Have you ever been frustrated with the output of LLMs? If so, I have a tip that could be a breakthrough for you. It was for me.
I was prototyping a “Gallery View” dashboard that would visualize textual and numerical data in interactive, expandable components. I wanted to get to a clickable prototype with real data, using AI tools alone and not burn Eng hours.
But I kept getting stuck:
🚧 Lovable doesn’t accept CSVs or any attachment other than image files
🚧 Claude rate-limited me multiple times, and prompted me to “Start a new conversation” (not helpful b/c my initial prompt was long and with 5 attachments)
🚧 ChatGPT-5 did output a prototype file, but not in a format I could easily open and use (.tsx)
Until I tried this hack on prompting GPT-5
If you get responses that don’t work for you from ChatGPT, Claude, or any other LLM, type this prompt into your chat thread next:
📝 On a scale from 1-10, how satisfied do you think I am with your response(s) in this chat?
It will give a score and likely suggest ways to improve. If you need a big improvement, tell the AI that your score is a few points lower than the score it rated itself, and briefly explain what you would need to see happen to be satisfied. For even better results, explain your persona (eg business or technical) and what you are expecting to see.
For me, GPT5 was able to deliver a clickable prototype within its Canvas feature, on the spot. I could interact with it, explore the real data and UI functionality, and make key product and design decisions in a fraction of the time, at a fraction of the cost of engineers manually coding. I did need to prompt and push it a few more times, but if you have the patience and will, you can usually get these LLMs where you want them.
Hope this tip helps you too. If you try it, please let us know how it goes!
