Custom GPT: A Multi-Property Maintenance Triage Assistant

What This Builds

A tenant text comes in at 9pm: something's wrong with the water heater at one of your units, but you own eight properties and you're trying to remember which vendor you used there last time and whether that unit even has a tankless system. This build is a Custom GPT that already knows every property's basics and your vendor list, so you paste in the tenant's message once and get back an urgency read, the right vendor to call, and a draft reply to the tenant, without digging through old texts first.

Prerequisites

• A Plus subscription ($20/month): chatgpt.com
• A list of your properties with age, known recurring issues, and system type (water heater, HVAC) for each
• Your vendor list with trade, name, and contact info
• Your own triage rules for what counts as an emergency versus something that can wait

The Concept

A Custom GPT is a version of ChatGPT you configure once with instructions and reference files. After that, you reuse it forever. It's less like starting a new conversation each time and more like keeping a dispatcher on staff who already has your property file and your vendor rolodex open. You paste in a tenant's maintenance message. The dispatcher reads it against what it knows about that specific property and hands back a plan, instead of you explaining your whole portfolio from scratch every time something breaks.

Build It Step by Step

Part 1: Open the GPT Builder

In ChatGPT, open Explore GPTs from the sidebar (or go to chatgpt.com/gpts) and select Create. You'll land on two tabs: Create, where you describe the GPT in plain language and let ChatGPT draft a first pass, and Configure, where you set each field directly. Use Create for a rough first draft. Then move to Configure to get the details right.

Part 2: Configure the Basics

In the Configure tab, fill in:

Name: "Maintenance Triage: [Your Portfolio Name]"

Description: "Triages tenant maintenance requests across my rental properties and drafts a vendor dispatch and tenant reply."

Instructions: paste something close to this:

You triage maintenance requests for a small residential rental portfolio.
You have property details and a vendor list in your knowledge files.

When I paste a tenant's maintenance message and tell you the property, do
the following:
1. Identify the likely issue and trade needed (plumbing, electrical, HVAC,
   appliance, general, pest).
2. Rate urgency: EMERGENCY (active leak, no heat in cold weather, no power,
   gas smell, security issue), URGENT (needs attention within a couple of
   days), or ROUTINE (can be scheduled normally).
3. Look up the property in your knowledge files and check for a known
   recurring issue or system detail that's relevant.
4. Pick the right vendor from the vendor list based on trade, and give me
   their contact info.
5. Draft a short, professional reply to the tenant confirming receipt and
   setting an expectation for when help will arrive.

Never guess a property's system type or vendor if it isn't in your
knowledge files. Say so and ask me instead. If the message describes
something that sounds like an emergency but doesn't clearly match the
EMERGENCY criteria above, ask a clarifying question rather than guessing.

Conversation starters: add a couple of examples, like "Triage this maintenance text for [property]" and "What's my plumber's contact info for [property]?"

Part 3: Add Your Knowledge Files

Still in Configure, scroll to Knowledge and upload two files:

Property Reference File

Property: [nickname or address]
Age: [year built or age]
Known recurring issues: [e.g. "upstairs unit AC has needed two recharges in
two years"]
Water heater type: [tank / tankless / age if known]
HVAC type and age: [system type, age]
Anything unusual about access (shared driveway, gate code, etc.): [notes]

Repeat for every property.

Vendor List File

Trade: Plumbing
Vendor: [name]
Contact: [phone / email]
Notes: [e.g. "used them at three properties, responsive, no emergency
after-hours line"]

Repeat for each trade you use.

Caution: keep this Custom GPT's knowledge files limited to property and vendor details. Do not upload a running log of actual tenant messages, since those often carry a tenant's name, phone number, or unit access details. When you paste in a live maintenance request to triage it, strip the tenant's name and phone number from the pasted text first and refer to them by unit number only. Check ChatGPT's data controls for how information shared with a Custom GPT is retained before you rely on it for anything beyond this session-by-session triage.

Part 4: Test and Refine

Under Capabilities, leave web search and data analysis on if you want, though this GPT doesn't need them. Skip custom actions for now.

Test with a real (or realistic) message: "Triage this for [property]: tenant says the upstairs bathroom sink won't drain and there's a bad smell." Check that it identifies plumbing, gives a reasonable urgency rating, references your notes about that property if relevant, names the right vendor, and drafts a tenant reply. If it guesses instead of checking your knowledge files, tighten the instructions to require it to say when information isn't there.

When you're satisfied, set sharing to Only me unless you have a reason to share it.

Real Example: The Saturday Text

Setup: The Custom GPT is configured with eight properties and four vendor entries (plumbing, electrical, HVAC, general).

Input: "Triage this for the Elm Street duplex: tenant in unit B says water is pooling under the kitchen sink and it's gotten worse over the last hour."

Output: The GPT identifies plumbing, rates it EMERGENCY given the active worsening leak, notes from the property file that this unit had a slow drain issue flagged six months ago, names the plumber used at that property before with contact info, and drafts a tenant reply acknowledging the issue and giving a realistic window for the plumber to call.

Time saved: What used to mean scrolling through old texts to remember which plumber worked on that unit now takes the time it takes to paste one message and read the response.

What to Do When It Breaks

• The GPT guesses at a vendor instead of saying it doesn't know → Strengthen the instruction: "If no vendor is listed for a trade, say so explicitly rather than suggesting one." Re-test with a trade you know isn't in your vendor file.
• Urgency ratings feel off → Your EMERGENCY criteria in the instructions are probably too broad or too narrow for your comfort level. Adjust the specific triggers listed rather than the general category, and re-test with a few real past requests to calibrate.
• The GPT doesn't seem to reference your knowledge files at all → Open a fresh conversation with the GPT rather than continuing an old one. If it still ignores the files, check that they uploaded correctly under Configure, since a failed upload won't always show an obvious error.
• You stop checking it and it goes stale → A Custom GPT doesn't notify you when a property's details change. Set a habit of updating the Property Reference File whenever you replace a major system or switch vendors, the same way you'd update a paper vendor list.

Variations

• Simpler version: Skip the property reference file and just upload the vendor list. You lose the property-specific context but still get consistent triage and the right vendor contact.
• Extended version: Add a third knowledge file logging past triage outcomes (issue, vendor called, resolution, cost range described in general terms) so over time you can ask the GPT which properties generate the most maintenance calls and where a bigger capital repair might be overdue.

What to Do Next

• This week: Build the GPT with your three or four most active properties and test it against a few real past maintenance texts.
• This month: Fill in the rest of the portfolio and vendor list, and start using it as your first stop when a maintenance message comes in.
• Advanced: Pair this with the weekly portfolio digest build so a pattern of repeated maintenance calls at one property shows up in your regular numbers, not just in your memory of a string of annoying Saturdays.

Advanced guide for real estate investor professionals. These techniques use more sophisticated AI features that may require paid subscriptions.