# The first file Everything I teach starts with one file. Here's who I am, here's what I'm working on, here's how I like things done. You tell your AI to read it first, every time. That's it. That's the whole trick that makes an AI feel like it knows you. Your AI doesn't know you yet. That's fixable, and this is how you fix it. ## How to use this 1. Get the template below into anywhere that holds text. A Word doc, a note on your phone, a plain text file - the format doesn't matter and the AI doesn't care. If your dictation works best in Word, use Word. 2. Fill in the blanks - out loud, not by typing. (More on that in a moment.) 3. Hand the result to your AI and let it do the filing. You don't need to know about file types or where things go - paste it in and tell the AI what it's for. The exact move per tool is below; each takes under a minute. 4. Update it as things change. A first file from six months ago describes a you from six months ago. Start tiny means start here. You do not need the perfect version. You need a version that exists. ## Fill it by talking, not typing Here's the part nobody told me and I now tell everyone: don't type your answers. Dictate them. Use the microphone key on your phone's keyboard, or whatever dictation your computer has, and just talk through the blanks - who you are, what you're working on, how you like things done. Two reasons. First, speed of capture: ideas evaporate. If I had to type everything out before a thought went away, I'd get maybe half of it - probably less. Talking gets it out while it's still whole. Second, and this took me longer to notice: I'm talking *to* the AI, not *with* it. I'm talking my ideas out, the way you would to a colleague who's good at listening - and the act of saying them out loud is what refines them. The AI's job is to read what I said and make it useful. It doesn't need to talk back to me (I don't like the talking-back kind, honestly - you might), and it doesn't judge the rambling. So: ramble your answers into the file, or straight into the AI, then say **"turn what I just said into my first file, using this template."** That's your first collaboration, done - and the file wrote itself. ## Give it to your AI - the exact move, per tool **ChatGPT.** Settings -> Personalization -> Custom Instructions: paste the "Who I am" and "How I like things done" parts there - now every conversation starts already knowing you. Or, if you use Projects, paste the whole file into the project's instructions. **Claude (the chat).** Make a Project (it's free on paid plans) and paste the file into the project's instructions, or add it as project knowledge. Every chat inside that project reads it automatically. **Claude Cowork or Claude Code.** Paste your filled-in answers into a session and say: **"This is my first file. Save it in this folder, and read it at the start of every session."** Claude handles the file part itself - letting the AI do the filing is rather the point of these tools. (Already have a `CLAUDE.md` from setting up Claude Code? Then say "add this to my CLAUDE.md" instead - that file is read automatically every session, so your context rides along for free.) **A work tool you can't configure (Copilot in the sidebar, a locked-down chat).** Paste the whole file as your first message: "Here's my context - read it, then help me with X." It holds for that conversation. Not elegant, but it works, and you'll feel the difference immediately. --- ## The template Copy everything below the line into your own file. --- # Who I am *A few plain sentences. What you do, what you're trained in, what you're not. Include the honest limitations - they help the AI help you.* - I'm a ______ (role / profession). - My background is ______ (training, field, years). - What I'm NOT: ______ (say what you don't have, so the AI doesn't assume it - e.g. "not a developer," "not a designer," "no formal stats training"). - The one thing to never get wrong about me: ______ (a credential, a fact, a boundary the AI should treat as fixed). # What I'm working on *The current picture. Update this often - it goes stale fastest.* - Right now I'm focused on: ______ - The projects that are live: ______ - Who I'm making this for (my actual reader / user / audience): ______ - What "done" looks like for me: ______ # How I like things done *Your working preferences. This is where the AI learns to sound like your partner instead of a stranger.* - Talk to me like: ______ (e.g. "a knowledgeable colleague, not a lecturer"). Plain language, no hype. - When you're not sure, ______ (e.g. "say 'I don't know' - don't guess and don't fill the gap with something that sounds right"). - Give me ______ (e.g. "a recommendation, not a menu of ten options" - or the reverse, if you prefer choices). - Before big work, ______ (e.g. "show me the outline / the brief first, then wait for my go-ahead"). - Formatting I like: ______ (e.g. "sentence case headings," "no emoji," "short paragraphs," "American spelling"). # What I care about being true *The verification standard. Tell the AI to hold your work to it.* - Everything that ships under my name gets checked. If you make a factual claim, tell me where it comes from. - Don't invent numbers, citations, or quotes. If you're not sure something is real, flag it instead of stating it. - If I ask "is that true" or "did you make this up," give me the honest answer, not the reassuring one. --- ## Why this works An AI is only as useful as what you give it. Out of the box it knows a general version of the world and nothing about you: not your job, not your reader, not the way you like things phrased. The first file closes that gap in one move. You write down the context you'd otherwise re-explain every session, and then you stop re-explaining it. This is also where my whole system started. One file, read first, every time. Everything else grew from that. You don't have to build the everything else. You just have to start the file. If I can do it, you can do it. Open a blank file and fill in the first three blanks. That's a real first file. Add to it tomorrow.