Writing Prompts: The Key to Better AI Results

A prompt is any input to an AI system — from a simple question to ChatGPT to a structured instruction to an API. The AI model has no goal of its own. It responds to what you enter.

This means: the same AI delivers fundamentally different results on the same topic, depending on how you ask.

• Weak prompt: "Write me a text about Drupal."
• Strong prompt: "Write a paragraph for IT project managers explaining why Drupal is suitable for enterprise websites with high content volume. Tone: factual, no marketing language. Max. 80 words."

1. Assign a role: Give the model a perspective. "You are an experienced SEO consultant" delivers different results than "You are a copywriter."

2. Provide context: The more relevant information you include, the more precise the result. Target audience, tone, length, format — everything helps.

3. Specify the format: Do you want a list, prose, a table? Tell the model. Otherwise it decides on its own.

4. Give examples (few-shot): Show the model one or two examples of the desired result. This is often more effective than lengthy explanations.

5. Iterate: The first prompt is rarely perfect. Refine based on the results. Good prompt work is an iterative process.

For productive AI use, ad-hoc prompts are not enough. Businesses need standardized prompts for recurring tasks:

Content creation: Prompt templates for product descriptions, social media posts, or internal documentation. Optimized once, usable team-wide.

Customer service: Chatbot prompts that match your brand's tone and reliably deliver the right information.

Data analysis: Prompts that guide AI models to draw the right conclusions from your data.

Since 2012, arocom has built Drupal platforms. In AI integration, prompt design is part of the scope — because the best model delivers mediocre results without good prompts.