How Should Brokerages Audit AI Listing Descriptions?

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Speaker for NAHREP, eXp Con, and brokerages nationwide.

Brokerages should audit AI listing descriptions with a standing system: one approved prompt template, a licensee review before publishing, and a monthly sample read by a delegated supervisor against a fixed scan list. The goal is catching template-level drift before it repeats across every listing. This guide covers the audit cadence, the scan categories, and the log.

Key Takeaways

• Audit the prompt template before you audit the agent, because one saved template repeats its phrasing across every listing it touches.
• A monthly random sample read by a delegated supervisor catches drift within one listing cycle, which a quarterly or annual review can’t.
• Keyword scanners are a first filter, not a compliance program, because the legal test is what the full description conveys to an ordinary reader.
• In Texas, TREC already requires sponsoring brokers to keep written procedures for advertising compliance, and AI-drafted copy is advertising.
• The audit log is the real deliverable: it shows supervision happened and tells you whether to coach one agent or fix the source for everyone.

What is an AI listing description audit?

An AI listing description audit is a recurring, brokerage-level review of the listing copy agents drafted with AI tools. It checks that copy against a fixed standard for fair housing language, factual accuracy, and advertising rules.

It’s different from an agent’s pre-publish review. The agent’s review catches one listing before it goes live. The brokerage audit catches the pattern across all of them after they do.

You need both. The agent-side version — prompt block, five checks, ninety seconds — is in my AI listing description compliance checklist. This post is the layer above it.

Why this matters for brokerage leaders

Volume moved faster than supervision. According to NAR’s 2025 Technology Survey (September 2025), 46% of agents report using AI-generated content, and NAR names listing descriptions as the example use. Most brokerage supervision was designed for a world where one agent wrote one description at a time.

Teams multiply the effect. According to NAR’s 2026 Member Profile (June 2026), 21% of Realtors work on a team. Team-based brokerage specialists reported a median of 32 transaction sides in 2025. That’s one team and one shared prompt producing dozens of public listings a year, all with the same phrasing.

The people reading those listings have changed too. According to the National Fair Housing Alliance’s 2025 Fair Housing Trends Report (November 2025), 32,321 housing discrimination complaints were filed in 2024. Private nonprofit fair housing organizations processed 74.12% of them.

Advertising is a small category in that data: 108 complaints in 2024. That’s exactly why brokerages under-invest here. But low frequency isn’t low exposure. A listing description is a published, dated, permanent document, so nobody has to reconstruct what was said.

The federal backdrop didn’t loosen the rule either. HUD withdrew its 2024 digital advertising guidance effective September 17, 2025. Its April 2026 Federal Register notice also states two things:

• Conduct that doesn’t comply with the Fair Housing Act remains subject to enforcement.
• Complainants can still file civil actions within two years.

The regulation that governs listing copy, 24 CFR 100.75, prohibits any statement indicating a preference based on race, color, religion, sex, handicap, familial status, or national origin. HUD didn’t touch it. I break down the withdrawal in can AI violate fair housing in a listing description?

Here’s the thing nobody wants to tell you: the brokerage’s exposure isn’t one agent’s bad sentence. It’s the saved template nobody has reread since the day it was written.

“A brokerage that audits agents finds mistakes. A brokerage that audits templates finds the one prompt producing the same problem across every listing it touches — and fixes all of them with a single edit.”
— Emily Terrell, Tom Ferry Coach

The five-part brokerage audit system

Who owns the prompt template?

The brokerage does. Publish one approved listing-description prompt with the fair housing constraint block built in, give it a version number, and make it the default for every agent.

Agents can add property facts and adjust tone. They can’t delete the constraint block. The version number goes at the top of the prompt (v1.0, v1.1) so every published description traces back to the template that produced it. Without version control, a finding tells you something went wrong. With it, the finding tells you where.

If your brokerage doesn’t have a written AI policy yet, start there. The four sections it needs are in why your brokerage’s AI policy isn’t enough.

Who runs the audit?

A licensed supervisor, delegated in writing. In Texas, TREC Rule 535.2 lets a broker delegate to another license holder the job of helping administer compliance. The broker can’t hand off overall responsibility for supervision, though, and the delegation must be in writing.

The same subsection requires anyone who leads a team to be delegated as a supervisor. That makes the team lead the natural owners of their team’s sample.

Don’t hand this to an unlicensed assistant. Pulling the sample is mechanical. Judging what a description conveys to a reader is licensed work. Brokers outside Texas should confirm their own commission’s delegation rules.

How big should the sample be?

Pull it monthly and cover every agent who published in the window, with a minimum of ten descriptions. Pull randomly; agents don’t choose what gets reviewed.

The sample includes:

• MLS public remarks
• Agent remarks and showing instructions
• Social captions that carry the brokerage name

Three situations get a full review instead of a sample:

• An agent’s first listings
• Any agent flagged in the prior cycle
• Every listing published after a template version change

The monthly cadence is the point. A quarterly audit lets a bad template run for ninety days of listings before anyone reads it.

What does the reviewer read for?

Four categories, from one scan list the brokerage owns:

• The buyer instead of the property. “Perfect for a growing family,” “ideal for empty nesters,” “great starter home for a young couple.”
• The people instead of the place. Characterizations of residents, or of area safety and desirability.
• Religious proxies. Named houses of worship offered as selling points.
• Accessibility as a limitation. Statements about who the home isn’t suitable for, rather than a description of the feature itself.

