Shop Notes
# How to Write Better AI Prompts
Most people treat large language models like search engines. They type a short phrase, hit enter, and hope for the best. While these models are capable of navigating vague requests, the quality of the result is directly tied to the clarity of the instruction. If you find yourself frustrated by generic, wordy, or slightly-off responses, the solution isn't a better model; it's a better prompt.
Learning how to write better AI prompts is less like coding and more like delegating a task to a smart, eager assistant who has no inherent context about your life or your goals. To get the best results, you must bridge that context gap.
## The Anatomy of a Good Prompt
A basic request usually says *what* to do. A professional prompt explains *how*, *why*, and *for whom*.
### 1. Define the Role
If you don't give the AI a role, it defaults to a general assistant. This often results in a neutral, encyclopedia-like tone. If you want a specific outcome, tell the AI who it should be.
* **Instead of:** "Write a blog post about gardening."
* **Try:** "Act as a professional landscape designer with 20 years of experience writing for a beginner audience."
### 2. Give Context (The Workshop Tip)
At the Branchstone workshop, we often say: A good prompt explains the task; a great prompt explains the context.
Context includes information that isn't strictly part of the command but influences how the command is executed. Who is the audience? What is the tone? What happened right before this request? If you are rewriting a rough idea into something more polished, using the [Prompt Improver](https://branchstone.app/tools/prompt-improver) can help you identify where these context gaps exist.
### 3. Set Constraints
Constraints are just as important as instructions. They prevent the AI from wandering into areas you don't need or using a style you dislike.
Examples of constraints include:
* "Do not use jargon."
* "Keep the response under 300 words."
* "Format the output as a bulleted list."
* "Avoid using common AI cliches like 'in today's digital landscape'."
## Leading by Example: Weak vs. Strong Prompts
Let’s look at how adding detail changes the output quality.
**Weak Prompt:** "Write an email to my boss about a project delay."
**The Result:** A generic, stiff email that might not sound like you or address the specific reason for the delay.
**Strong Prompt:** "I need to tell my manager, Sarah, that the Q3 Marketing Report will be two days late because we are waiting on the final data from the sales team. Act as a responsible employee who takes ownership. Keep the email brief, professional, and suggest a new delivery date of Wednesday at 9:00 AM. Do not make excuses; just state the facts and the solution."
**Weak Prompt:** "Give me some ideas for a healthy dinner."
**The Result:** A random list of salads or grilled chicken recipes that might require ingredients you don't have.
**Strong Prompt:** "Act as a nutritionist. Suggest three healthy dinner ideas that take less than 30 minutes to cook. I have chicken, spinach, and pasta in my pantry. Avoid recipes that require a food processor. Format each suggestion with a name, a list of ingredients, and one short sentence on why it's healthy."
## The Three-Step Polish Process
If you are struggling to move from a weak prompt to a strong one, follow this three-step cycle:
1. **The Brain Dump:** Write down exactly what you want in plain English, just as if you were talking to a friend.
2. **The Detail Layer:** Add the audience, the format, and the tone. Refer to specific names or data points.
3. **The Refinement:** Look at the prompt one last time. Is there anything the AI might misinterpret? If so, clarify it.
For those who want to speed up this process, the [Prompt Improver](https://branchstone.app/tools/prompt-improver) handles these steps for you. You provide the "rough draft" prompt, and it identifies the missing context and constraints needed to get a high-quality response from ChatGPT, Claude, or Gemini.
## Tailoring for Specific Models
While the core principles of prompt engineering remain consistent, different models have slight preferences:
* **ChatGPT (OpenAI):** Thrives on explicit instructions and step-by-step logic. If a task is complex, tell it to "think step-by-step."
* **Claude (Anthropic):** Excellent at handling long documents and maintaining a warm, human-like tone. It benefits from "XML tags" to separate different parts of your prompt (e.g., `<context>...</context>`).
* **Gemini (Google):** Great at integrating real-time information. It works best when you are clear about the creative versus factual nature of your request.
## Why Better Prompts Matter
Writing better prompts isn't just about getting a "prettier" result. It’s about efficiency. A bad prompt leads to a cycle of follow-up questions, edits, and manual corrections. A great prompt gets you 90% of the way there on the first try.
By treating your prompt writing as a craft—focusing on clarity, context, and constraints—you transform AI from a temperamental tool into a reliable partner in your daily work. Start small, be specific, and watch the quality of your AI interactions improve immediately.
7/2/2026 · 7 min read
How to Write Better AI Prompts: A Practical Guide to Prompt Engineering
A rough prompt gets a rough response. Learn how to refine your AI instructions with context, roles, and constraints for better results.
FAQ
What is the most important part of a prompt?
Context. Providing the AI with the 'why' behind the task and who the intended audience is usually makes the biggest difference in output quality.
Does the length of a prompt matter?
Not necessarily. A long prompt filled with fluff is worse than a short, precise one. However, detailed prompts that include specific constraints and examples usually perform better than one-sentence requests.
Can I use the same prompt for ChatGPT and Claude?
Yes, basic prompts work across all models. However, for complex tasks, you might want to tweak the formatting, as Claude likes structural tags and ChatGPT responds well to step-by-step instructions.
What are prompt constraints?
Constraints are 'rules' you give the AI, such as word counts, specific words to avoid, or a particular format (like a table or a list) that the response must follow.
Why does the AI sometimes ignore my instructions?
This usually happens when a prompt is contradictory or too vague. If the AI misses a step, try breaking the prompt into smaller, sequential instructions or putting the most important rule at the very end.