Setting Your Personal System Prompt
It takes about five minutes and it applies to every conversation you have from then on.
Why this exists
AI CTRL starts every conversation cold. It does not build a profile of you over time, it does not carry anything from one chat into the next, and it does not receive any information about you from your company's directory: it doesn't know your name, your job title, or what team you're on.
That is deliberate. Nothing is remembered unless someone deliberately puts it there.
The trade-off is that without any context, the model is guessing at three things every single time you open a chat: who you are, what you work on, and how you want to be answered. Your personal system prompt answers all three, once.
Where to find it
- Click your profile picture in the corner of the screen
- Open Settings
- Select General
- Enter your instructions in the System Prompt field
- Save
This setting is yours alone. It is not configured by your administrators, it does not affect anyone else, and you can change it whenever your work changes.
What to put in it
Three things, in this order.
1. Who you are
Your role, your team, and what you are responsible for. This is what lets the model pitch an answer at the right level instead of explaining things you already know or assuming expertise you don't have.
2. What you work on
The systems, products, clients, and subjects that come up repeatedly, and the vocabulary your organization actually uses. If your company says "engagement" where another says "project," this is where you say so.
3. How you want to be talked to
Format, length, tone, and level of detail. Bullet points or prose. Summary first or reasoning first. Whether you want to be asked clarifying questions or given a best attempt.
Be specific
Vague instructions produce vague results. A good instruction describes something only you would say; a vague one could describe anybody.
| Instead of | Write |
|---|---|
| I work in IT | I manage server infrastructure for a mid-sized manufacturer. My audience is usually non-technical leadership. |
| Keep it short | Lead with the answer in two sentences, then supporting detail in bullets. No preamble. |
| Be professional | Plain, direct language. No exclamation points. Avoid marketing phrasing. |
| I work with data | I spend most of my time in Excel building operational reports. Assume I know spreadsheet functions but not SQL. |
The fastest way to write one
Don't write it from a blank page. Open your preferred model and type:
help me build my personal system prompt
The model will ask you what it needs and draft one for you.
Paste the result into the field, use it for a week, and adjust the parts that don't fit.
What to avoid
Instructions that contradict each other. Asking for thorough and concise in the same breath gets you neither. If two instructions pull in opposite directions, pick one or say which wins.
Starting too broad. A short, specific prompt beats a long, general one. Start narrow and add to it as you notice what's missing. That is far easier than working out which of forty lines is causing a problem.
Sensitive content. Your personal system prompt is sent with every message you write, so treat it exactly as you would treat anything else you type into a conversation.
Personal system prompts and custom models
Both are called system prompts, and they do related but different jobs.
| Personal system prompt | Custom model system prompt | |
|---|---|---|
| Who it's for | You | Everyone who uses that model |
| What it describes | You and your preferences | A workflow and its rules |
| Applies to | Every conversation you have | Only conversations with that model |
| Who can change it | You | Whoever has write access to the model |
They stack. A custom model's instructions handle the job; yours handle how the answer reaches you.