Then two non-fair-housing checks: facts against the MLS record, and broker identification.

The reviewer reads every sampled description in full. A keyword scanner is fine as a first pass. But courts read Section 3604(c) through an ordinary-reader standard (Ragin v. New York Times Co., 2d Cir. 1991), which asks what the description conveys as a whole. Paraphrased preferences never trip a word list.

What happens to a finding?

Log it, then classify where it came from. The classification decides the fix:

• Template. Fix the template, issue a new version, and re-review every listing published on the old version.
• Agent edit. Coach that agent and put them on full review next cycle.
• Tool-native generator. An MLS or platform AI feature produced a copy that bypassed your template entirely. Add the tool to your inventory and decide whether it’s approved.

In every case, correct the live copy everywhere it was syndicated, not just in the MLS.

For Texas brokers, this procedure does double duty. TREC Rule 535.2(i) requires sponsoring brokers to keep current written policies and procedures ensuring agents comply with the Commission’s advertising rules. A written audit, actually followed, is that policy in practice.

Common mistakes

Auditing agents instead of templates. If three agents produce the same phrase, you don’t have three coaching conversations. You have one broken template.

Treating the word list as the standard. The list is a filter. The legal test is what a reader takes away from the whole description.

Sampling only MLS remarks. 24 CFR 100.75 reaches written and oral statements, including flyers, signs, and any document used in the sale. Captions and showing instructions count.

Ignoring AI built into the tools you already pay for. MLS systems and marketing platforms keep adding description generators. If one bypasses your template, your audit has a blind spot exactly where volume is highest.

Running it once. A one-time audit is a project. A monthly audit is supervision. Only one of those shows up in your records as an ongoing practice.

Delegating the read to unlicensed staff. Pulling the sample is administrative. Judging the language isn’t.

This is general information, not legal advice. Fair housing law is enforced at the federal, state, and local level, and broker supervision rules vary by state. Consult your state real estate commission and an attorney licensed in your state before setting brokerage policy.

Frequently Asked Questions

How often should a brokerage audit AI listing descriptions?

Monthly, on a random sample, plus a full review any time a saved prompt template changes. A monthly cadence catches template drift within one listing cycle instead of one quarter. Size the sample so it covers every agent who published in the window, with a floor of ten descriptions. Increase frequency for new agents and for anyone whose descriptions were flagged in the prior audit.

Who is liable if an agent’s AI listing description violates fair housing?

The publishing agent, and potentially the sponsoring broker. Federal regulation 24 CFR 100.75 reaches anyone who makes, prints, or publishes a discriminatory statement, and authorship by an AI tool isn’t a defense. In Texas, TREC Rule 535.2 makes a broker responsible for the authorized acts of sponsored agents and for ensuring their advertising complies with Commission rules.

Can a keyword scanner catch fair housing problems in AI listing copy?

Partially. A keyword scan catches obvious phrases like “perfect for families,” but courts read Section 3604(c) through an ordinary-reader standard that judges what the full description conveys, not whether a banned word appears. Paraphrased preferences and neighborhood characterizations slip past word lists. Use the scanner as a first filter, then have a licensed supervisor read flagged and sampled descriptions in full.

Does Texas require brokers to have written advertising policies?

Yes. TREC Rule 535.2(i) requires a broker who sponsors sales agents to maintain current written policies and procedures ensuring each sponsored agent complies with the Commission’s advertising rules. The rule doesn’t mention AI, but AI-drafted listing copy is still advertising. A written audit procedure is the most direct way to show those policies exist and are being followed.

Did HUD’s 2025 guidance withdrawal lower the risk for brokerages?

No. HUD withdrew eight guidance documents effective September 17, 2025, including its 2024 guidance on digital advertising, but the regulation governing discriminatory statements, 24 CFR 100.75, was not withdrawn. HUD’s notice states that conduct not complying with the Fair Housing Act remains subject to enforcement and that complainants may still file civil actions within two years.

Should a brokerage ban agents from using AI for listing descriptions?

No. A ban pushes drafting into personal accounts where the brokerage can’t see the prompt or the output, which removes the very supervision a broker is responsible for. NAR’s 2025 Technology Survey found 46% of agents already use AI-generated content. An approved template, a review gate, and a monthly audit give the broker visibility a ban never will.

What should a brokerage audit log record?

For each sampled description: the listing, the agent, the publish date, the channels it ran on, the prompt template version, any flagged language, and the corrective action taken. Record whether the issue traced to the template or to an individual edit. That last field tells you whether to coach one agent or fix the source for everyone.

Bring this to your team or event

Emily Terrell speaks at brokerage events, real estate conferences, and team trainings on AI, systems, and social media — the exact playbook in this post, delivered live to your audience. As a Top Coach and Speaker at Tom Ferry International and an active agent closing 70+ transactions a year, Emily speaks from the stage about what’s working right now, not theory. Recent stages include NAHREP and eXp Con.

Book Emily to speak at your next event:
Email: eterrell@yourcoach.com
Phone: (210) 400-9191
Web: coachemilyterrell.com

For real estate agents who want to implement this: Get the weekly real estate prompt library at weeklyrealestateprompts.com or follow @coachemilyterrell on Instagram for daily systems and AI breakdowns.