Lesson objective

Triage a brokerage task on consequence, data class, verifiability, fair housing exposure and the licensee who signs off, and route it before any AI tool is opened.

Plain-language explanation

Why this matters now

Level 1 explained how AI tools produce output, how they fail, and which duties stay with the licensee and the supervising broker. Level 2 is about an ordinary week: a new listing to write up, a buyer who wants a neighbourhood overview, three showings to report back on, a seller asking what the last offer really means, and an AI assistant open in another window.

Plenty of agents already work this way. Research (not binding): in the National Association of REALTORS® 2025 REALTORS® Technology Survey, published September 18, 2025, 20 percent of respondents said they used AI in their business daily over the past 12 months, 22 percent weekly and 27 percent a few times a month, while 32 percent had not actively tried it (the categories sum to 101 percent through rounding). NAR invited a random sample of 49,233 active REALTORS® (NAR members) in July 2025 and received 1,241 usable responses, a response rate of 2.5 percent. The results describe the members who answered, not every licensee.

Using a tool often does not mean using it well. The first decision on any task is made before a tool is opened: should this task go near AI at all, and on what terms? Triage is how you answer that question the same way every time.

A question to sit with

Two requests arrive at 9 a.m. from the same seller: turn the verified property data sheet into a first draft of the listing remarks, and answer her email asking whether she has to tell buyers about the old water stain in the basement ceiling. Both could be pasted into the same tool in the same minute. What makes them different tasks?

What goes wrong

Choosing the tool before the task. The assistant is open, so the agent works out what to ask it. Nobody asks whether AI belongs in the task at all.

Triaging on convenience. "It will save twenty minutes" is a benefit, not a triage criterion. A disclosure answer that saves twenty minutes and tells a seller something wrong has cost far more than it saved.

Asking the data question after the paste. Once a buyer's pre-approval letter or a seller's divorce details are in a tool, the decision about whether they should have gone there has already been made, by default.

Assuming every output can be checked. Bedroom count, lot size and HOA dues can be checked in seconds against the property file. A statement about what a seller is legally required to disclose can only be checked by someone qualified to give that answer, and it is not an answer the agent gives.

Forgetting fair housing until the end. Copy about who a home "suits", a neighbourhood description or a decision about who sees an ad carries fair housing risk from the first draft. Triage flags it at the start so the review is planned, not improvised.

Losing track of who signs off. A draft passes from a transaction coordinator to an agent to a team leader, and each assumes someone else owns it. If you cannot name the licensee who will review it, and know which broker supervises that licensee, the task is not ready.

The practical answer

Ask five questions, in order, and write the answers down. The questions are an AI Coalition Network recommendation; the duties they draw on are cited beside them.

# Question What you are looking for Typical answers
1 Consequence: if the output is wrong and nobody catches it, where does it land? The furthest place the output will travel Internal note; client email; MLS listing or public ad; a document a party signs
2 Data class: what would the tool see? The most sensitive item in the material Public listing data; confidential client information; financial or identity documents
3 Verifiability: can every part be checked against a document you hold? How the check will be done, and what it costs Fully, against the property file; partly; only by redoing the work
4 Fair housing exposure: does the output describe people or places, answer a consumer, or choose an audience? Copy, images, neighbourhood answers, ad targeting None; copy review needed; audience decision stays with a person
5 Sign-off: which licensee reviews and owns it, under which broker? A named person, not a role The listing agent; the buyer's agent; the broker; an attorney, appraiser or lender where the question is theirs

Question 1: consequence. Follow the output to its final destination. Remarks drafted "just for the seller to look at" usually end up in the MLS. Rate the task by where an error could finally land.

Question 2: data class. Module 13 covers classification in depth; at triage you need only the headline. Professional guidance (California): the California Department of Real Estate, in its Winter 2026 Real Estate Bulletin article on AI, advises licensees not to put confidential or sensitive client information into public or unsecured AI platforms and to understand what data a tool collects, retains and shares. That article is guidance to California licensees; check what your own commission and your broker's policy say.

Question 3: verifiability. If you cannot say how the output will be checked, you cannot check it. Where checking costs as much as doing, AI adds the risk of an unchecked error without saving time.

Question 4: fair housing exposure. Fair housing law binds everyone in the transaction, whatever the tool. Professional guidance (California): the same DRE article says licensees who use AI must do so consistently with existing duties, including fair housing, fair lending and consumer protection laws. Module 05 teaches the copy review; at triage, you only decide whether that review is needed. Any task that would let a tool decide which consumers see a listing is Route D: that choice stays with a person.

Question 5: sign-off. Law / rule (South Carolina): S.C. Code §40-57-820, effective May 21, 2024, makes a South Carolina licensee responsible for "any and all work product produced by him or with the assistance of artificial intelligence, machine learning, or similar programs", and a violation committed through those programs "will be treated as if the violation was committed directly by the licensee." It binds South Carolina brokers, salespersons and property managers only. Law / rule (Texas): 22 Tex. Admin. Code §535.2 makes a Texas broker responsible for a sponsored sales agent's authorized acts and requires written notice of the scope of those acts; it does not mention AI. Professional guidance (California): DRE states that using AI tools to conduct licensed activity "may be equivalent to asking an unlicensed assistant to do licensed activity." Check your own commission's rules; in every state, the tool holds no licence.

Now combine the answers into a route.

