Claude AI for Real Estate Agents: The Roadmap for What to Build First

Claude can already write my content, follow up with my leads, read my CRM, research properties, build CMAs and keep track of my transactions. That’s the problem.

When a tool can do fifty things, most agents build the thing that’s easiest to build. Then they spend a month on it, and their closings don’t move, because the thing they automated was never what their business was short of.

I’m William Zhang, a full-time agent here in Austin and the founder of Real Estate AI Society. This is the roadmap I use, and the one I give other agents, for deciding what to build first with Claude AI for real estate agents. It comes down to one question I ask before I build any automation, skill or agent:

What is limiting my business right now, and can Claude remove that constraint?

That question has saved me more wasted build time than anything else I’ve learned. Below is how I answer it, and the specific Claude workflows I run at each stage of my own business.

What to Automate First: The Three Stages Every Real Estate Business Gets Stuck In

When I look at my own business, almost every limitation falls into one of three stages.

Stage 1: Lead generation. Are enough new opportunities coming in? Not followers, not views, not likes on Instagram. Are people actually reaching out to you about buying, selling, investing or relocating in your area?

Stage 2: Conversion. Are those leads turning into appointments? Listing appointments, phone calls about selling, weekend showings. Leads don’t matter if they never get to an appointment, because you can’t get to a closing without one.

Stage 3: Client service. Can you serve the clients you already have without drowning? Constant CMA requests, rental comps, tour planning, weekly market updates, tracking transactions, answering the same questions over and over.

Your constraint is one of those three, and the honest answer decides what you build. If I don’t have enough leads and I’m not setting appointments, there’s no point in using Claude to automate a CMA report. It won’t get me a single extra deal. I’d focus everything on generating pipeline that’s going to buy or sell in the next two to three months.

A pattern I see a lot: most individual agents are stuck at stage one or two. Team leads and brokerage owners are usually stuck at stage three, because they’ve got a whole team running through the same back office. And plenty of agents who are sure they have a lead problem actually have a conversion problem.

Stage 1: AI Lead Generation for Realtors With Claude

There are two ways to generate leads. You trade time for attention (organic), or you trade money for attention (paid). Claude helps with both, but differently. I covered a few more lead sources, like cold-emailing relocating tech buyers, in 5 ways to get real estate leads with Claude AI.

Organic: YouTube

I’m a big advocate for YouTube as your main lead source, or at least one of them. I’ve done over $10 million in transactions from YouTube alone, and it’s where I’ve spent the most time building Claude agents.

A Claude agent, if you’re new to this, is an autonomous workflow that does a job from start to finish, 80% of the work if not all of it. On the YouTube side I run four:

• Topic research. My agent finds what I should be making, like “10 reasons to move to Austin” or the pros and cons of living here. That research used to take me two to three hours per video. Now it runs in the background.
• Outline and script. I film from an outline now, not a teleprompter, but I started on a teleprompter and that’s a fine place to start. I’ve also run an AI clone on one of my channels, and it’s closed three deals. As long as you disclose it clearly, it doesn’t hurt trust. If that’s interesting to you, I wrote up how I built my AI clone.
• Repurposing. An AI editor finds clips in my long-form videos and turns them into Shorts, working alongside my human editor.
• Upload, SEO and scheduling. I publish two to three long-form videos and 10 to 15 Shorts a week. Writing descriptions, SEO, thumbnails and scheduling used to eat two to three hours. Now I run one command and walk away.

Organic: social media repurposing

Same idea, more platforms. The agent pulls the long-form video, turns it into short clips, carousels and image posts, and schedules them to Facebook and Instagram. I’m not on TikTok or LinkedIn yet. I’m pretty happy where I am, and I’ll probably add them next year.

Organic: getting cited by ChatGPT and Google AI

This is the one almost nobody is talking about, and I think that’s because it isn’t competitive yet.

A few years ago, you searched “how do I negotiate on a new construction home in Austin,” got a page of links, and hoped your blog post ranked near the top. Now a lot of buyers just ask ChatGPT, Gemini or Google’s AI Mode, and they take the answer at the top of the page instead of reading five articles.

So I built a website for my niche and my market, and I use a Claude SEO agent to write long-tail posts for it. That’s gotten me recommended by AI. A Google AI answer that cited my site sent me a buyer who was under contract four days later. Actually, two deals came in that way this year. One didn’t work out, the other did.

Long-tail content means the specific questions nobody in your market has answered well. If you’re in Kansas City, I’d bet nobody has written a good post on buying a retirement home there, or what to watch out for when you upsize, or which schools matter for which neighborhoods. That’s open ground. I went deeper on this in AI SEO for real estate agents.

One honest caveat on everything organic: it takes time. That’s why I call it trading time for attention. But once the system is running (one or two videos, three or four blog posts, and daily social each week) it snowballs. I have videos from two and three years ago still sending me leads today, on a channel I haven’t posted to in a long time.

Paid: webinars, ads and portal leads

Webinars and events. A webinar is a one-to-many way to sell: gather everyone interested in “how to buy your first home in Austin” or “how to stop renting,” and pitch your service to all of them at once. In Austin it’s competitive, so plan on $500 to $1,000 a month in ad spend. My advice is to fund it with money from your organic channels once they’re producing. Claude helps me with topics and angles (doing okay, not great), builds my presentations, and writes the invites and reminders. I’m also testing a caller agent that phones registrants to remind them, built on the same idea as my AI voice agent for real estate leads. I’ll see if it lifts show-up.

