Stuck on AI Writing Topics? Here’s How to Choose. The 3-Step “Purpose-Based” Selection Method
You’ve likely had a similar experience when cooking. If you just say “how to cook,” it’s too vague. Even “cooking with pasta noodles” is too broad a scope.
However, the moment the goal becomes clear, like “how to make Napolitan spaghetti,” your hands start moving immediately.
AI is exactly the same. It looks like it can create anything, but that actually makes it easy to get stuck, unable to decide what to do. This isn’t because you lack ideas, but because the method of choosing a topic itself is inefficient.
Most people start like this:
- What should I do with ChatGPT?
- What can I make with Gemini?
- Where should I use NanoBanana?
This kind of “Tool → Topic” approach fails almost 100% of the time. Because there are too many things you can do with the tool, making a choice becomes impossible. In reality, completely flipping the order is much more stable. The most realistic way to choose a topic is the structure of:
“Work Purpose → Topic”.
Below is the 3-Step Automatic Topic Selection Method that simplifies this approach.
Step 1: Define the “Output” of this task in one line
This is a method of deciding the result (output) first, not the topic. Most creations using AI fall into one of the following categories:
- 1 Character Design
- 1 Illustration
- 5 Lookbook Cuts
- 3 Profile Images
- 1 Blog Post
- 1 Prompt Set
- A few Design Elements (UI Buttons)
- Report Outline
- Short Description Text
If the output is clear, more than half of the topic is automatically decided.
For example,
- I will make 5 lookbook cuts.
- I will create 1 character.
- I will finish 1 article.
- I will create 1 color prompt set.
Setting this up makes the start very fast. Conversely, if the output is unclear, it is difficult to start working no matter which AI model you use.
Step 2: Pick just one “Pain Point” you are facing now
All topics ultimately start from a problem. If you leave this vague, the topic becomes ambiguous. Just pick one problem you are stuck on right now.
Examples:
- The character’s face looks different every time.
- The lookbook colors are not unified.
- The prompts are always getting too long.
- Drafting text is too slow.
- GPT keeps going off-topic.
- The style is not maintained.
- Information is scattered, so I don’t know what to do first.
The moment you write one of these down, the topic transforms into a testable form.
For instance,
- Solving character consistency issues
- Lookbook color unification test
- Improving drafting speed
- Preventing ChatGPT topic deviation
A topic is effectively the name of the problem you want to solve right now.
Step 3: Decide on one “Test Element” for this task
Work utilizing AI isn’t about completing something in one go; the experiment of changing small variables itself becomes the topic. It is enough to choose just one element to test from the list below.
- Prompt Structure
- Model Comparison (GPT vs. Claude, Seedream vs. NanoBanana)
- Parameter (Settings) Changes
- Style Templates
- Color Presets
- Number of Samples
- Step-by-step Generation Method
- Rewriting Method Comparison
It becomes clearer with examples.
- GPT’s prompt structure A/B testing
- Experiment on maintaining identical styles with the NanoBanana series
- Seedream vs. NanoBanana lookbook comparison
- Testing GPT’s 3-step outline structure
- Comparing “Descriptive vs. List-style Prompts” for fixing character faces
Once you reach this stage, you can consider the topic virtually decided.
Combining these 3 steps automatically completes “Topic Selection”
Looking at the combinations, you’ll get the idea immediately.
Example 1)
- Output: 1 Character
- Problem: Face style changes every time
- Test: Compare prompt structures
Final Topic
Testing character consistency maintenance with Prompt Structure A/B
Example 2)
- Output: 5 Lookbook cuts
- Problem: Colors pop out randomly every time
- Test: NanoBanana style preset test
Final Topic
Unifying lookbook colors using style presets with NanoBanana
Example 3)
- Output: 1 Blog post
- Problem: Drafting speed is slow
- Test: GPT Outline 3-step routine
Final Topic
Creating a draft in 10 minutes with the GPT Outline 3-Step Routine
As you can see, a topic isn’t something you squeeze out of your brain; when you organize the Output, Problem, and Test Element, it naturally falls into a sentence.
Doing this eliminates the reason for getting stuck on topic selection
The reason a topic isn’t decided isn’t because you don’t have ideas,
- It’s because you try to choose from vague possibilities starting from the tool without criteria.
On the contrary,
- Output
- Problem
- Test Element
If you choose just these three, the topic changes into a unit of work with intention and purpose, and you can start working immediately.
NR Layer’s Perspective
These 3 steps look like a simple checklist on the surface, but in reality, it can be seen as shifting from a “method of creating topics by thinking” to a “method of extracting topics from the workflow.”
As AI tools become more powerful, “what you can do” increases endlessly, but what truly matters is the criteria for deciding “what to do right now.”
If you bundle Output-Problem-Test Elements together, the record itself becomes a reference for the next task, the experimental results become content, and it becomes a “self-manual” that you can use again later.
Conclusion
Next time you start a task, just writing down these 3 things is enough.
- What is the Output of this task?
- What exactly is the Pain Point you are experiencing now?
- If you had to pick just one Test Element for this time, what is it?
Once these three are decided, the topic creates itself. And this method applies exactly the same regardless of which AI tool you use—GPT, NanoBanana, Gemini, etc.
In the next post, I will organize a routine for extracting an outline and draft in 10 minutes using GPT-5.1 with the topic decided in this way.