Route When it applies What happens next
A: proceed, standard review Internal consequence; data cleared for the approved tool; fully verifiable; no fair housing exposure; licensee named Run the assisted-task workflow (Lesson 2); the licensee checks the checking
B: proceed, enhanced review Client-facing or public; data cleared; fully or largely verifiable; fair housing review planned; licensee named Run the workflow; the licensee checks every fact against the file and runs the fair housing review before release
C: redesign, then triage again Data not cleared; part of the output verifiable only by redoing it; or the task mixes agent work with someone else's work Remove data, split the task, or ask the broker; then triage again
D: no AI No licensee can be named; the output would be legal advice, an appraisal, a lending or rental decision, or an audience choice; or no approved tool exists Do it without AI, or route it to the attorney, appraiser, lender, housing provider or broker who owns it

Remember this

  • Triage happens before any tool is opened: consequence, data class, verifiability, fair housing exposure and sign-off.
  • Rate consequence by where an uncaught error could finally land, not where the draft goes first.
  • Legal advice, appraisals, lending and rental decisions, and choices about who sees a listing are Route D for the tool.
  • South Carolina's statute is explicit that AI-assisted work is the licensee's; elsewhere, check your commission, but the tool never holds the licence.
  • A tool's success on one task predicts nothing about a different task, so triage each one.

Real estate example

Larkfield Realty Group is a fictional twelve-agent brokerage in Texas. On a Monday, transaction coordinator Nadia Brooks, who is not licensed, has three requests from the team led by licensed sales agent Jonah Ortiz, and the brokerage's approved AI assistant is open. She triages first. Request one: turn the verified property data sheet for the fictional listing at 14 Wrenfield Court into draft MLS remarks. Consequence: public. Data: listing facts only, cleared. Verifiable: fully, against the property file. Fair housing: copy review needed. Sign-off: Jonah, under designated broker Marisol Lane. Route B. Request two: draft a reply to the buyers on another file, who asked whether the brokerage thinks the house is "worth" the $412,000 asking price. Consequence: client-facing, and it touches a value opinion. Nadia records Route C: the assistant may draft a short note saying Jonah will call to walk through the comparable sales, and Jonah will prepare that conversation himself. Request three: the same buyers ask the assistant, through a chat link on the team website, which nearby streets are "best for people like us". Route D: the question invites steering, no tool answers it, and Nadia flags it to Jonah, who will respond with objective, verifiable information about the homes themselves. Nadia writes each triage line into the transaction AI use record before she begins. Jonah initials the three routes within the hour, and because Texas requires the broker's written policies to cover advertising compliance, Marisol's policy tells Nadia exactly which tasks need her sign-off as well.

Responsible-use warning

Do not let an approved tool make triage feel unnecessary. Approval means your broker has cleared the tool for certain data and uses. It does not mean every task is suitable, that every output can be verified, that fair housing risk has been considered, or that a licensee owns the result. The costliest failures are a disclosure or legal question answered by a draft nobody qualified read, and a consumer question about "the right neighbourhood" answered by a chatbot. Triage the task, not the tool, and write the route down first.

Human-review requirement

The licensee who will sign off confirms the triage before the work starts, under the supervising broker's written policy. They check that the consequence has been followed to its final destination, that the data class is right and cleared for the tool, that a practical way to verify every fact exists, that a fair housing review is planned wherever copy, images, audiences or consumer answers are involved, and that their own name is on the task. They move legal, valuation, lending and rental questions to the professional who owns them. AI never assigns the route.

Practical exercise

Use fictional listings and clients only.

  1. Write down six tasks from a typical week in your role, such as listing remarks, a buyer neighbourhood overview, a showing feedback summary, an offer summary, a lead reply and a social media post. Replace every client and property detail with a fictional one.
  2. For each task, answer the consequence question by tracing the output to its final destination.
  3. Record the most sensitive item each task would expose to a tool.
  4. Describe in one sentence how each output would be verified, and mark it fully, partly or only by redoing the work.
  5. Mark the fair housing exposure of each task: none, copy review, or audience decision.
  6. Name the licensee who would sign off and the broker who supervises them, or the attorney, appraiser, lender or housing provider who owns the question.
  7. Assign a route, A to D, using the decision table in this lesson.
  8. For every task on Route C, rewrite it as two tasks: a verifiable part AI may prepare, and the part that stays with the licensee or another professional.
  9. Write the one-line triage entry for two of the tasks, in the format shown in the lesson, naming the licensee and the supervising broker.
  10. Ask a colleague to triage the same six tasks independently. Compare routes and note which of the five questions caused each difference.
  11. Take any task you both placed on Route D and write the sentence you would use to hand it to the person who owns it: the seller, an attorney, an appraiser, a lender or your broker.

Knowledge check

  1. What are the five triage questions in this lesson, and when are they asked?

    Answer guidance: Consequence, data class, verifiability, fair housing exposure and which licensee signs off under which broker. They are asked before any AI tool is opened.

  2. Why is a seller's question about what she must legally disclose Route D for the tool?

    Answer guidance: The disclosure is the seller's own statement and what the law requires is a legal question for an attorney. The agent does not ask a tool to answer it; at most a short acknowledgement is drafted and reviewed.

  3. What does South Carolina Code §40-57-820 say, and whom does it bind?

    Answer guidance: A licensee is responsible for all work product produced with the help of AI or similar programs, and a violation committed through them is treated as the licensee's own. It binds South Carolina real estate licensees only.

  4. Why can one good AI result not justify using the same tool on a different task?

    Answer guidance: Research on the jagged frontier found AI helped on tasks inside its capability and hurt accuracy on a task outside it. Tasks that look alike can sit on opposite sides, so each task is triaged on its own.

Sources and further reading