Give ChatGPT and Claude Persistent Context with 3 Markdown Files
If you use ChatGPT or Claude for real work, you've probably typed some version of this dozens of times: "I'm building a tool for freelancers that... our tone is friendly but not cheesy... please don't use buzzwords..." Every new chat starts from zero. The model doesn't know your product, your audience, or the things you hate seeing in its output. So you either re-explain everything (slow) or skip…
Working with ChatGPT or Claude for professional tasks often requires repeating the same explanation: the purpose of your tool, the tone you desire, and the rules you expect. Each new chat starts from scratch, unaware of your product, audience, or what you dislike in the output. This repetition slows down the process and leads to generic results.
A simple solution is to store your context in three small Markdown files and attach them at the start of every session. The files include:
1. product.md: a concise description of what you're building, who it's for, and why.
2. voice.md: guidance on the tone and style you want the output to follow.
3. rules.md: hard constraints and output format requirements.
Markdown files have several advantages over built-in memory features or custom instructions:
- Portability: These files work across different platforms and tools.
- Version control: Store them in a Git repository or notes folder to track changes over time.
- Precision: You decide precisely what the model sees, avoiding guessing about its memory from previous chats.
- Scope: Maintain different sets of files for various projects without interference.
The first file, product.md, serves as the source of truth for your product. Keep it brief, focusing on the essential aspects. The second file, voice.md, helps steer the model toward your preferred tone. The third file, rules.md, outlines non-negotiable rules regarding accuracy, output format, and the model's behavior.
Once created, these files require only five minutes to set up and seconds to reuse. To begin each session, attach the files or paste their contents with a brief instruction for the model to read them without summarizing. After that, provide a specific task for the model to complete, ensuring adherence to the established context, voice, and rules.
Iterate on the model's output and update the relevant files accordingly. Over time, these files will refine themselves based on your feedback. Using project features like ChatGPT Projects or Claude Projects allows you to attach the files once, making the process even more efficient.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
