What’s a Claude Project, and why it’s different#

You know ChatGPT. You type your question, it answers. It’s a stateless conversation: the system forgets what it knows about you in the next conversation.

A Claude Project is the opposite. It’s a permanent assistant configured with:

• Your permanent instructions (brokerage rules, communication tone, priorities)
• Your document base (brokerage book, templates, market data, listings)
• Your business context (property types managed, geographic area, client targets)

Once configured, every time you ask a question, Claude knows your context and gives perfectly personalized answers.

Before / after, on the same request#

• Without Project: “Write a listing for a 2-bed in NYC.” → Claude generates a generic listing.
• With Project: “Write a listing for this 2-bed” (you upload the listing sheet) → Claude generates a listing formatted to your brokerage standards, optimized for your area, with the strengths you identified.

It’s automation + personalization. That’s why 97% of US agents now use AI, and why those who don’t use it seriously are falling behind.

What it changes for a brokerage#

• 30-40% time saved on writing and document analysis
• Maximum consistency: your listings all look alike (good for the brand)
• Zero invention: AI only fabricates what’s in your documents
• Scalable: update your knowledge base once, everyone benefits

For generative AI applied to real estate overall (strengths, limits, Claude/ChatGPT/Gemini comparisons), we’ve published the complete cluster guide.

Why Claude, not ChatGPT#

Two concrete reasons for the Project format.

1. Claude handles long documents better#

A real estate brokerage project is often 20-50 documents (bylaws, listings, listing examples, client data). Claude handles this volume without quality degradation. ChatGPT slows down and loses precision when the context gets heavy.

2. Claude is more stable over time#

ChatGPT changes its models every few weeks, which can desync your workflows. Claude maintains longer compatibility, so your Project doesn’t break overnight.

Step 1: create your Claude Project#

Prerequisites#

• Claude Pro account (~$20/month) or free account with credits
• A browser (Claude runs in the web, no app needed)

Create the Project#

1. Allez sur claude.ai
2. Top left, open the “Projects” menu
3. Cliquez sur “+ Nouveau Projet”
4. Give it a name, e.g.: “[Your name] Brokerage”
5. Confirm

You’re now in a dedicated workspace where you can add documents, configure permanent instructions, and chat with Claude knowing it retains context.

Step 2: write your permanent instructions#

Photo: Lukas · Pexels

This is the key element of the Project. Permanent instructions tell Claude how to work for your brokerage. Copy-paste the template below into your Project’s “Instructions” section and adapt to your brokerage’s data.

YOU ARE AN AI ASSISTANT FOR A REAL ESTATE BROKERAGE

Brokerage context:
- Name: [Your brokerage]
- Geographic area: [Main market: Brooklyn Heights, Boston Back Bay, etc.]
- Property types: [Types managed: 1-3BR urban, luxury homes, investment, etc.]
- Brokerage size: [Number of agents, average revenue]
- Typical clients: [First-time buyers, investors, families]

COMMUNICATION RULES
- Tone: [professional, warm, direct, accessible]
- No useless jargon
- No invented numbers: verify against the listing sheet
- Every listing starts with the property's main strength

WRITING STANDARDS
- Listings: 150-200 words
- Client emails: short, 3-4 lines max
- No clichés: "bright", "spacious" only if true
- Clear call-to-action in every text

PRICE REFERENCE
- Area: [Average price per square foot: $700-$900]
- HOA dues range: [$300-$600/month typical]
- Investor target gross yield: [4-6%]

PRIORITY CLIENT DATA
- First-time buyers: transit access, parking, quiet
- Investors: net yield > 4%, university zone
- Families: nearby schools, parks, kids' space

If you need to write a listing: use the template below.
If you need to analyze a document: synthesize, highlight critical points.
If you need to send a client email: be concise, one CTA only.

This instruction set takes 5-10 minutes to write. It completely transforms response quality, because Claude stops guessing your context and applies it systematically.

