The Problem You Already Know
You open ChatGPT. You type: "Write a listing description for this 3-bed colonial." It spits out something generic. You rewrite half of it. You paste it into the MLS. Tomorrow you do it again — and ChatGPT has forgotten your voice, your market, and your brand.
Your designations. Your neighborhood expertise. The way you describe homes to buyers. Gone.
That's not an AI problem. That's a systems problem. You're using a tool that forgets. We're going to build one that remembers.
What You're Building
In the next 20 minutes, you'll build a listing description system. Not a prompt. A system — one that:
- Remembers your brokerage, your market area, your designations, and your voice
- Writes descriptions that sound like you, not generic AI
- Catches mistakes before they go live (wrong school district, claims you can't make)
- Works every time without re-explaining anything
By the end, you'll paste in property details and get back a description that sounds like you wrote it — because the system knows how you write.
Step 1: Give It Memory
The biggest frustration with ChatGPT? Starting from zero every conversation.
Here's the fix. Create a file called agent-profile.md with your details:
# Agent Profile
**Name:** [Your Name]
**Brokerage:** [e.g., Keller Williams Realty]
**Designations:** [e.g., CRS, ABR, SRS]
**Market Area:** [e.g., Baltimore metro — Canton, Fells Point, Federal Hill, Locust Point]
**Specialty:** [e.g., First-time buyers, historic rowhouses, waterfront condos]
## Voice
- Warm but authoritative — like a knowledgeable neighbor
- Lead with lifestyle, not square footage
- Neighborhood context in every description
- Never use "boasts," "nestled," or "won't last long"
- Keep it real — if it's a fixer, say "ready for your vision"
## Compliance Notes
- Never guarantee appreciation or investment returns
- Always include fair housing language where required
- Don't reference school ratings by number — say "highly rated schools nearby"
- Square footage from tax records, not measurements
This is memory. Instead of re-typing your credentials and style every time, the system reads this file and already knows who you are.
Step 2: Write the Instruction
Now create the system prompt — the instruction that tells the AI what to do with every listing:
You are a listing description writer for {{agent_name}}, {{brokerage}}.
When given property details, write a description that:
1. Opens with the lifestyle hook — what it FEELS like to live there
2. References the specific neighborhood and its character
3. Highlights 3-4 standout features naturally (no bullet lists)
4. Matches the voice profile — warm, authoritative, never generic
5. Stays under 250 words for MLS, 150 for social
6. Never claims "won't last long" or "best deal" — let the property speak
7. Includes a natural call to action
Read the agent profile before every description.
Notice rule #6. That's control — a constraint that prevents the AI from writing the kind of hype that makes agents look desperate. Without it, the AI defaults to every cliche in the book.
Step 3: Test It
Paste these property details into your system:
3BR/2BA colonial in Canton, Baltimore. 1,850 sq ft. Updated kitchen with quartz counters. Original hardwood floors. Finished basement. Walk to restaurants and waterfront. $385,000.
A bad description (generic AI, no system):
Welcome to this stunning 3-bedroom, 2-bathroom colonial in the heart of Canton! This beautiful home boasts 1,850 square feet of living space with an updated kitchen featuring gorgeous quartz countertops. Don't miss this amazing opportunity — it won't last long!
Every agent's AI writes that. Your buyers' eyes glaze over.
A good description (your system, with memory + instruction):
Sunday mornings in Canton start with coffee on the front steps and a walk to the waterfront. This 1,850-square-foot colonial puts you three blocks from the restaurants on O'Donnell Square and ten minutes from the harbor. The kitchen got the full treatment — quartz counters, modern fixtures — but the original hardwoods throughout the main level remind you this neighborhood has history. Finished basement gives you the flex space every Canton rowhouse owner wishes they had. $385,000.
The difference? The system knew your market. It knew Canton. It knew to lead with lifestyle. It didn't "boast" or claim anything would sell fast — it let the neighborhood do the talking.
What Just Happened
You built a system with three of the four components every AI system needs:
| Component | What It Does | What You Built |
|---|---|---|
| Memory | Persists your agent context | agent-profile.md |
| Instruction | Tells the AI what to do | System prompt with rules |
| Control | Prevents mistakes | "Never claim won't last long" rule |
| Flow | Multi-step automation | (Chapter 7 of the book) |
This same pattern works for market reports, follow-up emails, social posts, and open house recaps. Different instructions, same architecture.
Where This Goes Next
What you built works for one task. The book gives you two working systems — listing descriptions and social posts — then teaches you the framework to design your own (Chapter 13). Market reports, follow-ups, open house recaps — same architecture, your market expertise.
Chapter 7 adds flow — where one system's output feeds another. That's where "20 minutes of setup" turns into "hours saved every week."
The agents who figure this out first aren't just saving time. They're building a competitive advantage that compounds — because their systems get better every month while competitors are still copy-pasting from ChatGPT.
82% of agents use AI. Only 17% report significant impact. The difference? Systems, not prompts.
— NAR/RPR 2026 Technology Report