banner
akihaye

akihaye

Universe!
follow
github
steam
discord user
email

AI Communication Guide: How to Ask Questions to AI Correctly?

Do you often encounter situations like this:

  • Not knowing how to ask questions to AI?
  • After much effort to come up with a question, still not getting the ideal answer?
  • Why do others seem to use AI effortlessly while you frequently hit roadblocks?
    Is your little head filled with big question marks? Don't worry! It's not your fault; it's just that you haven't mastered the correct "way to consume" conversations with AI yet 🍴🍜.

In this guide, we will explore together:

  1. Why is it important to learn how to ask questions to AI?
  2. What is a Prompt? What is the basic structure of a Prompt?
  3. Personal experiences in conversing with AI
    ……
    I hope this blog post can be helpful to you!

Friendly reminder: This article is crafted using all-natural, handmade methods. If any AI-generated content is used, it will be specifically noted in the text~


ChatGPT Image 2025 年 4 月 27 日 下午 10_48_00

Why is it important to learn how to ask questions to AI?#

In today's rapidly developing generative AI technology, mastering the art of questioning has become an essential skill.
In the future, those who can interact efficiently with AI will have a super assistant like Iron Man's Jarvis! 💥
Whether in work, study, or daily life,
efficient AI conversations can save you time, enhance productivity, and predict trends, truly giving you a personal assistant comparable to Iron Man's Jarvis!


ChatGPT Image 2025 年 4 月 27 日 下午 10_45_38

What is a Prompt?#

In simple terms, a Prompt is the instruction text we carbon-based beings input when communicating with silicon-based beings—AI (like ChatGPT, DeepSeek, etc.).
It directly determines the quality of AI's responses.

For example:

  • Weak Prompt: "Write an article."
  • Medium Prompt: "Write a popular science article introducing the basic principles of Yin-Yang and the Five Elements, aimed at university students majoring in Traditional Chinese Medicine, requiring a clear structure and lively, easy-to-understand language."
  • Strong Prompt: Detailed explanation follows!

ChatGPT Image 2025 年 4 月 27 日 下午 10_52_49

How to correctly write a qualified Prompt?#

A qualified Prompt depends on your type of needs👇:

  • Want inspiration?
    • It's recommended to use a short and free Lazy Prompt. Let AI unleash creativity like bamboo shoots after the rain!
    • Applicable scenarios: brainstorming, writing outlines, rewriting styles, etc.

    Example: "Help me come up with a few creative ideas on the theme of 'The Future of Humans and AI'."; "Help me polish this article: insert content."

  • Is the task rigorous and complex?
    • Then obediently use a structured Prompt.
    • This type of task is not suitable for multiple adjustments afterward, as each adjustment may deviate from the original goal; it's best to express your needs accurately from the start, hitting the target 🎯!

How to write a qualified Lazy Prompt?#

A Lazy Prompt, as the name suggests, is a "lazy person's prompt"! How is it lazy?

  • Don't give too much interference; let AI play freely.
    • Just tell AI what you want, and don't worry about how it does it.
  • The prompt should have a bit of "vagueness"; don't lock the question too tightly.
    • But note, Lazy ≠ casual; the theme must be clear!
  • Allow the initial version to be imperfect; inspiration can be polished later!

✨ One-sentence secret: Clarity is essential, control should be minimal, and maintain a Zen mindset!

Little Tips: Don't be afraid to write freely; inspiration thrives in chaos, so trust AI to handle it boldly~

How to write a qualified structured Prompt?#

Let's first look at how AI thinks a structured Prompt should be written?

Divided into the following sectionsContent
Role Setting (Role)The role you want AI to play
Task Objective (Task)What task do you want AI to complete?
Input Information (Input)Provide necessary materials and background
Output Requirements (Output Requirements)Specify format, length, style
Constraints (Constraints)Prohibitions or special considerations

This is a quote from ChatGPT's response.
However, based on my personal experience, it can be simplified into three things:

  1. Background: What am I doing? (Briefly tell AI while protecting privacy)
  2. Objective: What do I want AI to help me with? (Describe in detail and clearly)
  3. Limitations: What do I not want AI to do? (Set boundaries to prevent AI from going off track)

Role Setting can be more personal; it's okay to omit it~
AI's Output Requirements can be flexibly included in Objective and Constraints!
A structured Prompt is best written in one go; frequent adjustments afterward can confuse AI!

Personal experiences in conversing with AI#

Let's put it into practice so you can understand instantly! 💥

Standard Play#

Applicable: Single task
Scenario: I am conducting a Traditional Chinese Medicine data mining task, but I need to standardize the names of syndromes and have saved the relevant extracted syndrome names in CSV format.

A regular Prompt might be:

Please help me analyze this CSV file and list the syndromes in a list.

What are the issues?
❌ It's too lazy! Rigorous and complex tasks are not suitable for a free approach; the results can easily become unrecognizable.

A structured Prompt should look like this:

 - Background:
 	- I have a CSV file with headers including `Category`, `Syndrome`, `Syndrome Name`, `Syndrome Manifestation`, with different data separated by `,` and synonyms separated by `/`.
 
