How to Write Prompts: Tips to Dramatically Improve Answer Quality

AI Navigate Original / 3/23/2026

💬 OpinionIdeas & Deep AnalysisTools & Practical Usage
共有:

Key Points

  • The foundation of a great prompt is three elements: <strong>role definition</strong>, <strong>providing context</strong>, and <strong>specifying the output format</strong>. With just these, you can significantly improve both accuracy and reproducibility.
  • In a Before/After comparison, prompts that clearly state the target reader, purpose, and constraints are more likely to produce outputs that are directly usable in real work.
  • Rather than forcing the model to reveal its <em>Chain-of-Thought</em>, it’s more practical to have it structure the decision process—e.g., <em>organize assumptions → compare options → recommend a plan</em>.
  • Few-shot is a technique where you show 1–3 examples so the model matches the tone and output pattern. It’s especially effective for reply messages to inquiries and classification tasks.
  • Step-by-step breaks down complex requests into phases to reduce failure. It’s useful for tasks like problem framing, root-cause analysis, and suggesting improvements.

Introduction: Even the same Claude can answer very differently depending on how you ask

When you’re using Claude and feel that the answers are “too shallow,” “not aligned with your intent,” or “too long,” the cause is often not the model’s performance—it’s in the prompt design. In particular, the first things beginners should learn are these three: role definition, providing context, and specifying the output format. Just doing this already improves answer accuracy and reproducibility a lot.

As of 2025, Claude is extremely good at understanding long-form text, summarization, writing, and analysis. At the same time, it tends to respond vaguely to vague requests. In other words, the key is to use it with the mindset of giving AI a better instruction sheet, rather than dumping work on it.

The basic structure of a “good prompt” you should learn first

In real work, the following template tends to work reliably.

You are[role].
Objective:[what you want to do]
Context:[background, target audience, constraints]
Input:[materials or situation]
Output format:[bullet points, table, headings, character count, etc.]
Notes:[what to avoid, points to prioritize]

For example, if you want a draft for an internal email, it looks like this.

You are an assistant knowledgeable about business documents from Japanese companies.
Objective:I want to create a email to coordinate a delivery date with a business partner.
Context:The message should be respectful to the recipient, and we also want to communicate our situation honestly.
Input:Due to a delay in procuring components, delivery is expected to be 3 days late. As an alternative, we can proceed with partial advance delivery.
Output format:Three subject line options, and one body message in formal polite Japanese.
Notes:Do not sound like an excuse; include specific countermeasures.

Just using this structure will produce results far more practical than simply asking, “Write a delivery-delay email.”

Seeing the effect of prompt improvements with Before / After

Example 1: Blog article outline ideas

Before

Claude, come up with an outline for an SEO article.

With this, the theme, the reader, the target keywords, and how deep the article should go are all unclear.

After

You are an SEO content editor.
Objective:Create a blog outline for the theme “I can’t focus while working from home.”
Context:The readers are office workers in their 1–3 years at the company who work remotely. The search intent is “I want improvement tips I can try right away.”
Output format:
- Three article title ideas
- h2/h3 structure
- For each heading, 1–2 sentences describing the key points to write
Notes:Focus on concrete, action-level measures—not motivational talk.

The improvement comes from clearly stating the target reader, search intent, and output format. As a result, you get an outline that’s not abstract, but actually easy to turn into a real article.

Example 2: Summarization requests

Before

Summarize this document.

After

Sign up to read the full article

Create a free account to access the full content of our original articles.