Meta and Google ads. I haven’t gotten a great return here yet. Those leads tended to be six to twelve months out, and I didn’t have the follow-up or targeting skill to find the two-to-three-month buyers. That’s what I’m working on for 2027. Where AI already helps a lot is the offer and the creative. A flat-fee listing offer did well for me on Google. And Meta’s Andromeda algorithm wants a constant flow of new ad variations, which is exactly what an AI-assisted workflow can produce: a human approves the launch, an agent monitors performance and generates new creatives. I wrote up my honest results from ChatGPT ads if you want the numbers.

Portal leads. I used to be a big hater of portal leads. Now they’re some of my higher-intent leads. A reasonable return is around one to three, maybe one to five: spend $1,000, ideally get $3,000 to $5,000 back. But the payback usually takes three to six months, sometimes a year. I just closed a seller lead that came in about a year and a half ago. Claude plugs in through Follow Up Boss: it reads each new lead, understands the context, drafts a first text, and hands off to a drip if we miss them.

Stage 2: Converting Leads Into Appointments With Claude

If you’re getting five or ten leads a day and they’re not turning into appointments, this is your stage.

Definitions matter here. I used to think of conversion as lead to closing. You can’t convert a lead straight into a closing. A lead converts into a listing appointment or a showing, and the closing comes after.

Speed to lead

The highest-leverage move in this whole stage is responding fast. A phone call in the first few minutes, 10 or 15 minutes of qualifying, is still the best return there is. The next best thing is a quick text or email that finds out what they’re looking for. If you want the phone call without being on the phone, I built an AI that calls new leads instantly.

Because Claude is connected to my Follow Up Boss, a new lead gets a first response drafted off their real context and starts a drip. The part I haven’t built yet, and it’s very buildable, is the next step: an intake that captures what they want, runs the search in the MLS or IDX, and sends them homes that fit before a human takes over and sets the appointment.

Follow-up that watches behavior

The fortune is in the follow-up, and this is where Claude has helped me most. Every morning it reads my CRM and tells me who to call based on triggers I gave it:

• Did they come back and look at homes after going quiet?
• Did they look at several homes on the same day?
• Did they share a home with someone else?

It also knows where each lead came from, so the sequence changes with the source. A lead from my downsizing Meta ad gets a downsizing sequence. A lead from a relocation YouTube video gets relocation material: school guides, things to do, commute info. (If you don’t have those yet, here’s how I build real estate lead magnets with AI.) That means my prospecting hours go to the people ready to move in one to three months, which is also how I avoid burning out. I broke down the full setup in how I use Claude to manage Follow Up Boss.

It’s boring work. The boring work is what gets you paid.

If you want a simple starting point for any of this, the Claude for Real Estate Agents Starter Guide is free.

Stage 3: Client Service Workflows

I don’t think most agents are stuck here, and if you are, it’s a good problem to have. But these workflows are where Claude has made the job more enjoyable for me.

CMAs and comps. I connected Claude to my MLS through a licensed feed, and it pulls the data, filters the comps and builds a clean comp report. I used to avoid investor clients. They’re often in the $200,000 to $300,000 range and want sales comps, rental comps and after-repair values on 20 or 30 properties. Now the agent does that research and I don’t mind at all. On the listing side, the same data powers the AI seller assistant I use for listing presentations. I’m still the one making the pricing judgment and presenting it. If you haven’t connected yet, read the three ways to connect Claude to your MLS first, because one of them got me a warning letter.

Showings and tour routes. I’m mostly a buyer’s agent, and the part I liked least was tours. A client sends ten screenshots, I narrow it to five or six, dig up the showing instructions, contact listing agents on the occupied ones, and figure out a route in Google Maps. Now Claude reads the buyer’s criteria, organizes the homes, pulls showing instructions and builds the route. For a client shopping in the $5 to $10 million range, it built a single-property one-pager for every home we saw that day. No Canva.

Weekly seller market updates. For each listing, my agent pulls what changed in the neighborhood: new listings, price cuts, pending, sold, withdrawn. Then it drafts a personal update for the seller. If you have 10 or 20 listings and do this by hand every Friday, that’s a lot of hours. I walked through real examples in the six jobs Claude does with my MLS data.

Transactions. I wouldn’t call what I have an AI transaction coordinator. It’s an agent that reads my contract documents and emails and puts the dates on my calendar: closing, the blue tape walkthrough, the final walkthrough, the end of the option period here in Texas. That alone keeps me from dropping a follow-up with the lender, the title company or my back office.

Repeat client questions. Relocation clients almost always ask the same thing: which school districts are close to my employer at my price point? I use Claude to capture those questions and turn them into blog posts and FAQ assets, which then feed back into stage one.

Find the Constraint, Build One System, Then Move

That’s the whole roadmap. It works whether you build on Claude, ChatGPT or something else, because the tool isn’t the point. The question is.

Are you short on leads, short on appointments, or buried in client work? Build one system for that stage. Get it running. Only then move to the next one.

If you’d rather build your first assistant with me live, join the next free workshop. And if you want the whole picture of how these pieces connect in one practice, here’s my full Claude system from lead gen to closing.