The 3 Levels of AI for Real Estate Agents, and How to Move Up One

Jerad Larkin

7 hours ago

10 min read

What are the three levels of AI for real estate agents? Level one is prompting, level two is projects, and level three is agents. Most agents sit on level one, and moving up one level usually takes a single afternoon of setup, not a new skill set.

Download the slide deck and the prompts

Everything I showed in this class is yours, including the write-like-me prompt, the listing description prompt, and the MLS market analysis prompt.

Full slide deck plus every prompt: https://bit.ly/whats-your-ai-level

Grab it now if that is what you came for. The written version of the class is below.

I teach this one live, and I open it the same way every time. Hands up if you are level one, meaning you have an account, it is probably the free one, and you barely use it. Usually about 60 percent of the room goes up.

That is not a knock on anybody. It is what happens when a tool gets dropped in your lap with no instructions. You open it, you ask it something, the answer sounds like a corporate press release, and you close the tab.

So I built this class as a buffet. I cover a lot of ground across three levels, and my only ask is that you take one thing and actually use it. Ignore the other two thirds. The worst outcome is walking out overwhelmed and doing nothing.

https://youtu.be/i9m28ewwDLA

What the three levels actually are

Level 1, Prompt. You have an account. You ask it things once in a while. You are probably on the free plan and scratching the surface.
Level 2, Project. You pay for it, you use it weekly, you are uploading files and getting real work out of it.
Level 3, Agent. AI stops answering and starts doing. You give it one instruction and it runs the steps.

Whatever level you are on, the goal is to get to the next one. Nobody jumps from one to three.

Level 1: your AI does not sound like a robot, it sounds like a stranger

The complaint I hear most is that everything AI writes sounds fake. I hear it about ChatGPT, Claude, Gemini, Grok, all of them.

Here is the thing. That is not an AI problem. It is a context problem.

You are handing it a blank page every single time you open a new chat. With nothing to go on, it defaults to a generic corporate voice, because that is the safest average of everything it has ever read. It is not writing badly. It is writing for nobody in particular, because you have not told it who you are.

The fix is to stop starting from scratch and build a permanent home that already knows who you are, what you do, and how you sound.

Where that home lives

Every major platform has a version of this:

ChatGPT calls it a project: https://chatgpt.com
Claude also calls it a project: https://claude.ai
Gemini calls it a gem: https://gemini.google.com

Same idea across all three. A folder that holds instructions and reference files, so every chat inside it starts with the AI already knowing your business.

The context files that teach it your voice

This is the part people skip, and it is the part that actually works. Feed it things you wrote or said, not things AI wrote:

Video transcripts. If you post video of yourself talking, this is the fastest way to teach it your rhythm and your phrasing. Export the transcript out of your editor and upload it.
Old emails you wrote yourself. Go back a year if you have to, before you started using AI to draft.
Listing descriptions you wrote. Same rule.
Social captions you wrote.
Call and Zoom recordings. The transcripts are gold.
Your bio and your About page.

If you have none of that, you are not stuck. Ask the AI to interview you. Tell it to ask you everything about who you are, your brand, who you serve, and how you talk, and let it build the voice profile from your answers.

Garbage in, garbage out. The less you feed it, the more generic it stays.

Building your first project, step by step

Open your AI and find the Projects section in the left sidebar. Click the plus button next to it.
Name it something obvious, like Sound Like Me.
Upload your context files into Sources or the knowledge area.
Open the three-dot menu and click Project settings or Instructions.
Paste in your instructions, telling it to source those files and match that voice.
Save, then start working inside that project instead of a blank chat.

One thing agents ask every single class: yes, you can drag an existing chat into a project after the fact. You do not have to start over.

Expect a couple of rounds of back and forth before it sounds right. That is normal. You are training it, not flipping a switch.

The other project worth building on day one

Listing descriptions. When I ask a room how they use AI, this is the first answer every time.

The version I built works like this: you drop in a property address, the AI goes out and searches for public record, prior listing history, HOA details, neighborhood and location data, then it separates what it can verify from what it needs you to confirm. You confirm the facts, and it writes the description.

