The One-Time Setup That Makes AI Feel Personal
There is a feature in Claude called Projects that lets you load your context once so every future conversation starts already knowing who you are, how you write, and what you expect. Most people have seen the button. Almost nobody sets it up properly.
The payoff is immediate: you stop re-briefing the model at the start of every chat, and output quality jumps because the assistant is working from your real context instead of generic defaults.
This guide gives you the exact three-file setup and a copy-paste instruction block you can run in the next 20 minutes.
What Is a Claude Project?
A Claude Project is a persistent workspace where context accumulates across every session, so the model does not need re-briefing each time you open a new chat. You add custom instructions and reference files once, and every conversation inside that project inherits them automatically.
Custom instructions are the rules that run before every conversation in the project. Reference files are documents the model can read, such as your brand guide or past reports.
A single project holds roughly 200,000 tokens of reference material, which is around 300 to 400 pages. That is enough to carry your entire working context without re-uploading anything.
Why Custom Instructions Beat Retyping Context
Custom instructions beat retyping context because they run automatically before every message, so the model never starts cold. Instead of pasting your role and preferences into each chat, you define them once and they apply to every conversation in the project.
The practical difference is large. A marketer who defines their industry, audience, and banned phrases once gets on-brand drafts from the first reply, not the third.
Keep the instructions focused. The best practice is to cover the 20 percent of preferences that shape 80 percent of your work, and keep the whole block readable in about 90 seconds. A bloated instruction set wastes tokens in every single call and leaves less room for the actual task.
The Three-File Setup Power Users Rely On
The most reliable personal-assistant setup uses three short files: an identity file, a voice profile, and an anti-patterns file. Together they tell the model who you are, how you sound, and what to avoid, which does more for output quality than switching to a bigger model.
The identity file states who you are and what you are working on: your role, your company, your current projects, and your standards.
The voice profile captures how you think and write: sentence length, tone, favourite structures, and the level of formality you use with clients.
The anti-patterns file lists the words, tones, and structures the model should never use when writing as you. This is the file most people skip, and it is the one that stops output from sounding generically AI-written.
Try This Custom Instruction Block
Here is a complete, copy-paste-ready custom instruction block. Drop it into a new Claude Project, replace the bracketed parts with your own details, and every conversation in that project will follow it.
You are my work assistant. Context about me:
--- Role: [your role, e.g. marketing manager at a Hong Kong SaaS company]
--- Audience I write for: [e.g. small business owners aged 30 to 50]
--- My voice: direct, concrete, no corporate filler. Short sentences. One idea per paragraph.
--- Always: give me a usable first draft, then list the assumptions you made.
--- Never: use the words leverage, synergy, or unlock. Never open with In today's world.
When a request is ambiguous, ask one clarifying question before writing.
Paste that, start a chat, and compare the first reply to what you usually get from a blank slate.
What Makes an AI Assistant Drift Over Time?
An AI assistant drifts when its instructions grow too long, contradict each other, or are never updated. The model gives more weight to recent messages in a long chat, so an overloaded instruction set gradually loses influence as the conversation fills up.
The most common mistake is treating instructions as a dumping ground. Every extra rule dilutes the important ones and burns tokens on every call.
A second trap is staleness. Timestamp your instructions with the month you last edited them, and review them when your role or priorities change, so the assistant reflects the current you rather than last quarter's.
Bringing It Together
A personal AI assistant is not a smarter model. It is the same model with your context loaded once, through a Project and three short files. Set it up today and every future conversation starts warm instead of blank.
Setting this up alone can feel like guesswork, and that is where a partner helps. We understand AI. We understand you better. With UD by your side, AI doesn't feel cold.
Turn Your Setup Into a Team-Wide System
A personal assistant is the start. Rolling the same context infrastructure out across your team is where the productivity compounds. UD's team will walk you through every step, from instruction design to deployment, so AI genuinely works for your business.