AI Listing Description Compliance Checklist for Agents

By Emily Terrell — Top Coach and Speaker at Tom Ferry International. Active San Antonio agent closing 70+ transactions a year.

An AI listing description compliance checklist is a five-step review that runs before publication: verify facts against source data, read for fair housing language, confirm broker identification, check altered media, and file the original. It takes about 90 seconds. This guide gives you the prompt block and the delegation-ready SOP.

Key Takeaways

• The compliance work belongs in the prompt and the review queue, not in your memory — a constraint you have to remember is a constraint you’ll skip at volume.
• The single highest-value line in a listing prompt asks the model to list every claim it could not tie to a fact you supplied.
• Broker identification is the violation happening most often, because AI never adds it and nobody notices it’s missing.
• A virtual assistant can run three of the five checks; two require a licensee’s judgment and can’t be delegated.
• Ninety seconds per asset is the target. If your review takes twenty minutes, you built a bottleneck, not a system.

What is an AI listing description compliance checklist?

An AI listing description compliance checklist is a fixed review sequence applied to every AI-drafted listing description before it reaches the MLS, a portal, or a social post. It exists because the speed that makes AI useful is the same speed that outruns a human review process built for one listing at a time.

It is not a legal framework and it is not a disclosure. Whether you disclose AI use is a separate question with a separate answer — I break that down in what real estate agents have to disclose when using AI. This is the operational layer underneath that decision: the thing you actually run on Tuesday morning with three listings to input.

Why this matters for real estate agents

The volume is already here. According to NAR’s 2025 Technology Survey (September 2025), 46% of agents report using AI-generated content, with listing descriptions cited as the primary use case. That same survey found 17% of agents saw a significantly positive impact from AI while 46% saw no noticeable difference at all.

Read those numbers together and the picture is clear. Half the industry is producing more listing copy than ever and getting nothing measurable from it — which means nobody is reviewing it either, because a review step that produces no value is the first thing to go.

The economics make the review step non-negotiable rather than optional. According to NAR’s 2026 Member Profile (June 2026), the typical individual agent reported nine transaction sides in 2025. Nine listings is not enough volume to absorb one advertising complaint, one fair housing inquiry, or one buyer who arrived expecting a third bedroom that the model invented.

“I run 70-plus transactions a year and every listing description passes through the same five checks in the same order. It takes ninety seconds, and it has never once been the reason a listing went out late.” — Emily Terrell, Tom Ferry Coach

The compliance block: what goes in the prompt

Most agents put their guardrails in the edit. That’s backwards. The edit is where you’re tired, rushed, and reading your own words for the third time. The prompt is where you’re calm and the constraint costs you nothing.

Paste this block at the end of any listing-copy prompt, in Claude or ChatGPT, every time.

COMPLIANCE CONSTRAINTS — apply to every output:

1. Use only the facts I supply below. If a detail is missing,

   write [VERIFY] rather than inferring it. Never estimate square

   footage, year built, school district, HOA dues, lot size,

   or tax figures.

2. Describe the property, never the occupant. Do not use language

   referring to or implying a preferred type of buyer or resident.

   This includes: family, families, kids, empty nesters, young

   professionals, bachelor, retirees, safe, quiet neighborhood,

   exclusive, walking distance, and any reference to nearby

   congregations or demographics.

3. No unverifiable claims or superlatives — best, rare, won’t

   last, turnkey, move-in ready, lowest price in the area —

   unless I supplied a fact that supports it.

4. Do not describe condition, systems, age, or repairs unless I

   supplied the fact with a date. No “new roof,” “updated

   electrical,” or “recently renovated” from inference.

5. End the output with a line reading:

   BROKER LINE: [Agent Name] | [Brokerage Name]

6. After the description, output a separate section titled

   UNVERIFIED CLAIMS listing every statement you could not tie

   directly to a fact I provided.

Why does the constraint belong in the prompt instead of the edit?

Because a draft that arrives close to compliant needs a review, and a draft that arrives non-compliant needs a rewrite. Those are different amounts of work, and only one of them survives a week with four listings and a stage commitment.

There’s a second reason. Constraint six changes the model from something you audit into something that audits itself. When the tool hands you its own list of inferences, you stop hunting for the fabricated detail and start confirming a list. That’s the difference between a 90-second review and a 15-minute one.

