Stop re-introducing yourself to your AI
Three files, written once, and your AI starts every chat already caught up.
By Chris Battis
Maria runs a solo recruiting shop. Six placements a month, no team. Monday morning she opens a new chat to draft outreach for a VP of Sales role. Before she can even ask for the email, she types: who the client is, what the role pays, her fee structure, the tone she wants, and the questions every candidate asks that she needs pre-empted.
Four paragraphs. Tuesday: the same four paragraphs, new candidate. Thursday: the same four paragraphs, different client.
By Friday she isn't writing outreach anymore. She's re-onboarding an assistant who forgot her business exists, five times that week.
- Who you areWritten once
- What good looks likeWritten once
- Right nowUpdated each week
Two kinds of context, and you're only writing down one
A model is trained, then it's frozen. Every new chat starts from that same frozen snapshot, no matter what you explained yesterday. Anything it knows about you has to come from what you hand it, and what you hand it splits into two kinds.
Frozen knowledge is the stuff that barely changes: who you are, who you sell to, your voice, your fee structure, the questions candidates always ask. Set it down once and it stays true for months.
Living context is state: what's actually happening right now. Which searches are open this week, what a client just changed about the comp range, what's on your calendar today.
Most people who "set up AI" only ever handle the first half, and only inside a single chat that evaporates when the chat ends. Maria never wrote her frozen knowledge down anywhere permanent, so she's retyping it forever. And even the businesses that do write something once almost never touch the living half at all. The model has no idea a client changed terms this morning, because nobody told it and nobody has a place to.
Three files. Fifteen minutes. Never explain it again
Paste these into a new chat, or drop them into a Claude Project or a custom GPT's files, and the model starts every session already caught up.
FILE 1: who-you-are.md (frozen, edit rarely)
- What you do, in one sentence
- Who you sell to
- Your voice: 3 adjectives, one thing you never sound like
- Your fee/pricing structure
- The questions people always ask you
FILE 2: what-good-looks-like.md (frozen, edit rarely)
- The format you want output in (bullets vs. prose, length, structure)
- One example of output you were happy with
- One anti-pattern: a mistake the model keeps making that you want stopped
FILE 3: right-now.md (living, edit weekly)
- What's actually open this week: active deals, projects, searches
- Anything that changed since last week
- What you're prioritizing the next few days
Files 1 and 2 are the onboarding document you write once. File 3 is the part almost everyone skips.
File 3 is the one that matters this week
Files 1 and 2 fix the first message of every chat. File 3 fixes every message after it. Without it, the model can recite Maria's fee structure perfectly and still have no idea she has three live searches open right now, or that one client moved the comp range yesterday, because that information only exists in her head and her inbox, not anywhere the model can read it.
The fix doesn't need to be automated. Two minutes on Monday morning, updating a text file, does more than a calendar connector nobody configured. If you already use Claude or ChatGPT connectors for email and calendar, that's the same job done live instead of by hand, but the plain file is where to start, because it works today with zero setup.
Same three files work in Claude, ChatGPT, or Gemini. The model is interchangeable. The files are the asset.
