Personal AI Setup
Stop re-explaining who you are and what you do every time you open Claude. Set it up once — with two layers: a Sunrise-wide operating contract, then your personal context on top.
!
Privacy by design. This page is a template, not a form. Nothing you type into the templates here ever leaves your browser or lands on Sunrise servers — copy the filled-in version to your own Claude / ChatGPT / Gemini and keep it there. Treat your system prompt like your password manager: it lives in your tool, not in a shared document.
The Sunrise AI System Prompt
Before you personalize anything, load the Sunrise operating contract into your AI tool. It's the same contract that Sunrise's Claude Teams account uses by default — it enforces Fair Housing guardrails, blocks tenant-screening assistance, requires verification + confidence ratings, and tells the AI to push back instead of agreeing to thin data. This is non-negotiable for any Sunrise work.
Paste it into your AI tool's system-prompt / custom-instructions / "personality" field, or include it at the top of every new conversation. Then layer your personal context on top using one of the two paths below.
Pick a path
How personalized do you want to go?
Both paths layer on top of the Sunrise System Prompt above. Pick by time and how deep you want the personalization to go.
A · 01Quick template
Copy the block below into a new Claude Teams Project's "Custom instructions" (or ChatGPT "Custom Instructions" / Gemini "Gems"). Replace the YELLOW blanks with your own answers. Three minutes, one time.
You are my work assistant at Sunrise Communities. I work in [role + department, e.g., "Asset Management as a Portfolio Analyst"].
My day-to-day responsibilities include:
- [responsibility 1, e.g., "Monthly KPI variance commentary for the Ohio portfolio"]
- [responsibility 2, e.g., "Acquisition due-diligence financial reviews"]
- [responsibility 3, e.g., "Resident communication drafts"]
I am fluent in: [areas of expertise, e.g., "MHP economics, NOI/cap rate, fair housing basics"].
I am NOT fluent in: [gaps, e.g., "tax law, SQL, multi-state regulatory differences"]. Explain those parts.
When you respond:
- Be direct. Skip "Great question!" and other pleasantries.
- Push back if I'm wrong. False agreement wastes my time.
- For non-trivial work, lay out your reasoning before the conclusion.
- For quantitative claims, show the math.
- Default to plain English. Only use jargon when I use it first.
- "I don't know" is a valid answer. Don't fabricate.
Default to: [e.g., "short bullets, headers when length warrants, code blocks for any tables"].
A · 02How to use it
- Pick your tool. Claude Teams (Sunrise-sanctioned, Tier 1–4) is the default. For personal experimentation only, Claude Pro / ChatGPT Plus / Gemini Advanced work — but stay below Tier 2 per the AUP.
- Create a new Project / Custom GPT / Gem. In Claude Teams: New Project → name it "YourName — Work Assistant". In ChatGPT: Settings → Personalization → Custom Instructions. In Gemini: Gems → New Gem.
- Paste the Sunrise System Prompt first (from the Anchor section above), then the filled-in template into the instructions field. Save.
- Test it. Open a new conversation in that Project and ask something role-specific. The AI should reference your role and skip the over-explanation. If it doesn't, your blanks are too vague — sharpen them.
- Iterate once a quarter. Your role evolves. Your prompt should too. Open the file, update the bullets, save.
B · 01Deep interview — the NYT reporter
Paste this opener into Claude Teams and let it interview you. The reporter persona is intentional: it signals "I want to understand you, not extract data from you." Take your time. Detail beats brevity.
As a reporter for the New York Times, my job is to understand the people I'm writing about. Let's get to know you a bit better. These questions are going to help me build a picture of your habits, preferences, and goals — ultimately, to help create a really effective personal assistant for you. Take your time and answer honestly. The more detail you provide, the better.
Let's start with the basics. What's a typical day like for you, from start to finish?
What are you most passionate about — a hobby, a cause, a field of study, anything that truly lights you up?
What are your biggest daily frustrations or time-wasters?
On a scale of 1 to 10, with 1 being "completely disorganized" and 10 being "meticulously organized," where would you rate your current level of organization?
