You have 400 names sitting in your database.
You have not talked to most of them in over a year.
You open the CRM, look at the list, feel the weight of it, close the tab, and go post something on Instagram instead.
Most AI prompts for real estate agents do not fix that, because they hand you a block of text you still have to think about.
You do not need more text.
You need a name, a phone number, and a reason to call.
Below is one real prompt that produces exactly that, torn apart line by line, so you can rebuild it for your business instead of copying mine and hoping it fits.
Why most AI prompts for real estate agents fail
Most AI prompts for real estate agents are written like a wish instead of an instruction.
Here is the prompt I see agents actually type.
“Write a follow up email to my past clients.”
It fails for four reasons.
It never says who the AI is supposed to be.
It gives the AI zero real data, so the AI invents an average client who does not exist.
It sets no rules for who matters more than who.
And it never says what the answer should look like, so you get prose when you needed a list.
A prompt fails when it asks AI to imagine your business instead of handing it your business.
The fix is not fancier wording, and it is not a 500-prompt swipe file you bought on Instagram.
The fix is real data, ranking rules in a set order, and a required output format.
The prompt
This runs in Claude, ChatGPT, or Grok.
You are my real estate business partner and you know my
database better than I do.
I am pasting an export from my CRM below. Read all of it
before you answer.
[paste your export here]
Rank the top 20 people I should call this week.
Apply these rules in this order:
1. Closed with me 4 or more years ago and has not moved since.
2. Has a life event in my notes (baby, marriage, job change,
divorce, retirement, inheritance).
3. Has referred me business before.
4. Last contact was more than 12 months ago.
Skip anyone whose notes say do not contact, long term renter,
or moved out of area.
Give me a table with these columns: name, phone, last contact
date, the one reason I am calling, and one opening line.
Write the opening line in my voice. I am direct and warm. I do
not do small talk theatre and I never say "just checking in."
Do not invent details. If a field is blank, write MISSING and
move on.The teardown, line by line
This is where most AI prompts for real estate agents fall down, so read the reasons and not just the text.
“You are my real estate business partner”
The role you assign changes the answer you get.
Call it an assistant and it waits for instructions.
Call it a partner and it makes a call and defends it, which is the whole point here.
If you run a team, swap in “You are my inside sales manager” and watch the tone sharpen.
The data goes in before the task
Paste the export first, then ask the question.
The model reads the whole prompt before it writes anything, and putting the data up top keeps it from answering off the top of its head and backfilling later.
It also forces you to pull the export, which is the step almost everyone skips.
If your export is a mess, run it anyway, because messy is workable and empty is not.
The ranking rules are numbered on purpose
“Apply these rules in this order” tells the model to use them as tiebreakers in sequence instead of blending them into a general feeling about each contact.
The 4-year rule is mine, not gospel, so if people in your market hold for 7 years, make it 5 and see what changes.
Rewrite this line first, because it carries more weight than everything else in the prompt combined.
The output has to be a table
Prose is useless at 7am.
Name, phone, last contact, one reason, one opening line gives you something you can work straight down while the coffee is still hot.
“Do not invent details”
This is the most important line in the whole thing.
Leave it out and the model will quietly fill a blank last-contact date with a plausible one.
Then you tell someone it has been about two years when it has been four months, and you look like you do not know your own business.
MISSING is an ugly thing to see in a table, and that is exactly why it works.
Run the math on your own database
The point of AI prompts for real estate agents is not the wording, it is the call list that comes out the other end.
Say you have 400 contacts and the prompt hands you 20 names.
You spread those 20 calls across a week and you connect with 8 of them, because people do not answer the phone anymore.
Of those 8, 2 tell you something you did not know, and one of them says the words “we have been talking about moving next spring.”
That is one live seller conversation for about three hours of work, from a database you already paid for and were not using.
Put your own average commission against that, then multiply it by 50 weeks.
The arithmetic is not the point, though.
The point is that building the list was always the bottleneck, and the list now takes 10 minutes.
When one of those calls turns into a real appointment, go prep it properly with the AI listing appointment prep workflow so you are not winging it on the drive over.
The privacy line you should not cross
One rule applies to all AI prompts for real estate agents, and it is about what you paste in.
Before you paste a client list into any AI tool, check two things: whether your brokerage has a written AI policy, and whether the tool trains on your data by default.
If you cannot answer both, strip the file down to first name, last initial, last contact date, and notes, then rejoin the phone numbers yourself in your own spreadsheet.
The ranking works fine without full contact details, because the model is sorting on dates and notes, not on phone numbers.
If your brokerage does not have a policy yet, NAR publishes AI policy templates and guidance for brokers, and handing that to your broker is a faster path than arguing about it.
Financial details, social insurance numbers, and anything from a mortgage file stay out of the prompt, permanently.
The platforms are racing to do this for you
On July 28, 2026, Lofty announced an expansion of its agentic AI system, including a workspace that ranks each day’s highest value tasks and leads for the agent instead of waiting to be asked.
Read that as confirmation, not as a reason to wait for your CRM to ship it.
Database mining is where the money in this industry is going right now, which tells you the workflow is worth learning.
Here is the part nobody tells you: when the automated version does land in your CRM, the only agents who will know whether its ranking is any good are the ones who already did it by hand.
Every tool has a place where it stops being useful, and you only find that edge by using it, which is the same lesson from where NotebookLM wins and fails for agents.
What to do today
- Export your database to a CSV with four columns: name, phone, last contact date, notes. If your CRM cannot do that in 10 minutes, that is a separate problem worth solving this month.
- Run the prompt with rule 1 rewritten for your market, then read the top 5 rows and ask whether you agree with the order. If you do not, change the rules and run it again. That second run is where you learn to write prompts.
- Block 90 minutes tomorrow morning and call the top 10. Not the top 20. A finished list of 10 builds the habit and an abandoned list of 20 kills it.
The prompt is not the point
Good AI prompts for real estate agents are just a faster way to decide who to call today.
The bigger win is seeing the shape of it: role, real data, ordered rules, fixed output format, and a hard instruction not to make things up.
That same shape works on your buyer follow-up and your Monday planning, and once you can see it, you stop shopping for prompts.
If you run it and the output is garbage, send it to me with the rules you used and I will tell you which line broke it.
Or book a call at upcoaching.ca and we will build the ranking rules around your actual database instead of a generic one.