“I tried talking to the AI, but the answers I got back were kind of underwhelming…”
Have you ever had that experience?
Actually, the quality of an AI’s response is almost entirely determined by how you ask. Even when using the same ChatGPT or Claude, the quality of the output changes dramatically depending on how you write your prompt (the instructions given to the AI).
As of 2026, generative AI has fully integrated into the business world. However, the gap between those who “just use it” and those who “master it to achieve their intentions” is gradually widening. The key to this difference is prompt engineering.
It has a difficult-sounding name, but once you grasp the basics, you can start using it today. In this article, I have narrowed down seven techniques that actually deliver tangible results, aimed at business professionals, engineers, and creators. Please be sure to read until the end.
What exactly is “prompt engineering”?
In a nutshell, prompt engineering is the “art of designing instructions to elicit the desired output from an AI.”
You don’t need to write complex code like in programming. It is all about the “structure of the text and the choice of words” you write in Japanese (or English).
For example, even with the same request to “think of a marketing strategy”â
-
â “Think of a marketing strategy”
-
– â “You are the CMO of a B2B SaaS startup. Please propose 5 measures to double the number of monthly leads in 3 months, assuming a budget of 1 million yen.”
The specificity of the answers returned by these two is completely different. Actually, this is quite important. Just by clearly stating “what, who, and under what conditions,” the AI becomes surprisingly smart.
Technique 1: Assigning a role with “Persona Setting”
Giving an AI a “character” changes the answer
The most effective basic technique is to assign a role to the AI. Simply adding a sentence like “You are an expert in XX” at the beginning will drastically change the tone, depth, and expertise of the response.
-
“You are a UI designer with 15 years of experience.”
-
– “You are a science teacher who uses language that even middle school students can understand.”
-
– “You are an editor known for being harsh.”
If you are a creator, using the AI as a “veteran editor who gives tough feedback” can help you identify weaknesses in your writing that you might not have noticed yourself. Setting a role is the foundation for all other techniques.
Technique 2: “Format Instructions” to specify the output format
Telling the AI “how you want it to answer”
By default, AI tends to answer in long, wordy sentences. However, in actual work, you often need specific, easy-to-use formats like “bullet points,” “tables,” or “3-line summaries.”
Simply specifying the output format at the end of your prompt will result in material you can use immediately.
-
“Please summarize the answer in 5 bullet points or less.”
-
– “Please output in Markdown format as a comparison table.”
-
– “Please write one sentence for the conclusion first, followed by three reasons.”
For engineers, specifying formats like Markdown, JSON, or code blocks works particularly well. Try designing your format with the goal of making the output “ready to copy and paste.”
Technique 3: “Context Injection” – Providing Background Information
AI cannot answer what it does not know
AI knows nothing about your company or the details of your project. Therefore, it is important to provide the necessary background information yourself. This is called “context injection.”
Examples of information to provide:
-
Target audience attributes (“For working mothers in their 30s raising children”)
-
– Constraints (“Within 800 characters,” “Do not use technical jargon”)
-
– Past context (“The response rate for last month’s campaign was 1.2%”)
The more background information you provide, the more realistic the AI’s response will be. If you feel like you want a more realistic answer, try increasing the context first. The act of providing information itself is also a great opportunity to organize your own thoughts, killing two birds with one stone.
Technique 4: “Chain of Thought” – Making the AI Think Step-by-Step
Accuracy improves when you make it follow a “thinking process”
When asking the AI to solve complex problems, if you demand an immediate answer, it may make mistakes. This is where the “Chain of Thought” technique is effective.
By simply adding the phrase “Think step-by-step,” the AI will provide an answer while showing its intermediate reasoning process.
How to use it specifically
It is particularly effective in situations where accuracy is required, such as numerical calculations, logical analysis, and complex decision-making.
-
“First, organize the current issues, then list three solutions, and finally, prioritize them.”
-
– “Examine the pros and cons of this strategy before stating your final recommendation.”
By specifying the steps, you prevent the AI from ‘skipping steps in its thinking,’ making it more likely to return a logically consistent answer.
Technique 5: Showing Examples with ‘Few-shot Prompting’
Showing a sample by saying ‘like this’
When you ask a human to do a job, you show them a sample by saying, ‘Write it in this style,’ right? The AI is the same. By showing it 1 to 3 samples of the text you want it to output, it will reproduce the tone, style, and structure almost perfectly.
For example, if you want to create a social media post:
Example 1: “I realized this during yesterday’s meeting. 80% of the preparation determines the result.”
> Example 2: “The focus before a deadline is something you can’t achieve in daily life. So, being pushed isn’t so bad.”
> Write one post about the importance of reading in the same style.
Try using it with the image of teaching the AI the ‘voice’ of your own brand. You will feel the speed of content creation change by 3 to 5 times.
Technique 6: Using Constraints to Your Advantage with ‘Negative Prompting’
Clearly stating ‘what you do not want it to do’
AI is helpful, so it may do things you didn’t ask for. Greetings, disclaimers, excessive preambles… If you don’t need them, tell it clearly not to do them.
-
“Please omit preambles like ‘Certainly’ or ‘Understood’.”
-
– “Please do not use emotional expressions or overly positive words.”
-
– “Do not use bullet points; write everything in full sentences.”
Negative prompts reduce noise in the output. If you are worried that your text sounds too much like AI, this alone will improve it significantly.
Technique 7: Iterative Thinking through Dialogue
Don’t expect perfection on the first try
The final technique is about mindset. A prompt is not something that produces a perfect answer on the first attempt. The essence is the process of gradually approaching your ideal output through repeated conversation.
Once you receive the initial response:
-
“Please be more specific.”
-
– “Please delve deeper into the third idea.”
-
– “Please change the tone to be a bit more casual.”
By engaging in this kind of dialogue, you will often find ideas and perspectives that you wouldn’t have thought of at the beginning. I believe the real thrill of prompt engineering lies in using AI not as a ‘tool,’ but as a ‘thinking partner.’
Summary
I have listed seven techniques, but you don’t need to try to use them all at once. Starting today, just try these two: ‘assign a role’ and ‘specify the output format’âthat alone will definitely start to change your relationship with AI.
