Prompts

Platforms: claude openai gemini m365-copilot

What Prompts Are

Prompts are the instructions you provide to an AI in natural language during a conversation. They’re ephemeral, conversational, and reactive — you provide context and direction in the moment.

Every AI interaction starts with a prompt. It’s the most fundamental building block — a well-crafted prompt is often all you need to get useful output, without any of the other blocks.

Key Characteristics

• Natural language — you write prompts the way you’d explain a task to a knowledgeable colleague
• Ephemeral — used in the moment, within a single conversation turn
• Reactive — you adjust and refine based on the AI’s response
• Range of complexity — from a single sentence to a detailed multi-section instruction with role, task, context, and format guidance

When to Use a Prompt Alone

A prompt by itself is sufficient when:

• The task is a one-off (no need to repeat it)
• The AI’s training data has everything it needs (no specialized knowledge required)
• The output format is simple or you can describe it inline
• You don’t need persistence across conversations

When you find yourself writing the same prompt repeatedly or needing to attach the same context every time, that’s a signal to consider other building blocks — Context, Projects, or Skills.

Anatomy of an Effective Prompt

The best prompts address four elements:

Element	What It Covers	Example
Role	Who the AI should be — expertise, perspective, personality	”You are a senior copywriter specializing in B2B SaaS”
Task	What the AI should do — the specific action or output	”Write three subject line options for this product launch email”
Context	Background the AI needs — constraints, audience, standards	”Our audience is enterprise CTOs. Tone should be authoritative, not salesy”
Format	How the output should be structured	”Present each option with a subject line and a one-sentence rationale”

Not every prompt needs all four elements. A simple question needs only the task. But as complexity grows, adding role, context, and format dramatically improves output quality.

Platform Implementations

Platform	How Prompts Work
Claude	Messages in conversation, system prompts, project instructions
OpenAI (ChatGPT)	Messages in conversation, system prompts, Custom GPT instructions
Gemini	Messages in conversation, Gem instructions
M365 Copilot	Chat messages, prompts within Copilot agents

Common Prompt Anti-Patterns

Vague instructions — “Help me with marketing” gives the AI nothing to anchor on. Be specific: “Draft a 200-word LinkedIn post announcing our Q3 product update, targeting engineering managers.”

Overloading a single prompt — Asking the AI to research, analyze, write, format, and review in one prompt leads to shallow results. Break complex work into sequential prompts or use a Skill.

Ignoring format guidance — If you don’t specify output structure, you get whatever the model defaults to. State what you want: bullet points, a table, a specific word count, or a particular template.

Repeating yourself every conversation — If you’re pasting the same preamble into every chat, you need a Project with custom instructions instead.

Guides

Guide	Description
Prompt Engineering	Core techniques — context windows, system prompts, few-shot learning, chain-of-thought
Project Instructions	When your prompts evolve into standing instructions for a project workspace

Related

• Context Engineering — the broader discipline that prompt engineering is part of
• Agentic Building Blocks — Prompts in the context of all building blocks
• AI Use Cases — see how prompts are used across content creation, research, coding, data analysis, ideation, and automation
• Projects — where prompts become persistent custom instructions
• Patterns — reusable prompt structures