I stopped prompting and started keeping files. My Claude output got dramatically better.
retained text
I used Claude every day for months and kept hitting the same wall. Long chats drifted. Every new chat, I re-explained who I am and how I write. And everything it wrote sounded like AI — clean, confident, and not me. So I stopped trying to write better prompts and started keeping files instead. Four of them. That's the whole method. 1. A profile file. Not a bio — a file that says who I am, how I write, what I'm building, and what I refuse to do. I built it by answering about 100 questions out loud (dictated, not typed), then compressing 20,000 words of answers into under 2,000. That file goes into every project now. 2. A banned-words file. I collected my own drafts, circled every phrase that sounded like AI wrote it, and banned them. About 40 entries. The file works because the words came from my drafts, not a list I copied off the internet. 3. A context folder. One folder with four short files: who you are, how you write, what you're building, what you refuse. New chat, attach the folder, no re-explaining. 4. A playbook file. For any task I repeat, I write the steps down once — the trigger, the steps, and a "not for" line saying when it doesn't apply. Then I reuse it instead of re-prompting. The difference, concretely: before, "write a LinkedIn post" got me generic AI voice. After, the same request with my profile + banned-words file attached gets me something I only have to edit, not rewrite. One more thing nobody tells you: when Claude starts ignoring your files, don't argue with it in the same chat. Restart the chat, re-attach the files, and check whether the file itself is the problem. Half the time it is — usually it's too long or contradicts itself.
document fetched 2026-11-15 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xw60r5/i_stopped_prompting_and_started_keeping_files_my