 - Task Objective: Please write a Python script with the following requirements:
 	1. Count the frequency of `Syndrome Name`, sort from high to low, and generate `syndrome.csv`.
 	2. Count the frequency of `Syndrome Manifestation`, generating `syndrome_manifestation.csv`.
 	3. Summarize each `Syndrome Manifestation` with related syndromes, generating `syndrome x related syndrome comparison table.csv` (Note: in `related syndrome`, syndromes are separated by `,`).
 
 - Constraints:
 	1. The file path is `./input/syndrome_standard.csv`,
	2. The output path is `./output`. 
	3. Ignore empty cells.
 - (⏫ Upload CSV file)

Did you see it? Clearly stated:

  • Background (explaining the CSV file content)
  • Objective (requirements must be clear)
  • Limitations (further refine the requirements)

AI can naturally complete the task perfectly!

Hidden Play: Chain-of-Thought Prompt#

When the task is more complex, for example:

I want to analyze the relationship between a certain disease and syndromes, prescriptions, find core prescriptions and core herbs, and also do visualizations…
You can use the Chain-of-Thought Prompt to guide AI to reason step by step and complete in phases.

Here's an example 🌰:

- I am conducting data mining in Traditional Chinese Medicine and want to analyze the relationship between `a certain disease` and syndromes, prescriptions, and find the core prescriptions and core herbs. I have already saved the relevant data in CSV format. Below is an explanation of this CSV file:
	1. The headers are `ID`, `Diagnosis`, `Comorbidity`, `Syndrome Name`, `Syndrome Manifestation`, `Prescription`. (The header content can vary based on collected data; for example, `Tongue`, `Pulse`, `Gender`, `Age`, etc. can also be included in the analysis, but this is just a simple example)
	2. Data in `Comorbidity`, `Syndrome Name`, and `Syndrome Manifestation` is stored as a string, with different data separated by `,`.
	3. If a cell has no information, it is considered `empty`, and if the number of empty cells in a row exceeds `xx`, that row will not be included in the analysis. (The value for empty cells depends on how strict you are with your data filtering requirements)
	4. Data in prescriptions is stored as a string, with different herbs separated by `,`, and `herb` and `dosage` separated by `:`.

- Please write a Python script with the following requirements:
	- First, check the quality of the included data and create a statistics table for included and excluded cases.
	- Then analyze the cases:
		1. Frequency analysis of single herbs and herb pairs, generating md tables and visualizations. Also, take the frequency statistics of single herbs and their dosage data to create a multi-herb dosage distribution curve, with the x-axis as dosage, the y-axis as frequency, each line representing a different herb, using different colors for different herbs, and annotating what each line represents below the graph.
		2. Analyze the correlation of `Diagnosis` with `Comorbidity`, `Syndrome Name`, and `Syndrome Manifestation` (the three generated cross tables will be stored in the "Diagnosis Correlation Analysis.md" file).
		3. Analyze the correlation of `Diagnosis`, `Comorbidity`, `Syndrome Name`, `Syndrome Manifestation` with `Herbs` (the four generated cross tables will be stored in the "Herbs Correlation Analysis.md" file).

- Suggestions and considerations:
	1. The front part of the Python script should include a control panel for me to customize parameters and a general function section.
	2. Set the CSV file path as `./input/table.csv`, and the output md files should be saved in the specified path: `./output/analysis_results`. The output image files should be saved in: `./output/analysis_result_images`.
	3. Tables should be stored in md format.
	4. All cross table data should be sorted by frequency significance.
	5. The generated visualizations should be saved in png format, and the graphs should support displaying Chinese characters (Arial Unicode MS).
	6. The resolution of the visualizations should be adjustable, such as dpi=300.
	7. The frequency data used for tables and visualizations should be adjustable, such as a frequency of more than 3 times.
	8. The visualizations should display high-frequency data that can be adjusted, such as the single herb frequency graph showing the top 15, high-frequency single herb dosage distribution showing the top 5, high-frequency herb pair graph showing the top 10, and the graphs for cross tables showing the top 10 rows/columns.
- (⏫ Upload CSV file & switch to reasoning model)

Did you understand? The core structure remains:

  • What am I doing? (Tell AI the background and details)
  • What do I want you to help me with? (Requirements, step-by-step reasoning)
  • What do I want to limit? (I want to see the entire reasoning process of AI, let it output the script source code, keeping all steps under control, minimizing the risk of letting AI play freely)

Using the above methods will yield high-quality responses.


Summary#

When asking AI questions, the biggest fear is ambiguity and assumptions; the best approach is clarity and specificity without excessive restrictions.

  • Need inspiration? —— Use a short and free Lazy Prompt to let AI soar!
  • Need precision? —— Use a structured Prompt to let AI hit the target!

Having written this,
Congratulations, you have mastered the beginner's guide to conversing with AI! 🏆
Next time someone complains that AI "misses the point," you can confidently pat them on the shoulder: "Bro, it's your fault for not asking!" 😎


Disclosure: This article's content has been polished using Lazy Prompt to enhance fluency (writing tutorials as I go!), but the core content and experiences shared are all personally written by the author.

All illustrations in this article are AI-generated.

Thank you for reading, and I hope it helps you!

Loading...
Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.