System prompts & personas

Give Claude a persistent role that applies across an entire session

What is a system prompt?

A system prompt is instruction text that sits above the conversation — it’s the first thing Claude reads, and it frames every response that follows. In Claude.ai, this is called “custom instructions.” In the API, it is the system parameter. In Claude Code, it is part of CLAUDE.md.

Think of it as setting the stage before the show starts. Everything a user types is interpreted through that initial framing.

System prompts vs. in-conversation instructions

System prompt In-conversation
When set Before conversation starts Mid-conversation
Persistence Applies to every message Applies from that point forward
Use case Persona, standing rules, context Task-specific adjustments

Anatomy of a great system prompt

## Who you are
You are Alex, a senior digital strategist at BConverse Consulting, a bilingual EN/ES digital operations agency based in Texas.

## Your communication style
- Direct and confident, but not cold
- Use plain language — avoid tech jargon unless the person is technical
- Bilingual capability: respond in Spanish if the user writes in Spanish

## Standing context
- BConverse serves two client tracks: established SMBs (Track A) and Hispanic-owned businesses in Texas (Track B)
- When drafting client-facing content, default to warm but professional

## What you don't do
- Don't provide legal or financial advice
- Don't promise deliverables without prompting the user to confirm specifics first

System prompt: Permanent identity, standing rules, durable context about your business, output preferences that never change.

User message: The specific task, the content to work on, details that change task to task.

Build it: Write a system prompt for your most common Claude use case. Include: who Claude is, communication style, 2-3 standing rules, and 1-2 things it should never do. Save it somewhere you can paste it at the start of relevant conversations.