Have you ever felt like, when talking to an AI, it just doesn’t give you the answers you’re looking for?
Actually, that is often not a problem with the AI’s performance.
As of 2026, generative AI, including ChatGPT and Claude, has become an everyday tool for many business professionals and creators. However, the gap between those who can use it effectively and those who cannot is gradually widening.
What is creating that gap is “prompt engineering.”
It sounds difficult, but the essence is simple. It is the skill of refining how you talk to an AI. And with just a few simple tips, the quality of the output can change dramatically.
In this article, I will introduce basic prompt techniques that you can use with zero prior knowledge, along with concrete examples. Please try them out starting today.
What exactly is a “prompt”?
“Prompt” refers to the input text you provide to an AI. In other words, it is the message you send to the AI itself.
To use a cooking analogy, the AI is the chef, and the prompt is the recipe. No matter how talented the chef is, if you just say “make something delicious,” you won’t get the dish you were imagining, right? If you give instructions like “use chicken breast, marinate it in shio-koji, and make it a light, set-meal style dish,” you are much more likely to get exactly what you expected.
Prompt engineering is the technique of improving the quality of these instructions.
Why is this skill important now?
As of 2026, while the use of generative AI has become commonplace, there are increasing reports of people saying, “I use it, but I don’t see results.” One survey found that among companies that have introduced AI tools, only about 40% feel that their operational efficiency has clearly improved. The results are in.
The key to bridging this gap is the quality of the prompt. This is a skill that relates to every profession, not just engineers, including planners, marketers, writers, and sales staff.
Basic Technique 1 | Assigning a Role (Role Prompting)
This is quite important. Simply setting a role for the AI, such as ‘You are an expert in X,’ will change the tone and expertise of the response significantly.
**Before (No role)**
> Tell me about marketing strategy
**After (With role)**
> You are a marketing consultant specializing in the growth of B2B SaaS companies. Please propose a content marketing strategy for small and medium-sized enterprises, including specific KPIs.
Even with the same question, the depth and accuracy of the returned answer will be completely different. Give it a try; you will be surprised.
The key is to clarify ‘who’ and ‘what’
What is important in role setting is specificity. Rather than just writing ‘expert,’ adding background information like ‘a UX designer with 10 years of experience’ will further increase accuracy. Also, adding the target audience for the response (e.g., ‘for beginners,’ ‘to explain to management’) will automatically adjust the writing style and difficulty level.
Basic Technique 2 | Adding Constraints (Constraint Setting)
‘Write it more concretely,’ ‘Make it easier to understand’—are you giving these kinds of vague follow-up instructions?
Actually, AI’s accuracy improves significantly when you apply constraints using numbers or conditions.
**Vague instruction**
> Explain it clearly
**Instruction with constraints**
> – Keep the character count under 300 characters
> – Do not use technical jargon; use language that even a middle school student can understand
> – Do not use bullet points; use a conversational tone
This difference is truly significant. By simply quantifying the conditions, the AI’s ‘interpretation drift’ is eliminated, making it easier to get an answer close to what you want on the first try.
Think of constraints in terms of three axes: ‘format,’ ‘style,’ and ‘target’
Constraints can be organized into three main axes: 1. Format (character count, bullet points, table format, etc.), 2. Style (formal, casual, polite tone, etc.), and 3. Target (who the text is for). If you create prompts with these three axes in mind, you will reduce the number of times you think, ‘This isn’t what I expected…’
Basic Technique 3 | Showing Examples (Few-shot Prompting)
It is difficult to convey intuitive nuances like “I want you to write in this tone.” That is when you can use the method of showing examples.
Simply attach a sample to your prompt, saying things like “Use this text as a reference” or “Follow the same style as the example below.” The AI learns the pattern from the example and attempts to reproduce its tone and structure.
This is especially recommended for writers and creators. If you provide a sample of text you have written in the past, the AI will mimic your “personal writing style.” Even if it cannot reproduce it perfectly, it is much easier to achieve your own personal touch than generating from scratch.
Even “one” example is effective
Despite the name “Few-shot,” it does not mean that multiple examples are required. In reality, providing just 1 to 3 examples is sufficient in most cases. In fact, if there are too many examples, the AI may get confused about “which one it should follow.” Start by providing one of your favorite samples.
Basic Technique 4 | Breaking Down Steps (Chain of Thought)
If you try to make the AI solve a complex problem all at once, the answer tends to be superficial.
That is where you can use the method of explicitly stating the steps of thought, such as “First think about X, and then propose Y based on that.” This is called “Chain of Thought.”
**Asking all at once**
> Come up with 10 ideas for a new business
**Breaking down the steps**
> First, list 3 unresolved issues in the aging society as of 2026. Next, for each issue, think of one business idea that can be solved with technology. Finally, choose the one you think has the largest market size among the three and explain why.
Even for the same “idea generation,” the latter will produce much more well-founded and usable ideas.
The magic syntax of “First…, next…, finally…”
When practicing Chain of Thought, it is recommended to use conjunctions like “first,” “next,” and “finally.” This helps the AI accurately grasp the order of processing and makes it easier to build up logical thinking. It is particularly effective for “thinking-based” tasks such as complex analysis, strategy formulation, and text composition.
Three Common Failure Patterns and Countermeasures
Knowing common failure patterns is even more effective than just knowing the techniques.
**1. Too little information**
Instructions like “make it look good” are difficult for an AI to interpret. Be sure to provide background information, objectives, and the target audience as a set.
**2. Expecting completion on the first try**
Prompting is a “dialogue.” The correct approach is to look at the initial response and continuously refine your instructions by saying things like “do more of this” or “change this part.” Repeating the dialogue will get you closer to your ideal result faster than aiming for a perfect one-shot prompt.
**3. Prompts that are too long**
If you pack in too many conditions, the AI may not be able to process all of the instructions. Try to focus on narrowing down the most important conditions to three to five points.
Summary
Prompt engineering does not require special qualifications or expert knowledge—it is a skill that anyone can start using today just by being a little more careful about “how they communicate.”.
Role prompting, constraints, Few-shot, and Chain of Thought. Try these four out little by little and develop your own unique “way of talking to AI.”
It is okay if things don’t go well. You can rewrite prompts as many times as you like. That trial and error itself is the shortest path to turning you into someone who can master AI.