The instructions also tell it to follow Fair Housing and your MLS advertising rules. That part is not optional. I had an agent in a recent class who got fined a few hundred dollars over a listing description because someone reported it. AI gets you most of the way there. You are still the licensed party, and you make the final call.

Your one step if you are level 1: build one project. Sound Like Me or listing descriptions, pick one. Then stop and ignore the rest of this post.

Level 2: you are paying for it and leaving most of it on the table

You use it weekly. You upload documents. You get real value. Here are three things agents at this level usually do not know about.

The intelligence setting nobody touches

Both ChatGPT and Claude let you choose how hard the model thinks before it answers. In ChatGPT it is a small slider under the chat box, moving from instant to medium to high. In Claude you pick the model at the top of the window.

Instant gives you a fast answer. Higher settings take longer and dig harder. For anything that matters, turn it up.

One honest caveat from a very sharp attendee in the room: higher intelligence burns through your usage faster. On a capped plan, that is a real tradeoff. A quick question does not need the smartest setting. A pricing analysis does.

Deep research

This is a specific mode, not just a longer prompt. It goes out, reads a stack of sources, and comes back with a cited report. It can take five to thirty minutes.

Things I have built with it for agents:

A fully branded neighborhood guide with parks, commute times, new development, and pricing history
A marketing strategy for a listing, including who the likely buyer is and where to reach them
A property deep dive from an address: permit history, past sales, zoning, flood
A competitive analysis of what other agents in your area are actually doing
Contact lists for referral relationships you want to build: divorce and family law attorneys, CPAs and financial planners, estate and probate attorneys, property managers, relocation contacts at large Denver employers, builders, HOA board members

On that last one, here is the trick. AI only works as hard as you tell it to. Do not ask for a list. Tell it you want at least 500 divorce and family law attorneys in the Denver metro, with phones, emails, and social accounts, compiled into a spreadsheet, and not to stop until it gets there. Then verify what comes back before you use it.

Your MLS is a gold mine

This is the one I would put money on being the highest-value thing in the class.

You have access to REcolorado, and that data can build your market report, your pricing conversation, your CMA, and your listing presentation on real numbers instead of feel.

Run your search in the MLS, select your results, and export. I use the full export format because it carries the most columns per property. Then drop that file into your AI and just talk to it. Subdivision, zip code, city, price band, whatever you need.

A couple of practical notes:

Your MLS plan caps how many records you can export per month. I pay extra for a much higher cap because the data is worth it to me.
If you are out of exports, a PDF export or a screenshot still works. AI can read both.
At level 3, the Claude Chrome extension can read the results on screen without downloading anything at all.

The reason this matters right now: listings are sitting. When a seller calls asking what is going on, the market is slow is not an answer. Pull the last 60 days of closed sales in that subdivision, have AI analyze it, and go into that conversation with numbers.

You know your neighborhoods better than any model does. Use AI to do the arithmetic and the formatting, not the judgment.

Your one step if you are level 2: export one MLS search and run one market analysis prompt on it.

Level 3: AI that does the work instead of describing it

Level three is where AI stops answering and starts doing. Call it AI 2.0 if you want. Version 1.0 hands you a result and you go act on it. Version 2.0 runs the steps.

One instruction, five steps. It runs the MLS analysis, builds the report, and drafts the email to the client. Or it opens your inbox, scans the senders, builds you a list, and unsubscribes you from the junk once you approve it.

The Claude Chrome extension

If you are not using this, it is the single biggest jump available to you right now.

Chrome has extensions. One of them is Claude: https://chromewebstore.google.com/detail/claude/fcoeoabgfenejglbffodgkkbkcdhcgfn

Install it, and anywhere you are on the internet you can open a side panel and hand off the screen. It interprets the request, shows you the plan, you approve it, and then an orange cursor takes over and does the work while you watch.

