How to Feed Reference Documents into System Prompts
Best practices for including reference material in system prompts and custom instructions without bloating every request.
System prompts are paid every turn
Whatever you put in the system prompt is sent with every message. A 5,000-token style guide costs 5,000 tokens per turn, so compression pays off fast.
Label and compress
Convert source material with the Markdown Converter, compress it with the Context Compressor while preserving product names and key terms, then confirm the size in the Token Counter.
Structure it
Use a clear layout:
- Role and goal (2 to 3 sentences)
- Rules and constraints
- Reference section wrapped in clear markers
- Output format
Try it free in your browser
- Context Compressor — Reduce token usage with compression strategies tuned for LLMs.
- Markdown Converter — Paste HTML, DOCX, or webpage text and get tidy Markdown.
- Token Counter — Live token, character, and context-window meter for any prompt. Uses official tiktoken encoders.
Tools mentioned
Frequently asked
Does the system prompt count toward tokens?
Yes, on every request. A bloated system prompt multiplies cost across a conversation.
Where should long documents go?
In retrieval or knowledge files, with only essentials in the system prompt.
Keep reading
How to Prepare PDFs for ChatGPT, Claude, and Gemini
A practical guide to extracting clean, AI-ready text from PDFs — born-digital and scanned — so ChatGPT, Claude, and Gemini answer accurately and don't waste tokens on headers, footers, and page numbers.
The Best Way to Feed Long Documents to Claude (and Other Long-Context Models)
Claude's 200K-token context is generous, but you'll still want to clean, compress, and structure long documents before sending them. Here's a step-by-step playbook.
Cleaning Whisper Transcripts for AI Summaries
OpenAI Whisper, Otter, and YouTube transcripts are full of timestamps, filler words, and speaker noise. Here's how to strip them before sending to ChatGPT or Claude — and why it matters.