Author: will_tygart

A system prompt is the instructions you give Claude before the conversation begins — the context, persona, rules, and constraints that shape every response in the session. It’s the most powerful lever you have for controlling Claude’s behavior at scale, and the foundation of any serious Claude integration. Here’s how system prompts work, how to write them well, and real examples across common use cases.

System Prompt Structure: The Five Elements

A well-structured system prompt typically covers these elements — not all are required for every use case, but the strongest prompts address most of them:

System Prompt Examples by Use Case

Customer Support Agent

Code Assistant

Content Writer

What System Prompts Can and Can’t Do

System prompts are powerful but not absolute. They can reliably control: Claude’s tone and persona, output format and structure, topic scope and focus, response length guidelines, and how Claude handles specific scenarios. They cannot override Anthropic’s core guidelines — Claude won’t follow system prompt instructions to produce harmful content, lie about being an AI when sincerely asked, or violate its trained ethical constraints regardless of what the system prompt says.

System Prompts in the API vs. Claude.ai

In the API, the system prompt is passed as the system parameter in your API call. In Claude.ai Projects, the custom instructions field functions as the system prompt for all conversations in that Project. In Claude.ai standard conversations, you can prepend context at the start of a conversation — it’s not a true system prompt but achieves a similar effect.

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    system="You are a helpful assistant...",  # ← system prompt here
    messages=[
        {"role": "user", "content": "Hello"}
    ]
)

For a full library of tested prompts across use cases, see the Claude Prompt Library and Claude Prompt Generator and Improver.

Frequently Asked Questions

Claude’s context window determines how much information it can hold and process in a single conversation. Claude Sonnet 4.6 and Opus 4.6 support 1 million tokens; Haiku 4.5 supports 200,000 tokens — one of the largest in the industry. Here’s what that means in practice, what you can actually fit inside it, and how context window size affects your work.

Claude Context Window by Model (April 2026)

What Fits in 200K Tokens

Context Window vs. Output Length

The context window covers everything Claude processes — both input and output combined. If your prompt is 50,000 tokens (a long document), Claude has 150,000 tokens remaining for its response and any further back-and-forth. The window is shared between what you send and what Claude generates.

Maximum output length is a separate constraint — Claude won’t generate an infinitely long response even within a large context window. For very long outputs (full books, extensive reports), you typically work in sections rather than expecting Claude to produce everything in one pass.

Why Context Window Size Matters

Context window size is the practical limit on how much work you can give Claude at once without losing information. Before large context windows, working with long documents required chunking — splitting the document into pieces, analyzing each separately, and manually synthesizing the results. With 200K tokens, Claude can hold the entire document and answer questions about any part of it with full awareness of everything else.

This matters most for: document analysis and legal review, code understanding across large files, research synthesis across many sources, and long multi-step conversations where earlier context affects later decisions.

How Claude Performs at the Edges of Its Context Window

Research on large language models has found that performance can degrade somewhat for information buried in the middle of a very long context — sometimes called the “lost in the middle” problem. Claude performs well across its context window, but for maximum reliability on information from a very long document, referencing specific sections explicitly (“in the section about pricing on page 12…”) helps ensure Claude focuses on the right part.

For the full model spec breakdown, see Claude API Model Strings and Specs and Claude Models Explained: Haiku vs Sonnet vs Opus.

Frequently Asked Questions

Claude has usage limits on every plan — but Anthropic doesn’t publish exact numbers. Instead limits are dynamic, adjusting based on model, message length, and system load. Here’s what the limits actually look like in practice, what triggers them, and what your options are when you hit them.

Rate Limits by Plan

What Counts Against Your Limit

Claude’s limits are usage-based, not message-count-based. A single message asking Claude to write a 3,000-word article uses more of your limit than ten quick back-and-forth questions. What consumes the most limit, fastest:

• Long outputs — requests for long articles, detailed analyses, or extended code
• Long context — uploading large documents and asking questions about them
• Opus model — the most powerful model consumes limits faster than Sonnet or Haiku
• Agentic tasks — multi-step autonomous operations use significantly more than conversational use

API Rate Limits: How They Work

The API uses a different limit system from the web interface. API limits are measured in:

• Requests per minute (RPM) — how many API calls you can make
• Tokens per minute (TPM) — total tokens (input + output) processed per minute
• Tokens per day (TPD) — total daily token budget

New API accounts start on lower tiers and can request higher limits through the Anthropic Console as usage establishes a track record. The Batch API has separate, higher limits since it’s asynchronous and non-time-sensitive.

What To Do When You Hit a Limit

Frequently Asked Questions