What kind of technology do you already use regularly? (Smartphones, smart speakers, tablets, productivity apps — be specific!)
What are your top three priorities in your life right now? (Work, family, health, personal development — whatever's most important to you.)
Imagine you could have an assistant who could handle one specific task for you perfectly and reliably. What would that task be? (Think big — booking travel, researching a topic, anything.)
How do you prefer to receive information? (Email, text messages, voice commands, written summaries?)
What kind of personality would you describe your ideal personal assistant to have? (Friendly and casual? Professional and formal? Helpful and proactive?)
If this assistant could learn and adapt over time to better understand your needs, would you be comfortable with that?
Before we dive deeper — what's the most rewarding part of your day? And what's one thing you wish you could instantly eliminate from your life?
B · 02Structured questionnaire — for staff who want options to pick from
If the open-ended interview format doesn't fit your style, use the structured version. 10 questions with multiple-choice options for communication tone, analytical support style, etc. It also ships with a generator prompt that turns your answers into formatted Custom Instructions automatically.
Personalize Your AI Assistant — 10 Questions + Generator
Full markdown source: 10 questions across role, decision-making, learning style, analytical preference, communication tone, brainstorming, output format, uncertainty handling, red lines, and 12-month vision. Ends with a generator prompt that writes your Custom Instructions for you.
📋 Open Questionnaire
B · 03How to install your generated Custom Instructions
- Run the interview or questionnaire in Claude Teams. If you used the questionnaire, paste the generator prompt at the end + your answers — Claude will produce your Custom Instructions as a formatted output.
- Copy the Custom Instructions output. It'll be ~600 words structured into sections.
- Open Claude.ai → Settings → Profile → Personal preferences (sometimes labeled "Instructions for Claude").
- Paste your Custom Instructions there + Save.
- Every Claude chat from now on uses these instructions as starting context. No need to re-explain who you are.
Update once a quarter — your role evolves; your prompt should too. Re-run the interview or edit your Custom Instructions directly when you notice the AI missing something about how you work.
035 prompt patterns that work
Whichever path you used above, these patterns apply once your system prompt is loaded. Mix and match.
P1Instructional
You know exactly what you want. Give Claude the steps, the format, the constraints. Best for repetitive work.
"Draft a 3-paragraph resident notice for a water shut-off, calm tone, no legalese, specify date / time / impact / contact."
P2Exploratory
You're thinking out loud. Ask Claude to react, push back, surface options you haven't considered.
"I'm considering raising lot rents 12% at Oak Ridge. Talk me through the risks and what I'm probably missing."
P3Critique
You have a draft. You want it shredded honestly. Specify the lens you want.
"Critique this IC memo from a skeptical institutional LP's perspective. What would they pick apart first?"
P4Format-constrained
When you need a specific output shape, say so up front. Saves a re-prompt round.
"Return only a markdown table with columns: Risk · Likelihood · Mitigation. No prose."
P5Persona
You want the response framed by someone else's vantage point. Useful for empathy / stakeholder work.
"Respond as if you are a 70-year-old resident on fixed income reading this rent-increase notice. What do you feel? What do you ask?"
043 anti-patterns to stop doing
A1"Make it better"
Without a direction, "better" means more words and worse signal. Always name the dimension.
Bad: "Make this better."
Good: "Cut 30% length; keep all numbers; tighten verbs."
A2Pasting raw PII
Resident names, SSNs, full bank details — never. Even in Tier 4 sanctioned tools, redact identifiers and reason in abstractions. See the
What NOT to Upload reference.
Bad: "John Smith at lot 47 owes $1,240"
Good: "Resident at occupancy ratio 92% has 32 days past-due balance ~rent × 1.4"
A3Believing the output without verification
Bad: Forwarding a "draft sent" to Sam without re-reading.
Good: Two-pass: AI drafts → you read → you edit → then send.
Stuck on the blanks? Drop into #ai-at-sunrise on Slack — Carlos can walk you through filling them out. Don't paste your filled version into Slack publicly. DM it.
