How a New York landlord can screen maintenance requests with AI before dispatching a vendor
A repeatable workflow for sorting tenant requests into urgency tiers, drafting vendor instructions from the tenant's own words, and keeping a record of every decision, with the emergency calls left to a human.


- AI sorts and drafts the request; the human still decides dispatch
- Anything threatening safety or causing active damage skips AI and goes straight to a person
- The tenant's own words become the vendor instructions, so nothing gets lost in a rewrite
- Save every prompt, tier and decision so you can show what you did and when
It is Friday night and a tenant sends a photo of a brown stain spreading across the bathroom ceiling. The property manager reads it once, then again, and starts the mental math: pay for an emergency plumber tonight, or risk a bigger repair and an angry tenant by waiting until Monday. The message itself gives the answer, but only if someone has time to pull the details out of it. That is the bottleneck in most small portfolios, not the lack of a vendor.
AI can sort, categorize and document that request in about a minute, so the human decides dispatch with the facts in hand. The work happens before anyone calls a truck, and it takes a general assistant such as ChatGPT or Claude plus a repeatable prompt. One subscription is usually enough to run this, and the cost is whatever the provider charges on its own pricing page, visible before you pay.
What the workflow actually does
The system has three parts. First, a triage prompt reads the tenant's message and assigns an urgency tier based on the words the tenant used. Second, a drafting prompt turns that same message into instructions a vendor can act on, with the unit number, the symptom and any photos described. Third, a records habit files the prompt, the tier and the dispatch decision in one place, so a month later you can answer why you treated a leak one way and a dripping faucet another.
The tiers are yours to define, but a workable starting set is four: emergency, urgent, routine and low priority. An emergency threatens safety or is actively damaging the building, such as a gas smell, flooding or no heat in winter. Urgent means damage is likely within days, like the ceiling stain that grows overnight. Routine covers working-but-failing fixtures, and low priority covers cosmetic issues. The important rule is that AI never assigns the emergency tier by itself; it only flags candidates for a human to confirm.
Who this is for, and who should skip it
This fits a landlord or property manager handling dozens of tenant messages a week, especially one wearing several hats without a dispatch desk. It also fits a small team where one person triages and another dispatches, because the written tier and notes keep both on the same page. It is not for someone hoping to remove humans from the loop. Every dispatch decision, and every emergency, stays with a person. It is also not worth building if you handle one or two requests a month; the overhead of writing prompts would exceed the time saved.
Set it up
- 01Define your four tiers
Write one line each for emergency, urgent, routine and low priority, with New York-specific examples such as no heat in winter.
- 02Write the triage prompt
Tell the assistant to read the tenant message, pick one tier, quote the exact words that justify it and list any missing details.
- 03Paste the message, never the tenant's identity
Use the text of the request and the unit number; leave out names, rent history and anything not needed for the repair.
- 04Handle the emergency check first
Before using any output, ask whether the message involves safety or active damage; if yes, stop and call the vendor yourself.
- 05Run the drafting prompt
Ask for vendor instructions built from the tenant's words: location, symptom, when it started and what the tenant already tried.
- 06Save the record
File the original message, the tier, the vendor instructions and your dispatch decision in one folder or doc per request.
The emergency exception stays human
Anything that threatens a person's safety or is actively damaging the building goes straight to a human and a vendor call, no AI step in between. Gas smells, electrical hazards, flooding, sewage backups and loss of heat in cold weather all belong in that lane. The reason is simple: an AI can misread a vague tenant message, and the cost of a wrong emergency call is measured in safety and damage, not in minutes saved. The workflow earns its keep on the routine middle of the pile, where the risk of delay is small and the volume is large.
The records habit
Keep four things for every request: the tenant's original message, the tier the prompt assigned, the vendor instructions as sent, and your dispatch decision with a date. This takes one folder and a naming convention, such as the unit number and the date. The payoff appears when a tenant disputes response time, when a vendor bills against vague instructions, or when you want to see which units and which systems generate the most requests. The record also protects the tenant's information, because it keeps only what the repair needed.
Prompt templates to start with
- Triage prompt You are helping a property manager sort maintenance requests. Read the tenant message below and assign one tier: emergency, urgent, routine or low priority. Quote the exact words that justify the tier. List any detail a vendor would need that the message leaves out. Do not recommend a vendor or decide dispatch. Tenant message: [paste]
- Vendor instructions prompt Using the tenant message below, write instructions for a vendor. Include the unit, the location of the problem, the symptom, when it started, anything the tenant tried and any photo described. Use the tenant's own words for the symptom. Do not add guesses about the cause. Tenant message: [paste]
Both templates share one rule: the assistant reports and drafts, it does not decide. If you find an output quietly recommending a course of action, tighten the prompt and rerun. The value of the workflow is that it puts the facts in front of you quickly, and the judgment stays where it belongs.