What agents in my classes actually use it for:

Running MLS searches and reading the results without burning an export
Building ShowingTime analysis reports
CRM notes and data entry
Clearing out promotional email
Finding and applying promo codes before checkout, which is the one that always gets a reaction

My own favorite use is the one nobody expects: have it do the things you do not know how to do. When you hit something technical on a website and you are stuck, open the panel and tell it to figure it out, then watch how it does it. I have also handed it workflows I have done manually for years and watched it find a shorter path.

It runs about twenty dollars a month. If it saves you an hour, it paid for itself twice.

Claude Cowork

This is the deeper end of the same idea. You can point it at a folder on your computer, connect it to your email and calendar, and have it work across all of it in one place: https://claude.com/download

Agents I work with are using it for inbox triage, reverse prospecting, drafting blog posts and email replies, sending texts through iMessage, and a morning briefing of the day ahead. You can also save repeatable instructions as skills so you are not re-explaining yourself every time.

Start small. Give it one folder and one job.

Your one step if you are level 3 or want to be: install the Claude Chrome extension tonight and ask it to clean up your inbox.

A fair question about what you feed it

Someone in the room asked how much personal information you should hand an AI. It is the right question and I told them the truth, which is that it is a judgment call and the rules are still moving.

Where I land: keep client personal information out of it. Social Security numbers, account numbers, anything you would not want sitting on a server you do not control. Business context and market data are a different category from a client's private details.

A related question came up about new Colorado rules on automated decision making. The distinction I would draw is between using AI to write a description or a social post, which is marketing, and using AI plus someone's personal information to decide who qualifies or who gets a lease, which is a different animal entirely and sits right next to Fair Housing. Confirm the specifics with your broker or your attorney rather than with me or with a chatbot, because this area is changing fast.

Where Chicago Title fits

As an Account Executive with Chicago Title of Colorado, I am in front of Denver metro agents all day, and this class came directly out of those conversations. I have sat down one on one with more than a thousand real estate professionals since 2017, and the pattern is always the same: everyone is at a different level and nobody wants to feel behind.

My job is to hand you the data, the marketing, and the systems that help you win listings, and then make the closing smooth when you get one. The classes are part of that, not separate from it.

FAQ

What is the difference between a prompt and a project?

A prompt is a set of instructions you paste into a chat. A project is a folder that stores those instructions plus your reference files, so every chat inside it starts with the AI already knowing your voice and your business. Prompts are one-offs. Projects are permanent.

Do real estate agents need the paid version of ChatGPT or Claude?

For level one, the free version is enough to learn on. Once you are using it weekly, the paid plan is worth it, mostly because it unlocks higher intelligence settings, larger file uploads, and the agentic features. That is usually around twenty dollars a month.

Can AI write my MLS listing descriptions?

It can draft them, and it can do a good job if you feed it verified property details and instructions that include Fair Housing and MLS advertising rules. You are still the licensed party responsible for accuracy and compliance, so review every word before it goes live.

Which AI is best for real estate, ChatGPT or Claude?

Both are strong, and the honest answer is that the one you set up properly will beat the one you did not. I use Claude because it moved early on agentic features, connecting to a computer and running tasks. Plenty of agents I work with are getting excellent results out of ChatGPT. Pick one and go deep rather than dabbling in four.

How do I use MLS data with AI?

Run your search in REcolorado, select your results, and export in the full export format. Upload that file to your AI and ask it questions about the subdivision, the price band, or the days on market. If you are out of exports, a PDF or a screenshot works too.

Take one thing

That is the whole ask. Prompt, project, agent. Pick the level you are on, take the one step for that level, and stay curious. Some of the best questions I got in this class were ones I did not know the answer to, so I asked AI in front of the room and we found out together.

If you have a listing sitting and you want an analysis run on it, or you are looking for marketing ideas to get it in front of the right people, reach out to me.

Want more tools, tactics, and resources like this? Subscribe to my weekly emails at https://www.milehightitleguy.com. I share real estate marketing ideas, AI tools, and exclusive invites to upcoming classes and events across Colorado.

Get the class materials: https://bit.ly/whats-your-ai-level