What does the UNVERIFIED CLAIMS section actually catch?

Square footage the model rounded. A “spacious” primary suite you never described as spacious. A “quiet cul-de-sac” it inferred from an address. A “recently updated kitchen” built out of the word “granite” in your notes. None of these read as errors — that’s exactly the problem. They read as good copy, which is why they get published.

The 90-second review: five checks in fixed order

Order matters. Run them in sequence, every asset, no exceptions.

1. What do you verify first?

Facts against source, not against the draft. Open the MLS input record or the seller’s disclosure and check every number and feature claim in the description against it. Read from the source to the draft, not the other way around — reading the draft first primes you to confirm what’s already on the page.

Start with the UNVERIFIED CLAIMS list of the model generated. Everything on it either gets a source or gets deleted.

2. How do you run a fair housing read in 20 seconds?

Read only the words that describe people or lifestyle, and ignore everything else. Section 3604(c) of the Fair Housing Act makes it unlawful to publish any advertisement regarding the sale or rental of a dwelling that indicates a preference, limitation, or discrimination based on a protected characteristic (42 U.S.C. § 3604). The statute reaches the advertisement, not the author — so who drafted it has no bearing on liability.

The practical version: scan for any noun or adjective that describes a person rather than a structure. “Three bedrooms” describes a structure. “Room for a growing family” describes a person. Delete the second category entirely and you’ve cleared most of the exposure.

3. Where does broker identification break?

In the batch. TREC defines an advertisement to include social media, email, text, and the internet, and Rule 535.155 requires every advertisement to carry the license holder’s or team’s name plus the broker’s name at no less than half the size of the largest agent contact information in the ad (TREC). Agents outside Texas need the equivalent rule from their own commission — most states have one.

Here’s the thing nobody wants to tell you: ask a model for thirty captions and you’ll get thirty captions with zero broker names, because the tool has no idea it’s required. Publish them straight from the output and you just produced thirty non-compliant advertisements. That’s the violation happening at volume right now, and I cover why it outranks every AI-specific headline in what actually risks your license.

Constraint five in the prompt block handles this on the way in. Check five confirms it on the way out.

4. What if the listing has altered images?

Then a different rule applies, and it’s the one with actual statutory teeth. California’s AB 723 added Section 10140.8 to the Business and Professions Code, requiring that any digitally altered listing image carry a reasonably conspicuous disclosure on or adjacent to the image plus a link, URL, or QR code to the original, unaltered version. The statute defines alteration to include changes made with photo editing software or artificial intelligence to add, remove, or change elements such as fixtures, furniture, flooring, paint color, landscape, facade, or views through windows — while excluding routine lighting, cropping, white balance, and exposure corrections that don’t change how the property is represented (California AB 723, Chapter 497, Statutes of 2025).

Wisconsin follows with a broader trigger. Under 2025 Wisconsin Act 69, effective January 1, 2027, Wis. Stat. 452.136(1m) requires a licensee to disclose in all advertising that has been altered or modified using technology, including artificial intelligence, to add, remove, or change elements of the property in a way that creates a false or misleading impression (Wisconsin Legislature). Note the wording — that provision reaches advertising generally, not images specifically, and it hinges on the impression created rather than the medium.

Check your own MLS rules separately. Many require virtually staged labeling independent of any state statute.

5. What gets stored, and where?

The original image, the published version, and the final description — all three into the transaction file, on every listing, whether your state requires it or not. This is the step everyone skips because nothing happens when you skip it, right up until something does.

It costs one drag-and-drop. It removes an entire category of future problems.

How to hand this to a VA or TC

The reason most compliance SOPs die is that they’re written for the person who built them. This one splits cleanly.

Your assistant can run: the fact verification against source data, the broker line check, and the file storage step. All three are mechanical — a comparison, a presence check, and an upload. Write them as a three-line checklist and they’re done.

You have to run: the fair housing read and the altered-media determination. Both require judgment about what a reasonable consumer would believe, which is licensed work. Confirm the boundaries of what unlicensed staff can do with your broker and your state commission before you delegate anything — TREC and most commissions have specific rules on unlicensed assistants.

