using the three parts. State, Change, Keep.
You do not even need to use it for a build right now. Just rewrite it. The exercise will change how you think before you type.
Reply and send me the before and after. I read every reply.
If you want a collection of ready-made prompts you can adapt for your own builds, the VibeShip Vault has a growing prompt library. And if someone you know is getting garbage outputs from AI tools, send them here to join the community free.
See you next week.
Prathamesh
Founder, VibeShip