Step 3: upload your reference documents#

This is your Project’s knowledge base. Six document families to load as priority.

1. Brokerage book (1 document)#

• Brokerage and agent intro
• Office photos
• Brokerage values
• Areas covered

2. Listing templates (1 grouped document)#

• Successful 1-bed urban listing example
• Successful family home example
• Successful investment studio example
• Successful loft / unusual property example

Put in 3-5 “model” listings you like. Claude will draw from them.

3. Market data (1 document)#

• Average price per square foot by area (neighborhood by neighborhood)
• Estimated gross yields
• Local market trends
• Seasonal calendar (peak season, off-season)

4. Sample listing sheets (1 or more documents)#

• At minimum 2-3 varied listings (urban 1-bed, family home, investment studio)
• For each, the complete listing sheet with all features

5. Internal communication guide (1 document)#

• Privacy rules
• Approval process before sending to client
• Who signs emails (broker, lead agent)

6. MLS / public records data (optional but powerful)#

• Recent transactions in your area (sale prices, conditions)
• Claude can use this to estimate a comparable property’s price

How to upload#

1. In your Claude Project, open the “Documents” section
2. Click “Add a document”
3. Drag and drop your file (PDF, Word, .txt accepted)
4. Claude indexes automatically

Each document must be under 25 MB. If larger, split it.

Once your Project is fed, you stop repeating context. You ask, Claude does.

Four use cases once configured#

1. Write a listing that respects your standards#

Here is the complete listing sheet for this 2-bed.
Write a listing according to my brokerage standards.

[YOU COPY-PASTE THE COMPLETE LISTING SHEET]

Claude produces a listing formatted to your standards (150-200 words), adapted to your market (local prices, typical average size), inspired by your best models, with no invented numbers.

2. Analyze HOA minutes with your area’s context#

Analyze these HOA minutes. Flag dues that are abnormal
compared to the area, and work that would impact value in my market.

[YOU UPLOAD THE MINUTES PDF]

Claude answers with context: “$450/month HOA dues are at the high end for Park Slope, range $350-$500”, “Façade restoration is roughly +8% value in your area”, “An unmaintained elevator is -5% value immediately.” For a complete method guide, we’ve published how to analyze your HOA documents in 2 minutes with Claude.

3. Prep a seller appointment#

I have a seller appointment tomorrow for this 3-bed family home.
Give me 5 key selling arguments and 3 likely seller objections
with their responses. Use my comp database.

[LISTING SHEET]

Claude produces 5 arguments based on your real market data, anticipates recurring objections you’ve documented, and gets you to the appointment fully prepared.

4. Generate a personalized client email#

Send an email to the couple who asked for more info
on the 2-bed in Park Slope. Be welcoming, propose an appointment, use my tone.

Claude produces a short email, signed per your process, written like you would. Not a generic blob you can spot from a mile away.

Sonnet or Opus, which model to pick#

Verdict: start with Sonnet. You get 99% of the quality at ~20% of the cost. Switch to Opus when your Project is adopted by the whole team.

Complete setup, 30 minutes flat#

• 10-15 minutes: write the permanent instructions
• 10 minutes: select and group the documents
• 5 minutes: upload the documents
• 2-3 minutes: test the first prompt

Thirty minutes for a Project that will save you dozens of hours in the weeks ahead. One of the best investment-to-return ratios in the whole brokerage AI stack.

To go even further#

Photo: cottonbro studio · Pexels

If you want to automate to the max, the next level is Claude Code: writing Python or JavaScript code that uses the Claude API to fully automate certain workflows (scrape MLS data, generate 50 listings in batch, analyze a hundred listings, monitor a market). But 95% of brokerages get what they need from well-configured Claude Projects alone.

The natural next step is packaging this Project into a shared tool for the whole team: that’s called building a custom AI assistant for the brokerage. Same building blocks, different deliverable: an assistant every agent can query about their market, listings, clients, without reconfiguring their own Project.