# Personalize Your AI Assistant **For Sunrise staff using Claude (claude.ai)** This document helps you customize Claude so it works the way *you* work. Once Claude knows your style, your role, and your preferences, it becomes far more useful — and you'll spend less time re-explaining yourself in every chat. This is about **scope expansion, not measurement.** We're not capturing this to track output or optimize your workflow on someone else's behalf. We're capturing it so the AI can take on more of the routine work and free you up for the higher-judgment parts of your job. **Time to complete:** 20–30 minutes **You can pause and resume.** **There are no wrong answers — be specific and honest.** --- ## How This Works (3 Steps) 1. **Answer the 10 questions below.** Fill in your responses under each question. Concrete examples beat abstract preferences. 2. **Paste your finished answers + the generator prompt at the bottom into your assigned Sunrise Project on claude.ai** (e.g., the Finance Project, Asset Management Project, etc.). Running it inside the Project gives Claude the Sunrise context it needs to write better instructions for you. 3. **Paste the output into claude.ai → Settings → Profile → Personal preferences (Instructions for Claude).** Save. Done. From that point on, every Claude chat starts with your context loaded. You can iterate later — re-run this anytime your role shifts or your preferences evolve. --- ## The 10 Questions ### 1. Role grounding In your own words, what's your role at Sunrise? Not your job title — what you actually spend your time on, what decisions land on your desk, and what "a great quarter" looks like for you. > *Your answer:* --- ### 2. Hard call Tell me about a hard decision you made recently at work. What were you weighing, and how did you land? Help me understand what you considered, what you discounted, and what made it hard. > *Your answer:* --- ### 3. Learning style When you have to learn something new for work, how do you actually learn — reading, asking someone, video, hands-on experimenting, or a mix? Give an example from the last six months. > *Your answer:* --- ### 4. Analytical support When you're working through something complicated, what do you want from the analysis? Pick the closest: - **(a)** A clean recommendation with reasoning - **(b)** Several options with trade-offs so I can choose - **(c)** My assumptions challenged so I can stress-test my thinking - **(d)** Raw data — I'll do the synthesis myself And: does that change depending on the situation? > *Your answer:* --- ### 5. Communication tone How should an AI talk to you? Pick the closest: - **(a)** Brief and direct — get to the point - **(b)** Detailed and educational — walk me through your reasoning - **(c)** Conversational, like a smart colleague - **(d)** Formal, like a consultant's deliverable Then: when does your answer change? (e.g., "A when executing, B when learning.") > *Your answer:* --- ### 6. Brainstorming style When you're stuck or generating ideas, what kind of help works best — lots of options without filtering, pushback on your assumptions, building on your direction, or playing devil's advocate? Describe what "genuinely helpful" looks like versus "just agreeable." > *Your answer:* --- ### 7. Output format How do you like receiving information — bullet points, tables, full prose, numbered steps? What's your least favorite, and why? Do you want a summary at the top, or straight to the answer? > *Your answer:* --- ### 8. Honesty under uncertainty When the AI doesn't know something for sure, what should it do — always give a definitive answer, flag uncertainty, refuse to answer, or give confidence levels (HIGH / MODERATE / LOW)? What's worse for you: confidently wrong, or constantly hedging? > *Your answer:* --- ### 9. Red lines Name 2–3 things that would make you stop trusting the AI. Could be making things up, sycophantic agreement, mixing personal and work context, taking action without confirming, sharing your data with anyone — whatever feels like a deal-breaker. > *Your answer:* --- ### 10. Signature moves & 12-month vision If this AI became your most useful work tool, what would it consistently do that no one else does for you today? And looking out 12 months — what would you want it to grow into? > *Your answer:* --- ## The Generator Prompt After answering all 10 questions above, copy everything between the `---START---` and `---END---` markers below into a new Claude chat (claude.ai web). Replace `[paste your answers here]` with your filled-in answers. ``` ---START--- You are writing personalized AI assistant instructions for a Sunrise Communities employee based on their interview answers below. Sunrise is a manufactured housing portfolio operator (45 communities, 8,250+ lots) with a value-add investment model. Your job is to translate their answers into a Custom Instructions document they will paste into claude.ai's "Instructions for Claude" setting. OUTPUT FORMAT (must match exactly, including markdown headers): # AI Custom Instructions for [Name, Title] ## I. My Context & Role [2-4 sentences. What they actually do, who they work with, what success looks like. Specific to their answers, not generic.] ## II. How I Think & Work ### Problem Solving & Idea Generation [Their decision-making texture from Q2. What they weigh, where they feel friction.] ### Learning Style [How they learn from Q3. Use their example.] ### Analytical Support [What kind of analysis they want from Q4. Note context-shifts if mentioned.] ## III. Communication & Output Preferences ### Tone & Voice - **For information/answers:** [from Q5] - **For idea generation/brainstorming:** [from Q6] - **For work communications:** [if they mentioned a different mode for drafts/emails; if not, default to their information voice] ### Deliverables [Format preferences from Q7.] ### Accuracy & Honesty [Uncertainty handling from Q8.] ### Ethical Considerations [Red lines from Q9.] ## IV. My Ideal AI Assistant ### Signature Moves [2-4 distinctive things from Q10, first half. Not generic AI cliches.] ### Future Aspiration [Where they want it to grow from Q10, second half.] RULES: - Use THEIR language where possible. If they said "punchy," don't say "concise." - Be specific. "Brief and direct" is generic. "Brief, under 5 bullets unless I ask for more" is specific. - No AI flattery. No "you are a brilliant strategic partner." Just describe what they need. - If they contradict themselves, surface the context-dependent rule (e.g., "Brief when executing; detailed when learning.") - Maximum 600 words total. - End with a single line: "Last updated: [today's date]" INTERVIEW ANSWERS: [paste your answers here] ---END--- ``` --- ## Installing Your Custom Instructions Claude will return your Custom Instructions in the format above. To install: 1. Go to **claude.ai** → click your profile icon (top right) → **Settings** 2. Find **Profile** → **Personal preferences** (sometimes labeled "Instructions for Claude") 3. Paste the Custom Instructions there → **Save** Every Claude chat from now on will use these instructions as the starting context — no need to re-explain who you are or how you work. --- ## Updating Over Time Re-run this questionnaire any time: - Your role significantly shifts - You realize the AI is consistently missing something about how you work - 6 months have passed and your preferences have evolved You can also hand-edit the Custom Instructions directly in claude.ai whenever you notice something to refine. The questionnaire gets you ~80% of the way; your own tuning gets you the rest. --- ## A Few Notes - **No one reads your answers.** They go from your draft → into Claude → out as your Custom Instructions. Your answers don't get stored or reviewed by HR, your manager, or Carlos. - **Be honest about red lines.** If you don't trust AI with personal context or a specific kind of task, say so. The instructions will tell Claude to respect that. - **It's OK to be ambitious in Q10.** "What would you want it to grow into?" is genuinely open. If the answer is "I want it to draft my whole monthly board memo from raw data," write that down. We can build toward it. --- ## Questions? Ask Carlos.