I Used GPT Wrong for a Year. These 5 Prompt Principles Changed Everything
The secret to getting better and higher-quality responses
I used ChatGPT wrong for over a year.
My mistake? Feeling like it was a “smarter Google”.
I asked questions. It gave answers. And while it was often helpful, something always felt … off. The results were generic, inconsistent, and sometimes just completely wrong. Other times, I found myself “arguing” with it because it fell into a loop of hallucinations.
However, the problem was not the tool but how I used it.
Especially, how I instructed the AI (with prompts) to give me an answer.
In a book called Prompt Engineering for Generative AI: Future-Proof Inputs for Reliable AI Outputs, James Phoenix and Mike Taylor proposed 5 principles for a solid prompt.
These changed the way I work with ChatGPT.
☕️❤️ Support my work to continue writing by buying me a coffee.
Principle 1. Give Direction
Stop thinking about ChatGPT as a mind reader.
If you want responses that sound a certain way, informal vs formal, for example, you have to say so. Here, you must describe the desired style in detail. The more intentional you are, the better responses you will get.
Another good tip is to reference a persona.
Let’s see an example:
Bad prompt:
“Write a blog post about the neuroscience of consciousness.”
Better prompt:
“Write a blog post about the neuroscience of consciousness in the style of Anil Seth: concise, insightful, and story-driven.”
Principle 2. Specify Format
Without structure, AI will guess.
I know. Sometimes this is quite good. But again, GPT is not a mind-reader. You may want the results in plain text and not in bullets, but without telling GPT this, it will only guess.
So, define what rules to follow and the structure of the response you want.
Do you want:
Bullet points?
A table?
A JSON object?
A tweet thread?
Say so!
Bad prompt:
“Summarize this.”
Better prompt:
“Summarize this into 3 bullet points, each no more than 15 words.”
Principle 3. Provide Examples
This is where things get interesting.
And more personalizable. ChatGPT can learn what a “good” answer looks like. So, the idea is to insert a diverse set of test cases where the task was done correctly.
Let’s see more examples:
Bad prompt:
“Write 30 headlines for a Medium article.”
Better prompt:
“Write 30 headlines for a Medium article using these examples I like as a template: [insert headlines]”
Principle 4. Evaluate Quality
Although tedious, this could enhance GPT’s responses.
Here, the idea is to identify errors and rate responses, testing what drives a better performance. In other words, compare multiple outputs. Spot the problem. Give feedback.
Treat GPT as an intern.
Principle 5. Divide Labor
Some tasks are too complex for a single prompt.
So, the solution is to use the Lego mindset: break down the task into smaller and actionable chunks. This is crucial for a solid prompt. Here are some guidelines for splitting tasks into multiple steps, chained for complex goals:
Step 1: Outline the task.
Step 2: Gather input.
Step 3: Generate a draft.
Step 4: Review and revise.
This will not only give you a better response but also enhance your logical and critical thinking.
Here’s one prompt I used some time ago:
Bad prompt:
“Should I switch from Notion to Obsidian?”
Better prompt:
“Step 1: List the pros and cons of Notion vs. Obsidian for note-taking. Step 2: Create a decision matrix using cost, usability, and offline access. Step 3: Recommend the better option for a self-taught learner.”
All Together
So, this is how to prompt smarter.
Once you apply these principles, I promise you’ll notice the difference immediately. To recap, these are the five principles:
Give direction.
Specify format.
Provide examples.
Evaluate quality.
Divide labor.
Ultimately, this is about using AI as a tool rather than a simple generator. Plus, GPT is not a mind-reader, and it needs clear step-by-step guidelines. Otherwise, it will guess or give bad responses.
These five principles tackle this.
I hope you find them useful too!
Ultra-learning saved my career.
It helped me learn anything without overstudying or overworking. But remember, each person has a different life. Perhaps you work full-time and also study. Or maybe you have a family, etc.
Be reasonable, and don’t fall into toxic productivity.
Here are some last recommendations:
Don’t compare yourself with others. Take your time.
This is not a competition. This is about you. Make it fun and healthy.
Please take care of yourself. That’s the most important thing!
Now it is your turn.
What are you going to learn next?
☕️❤️ Support my work to continue writing by buying me a coffee.