Build it as one document with the split marked, and the handoff stops being a conversation you repeat every quarter. That’s the same principle behind the compliance segment I teach in AI training for real estate agents.

This is general information, not legal advice. Advertising, fair housing, and AI disclosure rules vary by state, MLS, and brokerage, and they’re changing quickly. Confirm your obligations with your broker and an attorney licensed in your state before you set policy.

How I use this in my own business

I close 70+ transactions a year in San Antonio in roughly five hours a week of active management, and the only reason the listing marketing side holds together is that the compliance step lives inside the workflow instead of on top of it.

My compliance block is saved as a project instruction in Claude, so it applies to every listing prompt without me pasting anything. The draft comes back with the broker line already attached and an UNVERIFIED CLAIMS list underneath it. On a recent Stone Oak listing, that list flagged “generously sized primary suite” — I’d given the model dimensions, not an adjective. Thirty seconds to check the measurement, and either it earns the word or the word comes out.

Then the description moves to the review queue in Follow Up Boss with the transaction, where my TC runs the three mechanical checks and I run the two that need a license. Feet on the desk, coffee in hand. That’s the system working — not me being more careful, just a process that doesn’t depend on me being careful.

Common mistakes

Putting the guardrails in the edit. You’ll catch it the first three times and miss it the fourth, which will be the listing that gets syndicated to six portals before anyone reads it.

Reading the draft before the source. Once you’ve read the polished version, your brain confirms rather than verifies. Source first, always.

Delegating the fair housing read. It looks like a proofreading task. It isn’t — it’s a judgment about what language implies, which is exactly the work your license covers.

Disclosing everything as a hedge. A blanket “AI-assisted” label on every asset is not compliant. It’s noise that trains clients to ignore the label in the one place it actually matters.

Treating a text-only rule set as complete. The statutes with real specificity right now govern images, not words. If your checklist stops at the description, it stops before the part with a criminal penalty attached in California.

Building a twenty-minute review. A process that costs more than it saves gets abandoned inside a month. Ninety seconds is not a shortcut — it’s what makes the thing survive.

Frequently Asked Questions

Do I have to disclose that I used AI to write a listing description?

Generally no. No state currently requires a licensee to label listing text as AI-authored. What every state requires is that the description be accurate and not misleading, and you own that obligation regardless of what produced the first draft. The disclosure statutes that do exist target digitally altered images, not written copy. Confirm your MLS rules separately.

What should an AI listing description compliance checklist include?

Five checks in fixed order: verify every fact against source data rather than the draft, read the copy for fair housing language, confirm broker identification is present and correctly sized, determine whether any accompanying media was altered and needs disclosure, and file the original alongside the published version. Run them in that sequence on every asset, without exception.

How do I stop AI from writing fair housing violations?

Put the constraint in the prompt rather than the edit. Instruct the model to describe the property and never the occupant, and give it the specific banned vocabulary — family, kids, safe, quiet, walking distance, young professionals. Then run a 20-second read of only the people-describing words before publishing. Prompt-level constraints prevent; edit-level catches fail at volume.

Can my assistant review AI listing copy for compliance?

Partially. Fact verification against source records, broker line presence, and file storage are mechanical and delegable. The fair housing reading and the altered-media determination require judgment about consumer perception, which is licensed work. Split your checklist explicitly and confirm what unlicensed staff may do with your broker and your state commission before delegating.

Does AI-generated listing copy violate TREC advertising rules?

Not by being AI-generated. It violates them the same way any copy does — by being misleading, or by omitting the broker’s name. TREC Rule 535.155 requires broker identification on every advertisement including social media and internet content, and AI tools never add it. Batch-published output is where this fails most often.

What is the single most useful line in a listing prompt?

Ask the model to output a separate list of every claim it could not tie to a fact you supplied. That one instruction converts your review from hunting for fabrications to confirming a short list. It surfaces rounded square footage, inferred conditions, and adjectives with no source behind them — the errors that read like good copy.

Do I need to keep original listing photos?

Yes, as a default habit regardless of your state. California requires access to the original whenever an image has been digitally altered, and Wisconsin’s rule arrives in 2027. Even where no statute applies, keeping the original and the published version in the transaction file costs one upload and eliminates an entire category of dispute later.

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.