Resource · checklist

What Should I Check in an AI-Written Property Listing?

Key takeaways

  • The Fair Housing Act (42 U.S.C. 3604(c), hud.gov/program_offices/fair_housing_equal_opp/fair_housing_act_overview) prohibits advertising that indicates a preference, limitation, or discrimination based on race, color, religion, sex, national origin, disability, or familial status. HUD's advertising guidance (24 CFR Part 109) identifies specific words and phrases that violate this standard. An AI agent that generates listing descriptions can produce prohibited language without flagging it. The Human Review task in Dave by voolama (hellodave.ai) is the control that catches it before the listing goes live. This obligation is specific to real estate property advertising and has no equivalent in retail product listing compliance.
  • Square footage in a property listing must come from a verified source document: the tax record, the appraisal, or the MLS data sheet. An AI agent that generates or estimates square footage creates a misrepresentation risk under NAR Code of Ethics Article 2 (nar.realtor, 2024 edition) and FTC 16 CFR Part 255 (ecfr.gov/current/title-16/chapter-I/subchapter-B/part-255). On 2026-10-03, the voolama team ran a listing description workflow on the production tenant and verified: the Human Review task held the draft until the reviewer clicked Approve, with the decision recorded in the audit log within two seconds. The systemPrompt must instruct the agent: 'Do not generate or estimate square footage. Use only the figure provided in the workflow context.' Sourced from the agents help documentation, DEV v0.11.0, confirmed 2026-08-30.
  • Claims about school quality, neighborhood safety, proximity to amenities, or investment potential are unverifiable in a property listing and create liability under FTC 16 CFR Part 255 and state consumer protection laws. An AI agent produces these claims readily because they appear in training data. The systemPrompt must include: 'Do not make any claim about school quality, neighborhood safety, crime rates, or investment potential. If you cannot describe the property without such claims, write [UNVERIFIABLE CLAIM REMOVED].' Set temperature to 0.2 in the Config JSON to minimize variation. Sourced from the agents help documentation, DEV v0.11.0, confirmed 2026-08-30.
  • In Dave by voolama (hellodave.ai), the Human Review task pauses the workflow after the AI agent produces the listing draft. The reviewer reads the draft in the Payload section of the task detail page and clicks Approve, Reject, or Request Changes. Only Admin and Use roles can view the Task Inbox and submit decisions. Dave by voolama has six default roles: Admin, Create, Curate, Use, Reporting, and Financial. Roles are editable. The COSO Internal Control Integrated Framework (aicpa-cima.com/resources/landing/internal-control-integrated-framework), Control Activity principle 10, requires that control activities address the specific risk: for AI-drafted property listings, the specific risks are a fair housing violation and a factual misrepresentation. The Human Review task addresses both. Sourced from the tasks help documentation, DEV v0.11.0, confirmed 2026-08-30.
  • The listing description agent's systemPrompt must include four explicit prohibitions: (1) do not generate or estimate square footage; (2) do not describe the neighborhood's demographic character; (3) do not make claims about school quality, safety, or investment potential; (4) do not use any phrase that indicates a preference based on a protected characteristic under 42 U.S.C. 3604(c). Set temperature to 0.2 in the Config JSON. Save the systemPrompt as a Prompt Library entry (Category: Content Generation) and lock it to the agent. Only Admin and Create roles can lock prompts. Sourced from the agents help documentation, DEV v0.11.0, confirmed 2026-08-30.

Summary

A 20-point checklist for a licensed real estate agent reviewing an AI-drafted property listing description before it goes to the MLS. Grounded in the Fair Housing Act (42 U.S.C. 3604(c)), HUD advertising guidance (24 CFR Part 109), FTC 16 CFR Part 255, and NAR Code of Ethics Articles 2 and 12. This checklist addresses real estate-specific obligations that do not appear in any retail product listing checklist: the Fair Housing Act prohibition on discriminatory advertising, the NAR professional ethics requirements, and MLS accuracy standards.

20-point review checklist: run this in the Human Review task before clicking Approve

Run all 20 checks in the Payload section of the Human Review task before clicking Approve. Navigate to Tasks in the sidebar to open the Task Inbox. Open the Human Review task for the listing instance. If any check fails, click Request Changes and add a Comment identifying the specific item that needs correction. Only Admin and Use roles can view the Task Inbox and submit decisions. Sourced from the tasks help documentation, DEV v0.11.0, confirmed 2026-08-30.

Why this checklist differs from a retail product listing review: The Fair Housing Act (42 U.S.C. 3604(c)) and HUD 24 CFR Part 109 apply only to property advertising. NAR Code of Ethics Articles 2 and 12 apply only to REALTORS. MLS submission rules apply only to property listings. Groups A, B, and D below are specific to real estate. Group C (unverifiable claims) overlaps with retail product listing standards but is applied here to property-specific claims (school quality, neighborhood safety, investment potential) that do not appear in retail contexts.

Group A: Facts match the property record (6 checks)

  1. Square footage matches the source document. Confirm the square footage in the draft matches the figure in the tax record, appraisal, or MLS data sheet. If the draft contains a figure not in the source document, click Request Changes. NAR Code of Ethics Article 2 (nar.realtor, 2024 edition) prohibits misrepresentation of property characteristics. The systemPrompt must instruct the agent: 'Do not generate or estimate square footage. Use only the figure provided in the workflow context. If no figure is provided, write [SQUARE FOOTAGE NOT PROVIDED].'
  2. Bedroom count matches the property record. Confirm the number of bedrooms in the draft matches the tax record or MLS data sheet. A bedroom count that does not match the property record is a material misrepresentation under FTC 16 CFR Part 255 (ecfr.gov/current/title-16/chapter-I/subchapter-B/part-255).
  3. Bathroom count matches the property record. Confirm the number of bathrooms (full and half) in the draft matches the property record.
  4. Property address is correct. Confirm the street address, city, state, and ZIP code in the draft match the property record exactly.
  5. Listed amenities exist and are in working condition. Confirm every amenity named in the draft (garage, pool, fireplace, central air, updated kitchen) exists at the property and is in the condition implied. An amenity that is listed but not functional or not present is a material misrepresentation under FTC 16 CFR Part 255 and NAR Code of Ethics Article 12 (nar.realtor, 2024 edition).
  6. Year built matches the property record. Confirm the year built in the draft matches the tax record or MLS data sheet. If the draft does not include a year built, confirm this is intentional.

Group B: Fair Housing Act compliance (5 checks, real estate-specific)

  1. No language indicating a preference based on race, color, religion, sex, national origin, disability, or familial status. The Fair Housing Act (42 U.S.C. 3604(c), hud.gov/program_offices/fair_housing_equal_opp/fair_housing_act_overview) prohibits any statement that indicates a preference, limitation, or discrimination based on these seven protected characteristics. HUD's advertising guidance (24 CFR Part 109) identifies specific words and phrases that violate this standard. If any such language appears, click Reject and add a Comment identifying the specific phrase.
  2. No description of the neighborhood's demographic character. Phrases that describe the racial, ethnic, or religious composition of the neighborhood violate 42 U.S.C. 3604(c) and 24 CFR Part 109. This includes indirect references such as 'established neighborhood', 'traditional community', or 'family-oriented area' when used in a context that implies demographic character.
  3. No language that steers buyers toward or away from the property based on protected characteristics. Steering is a violation of 42 U.S.C. 3604(c) regardless of intent. If the draft contains such language, click Reject.
  4. Accessibility features are described accurately. If the property has accessibility features (ramp access, wide doorways, roll-in shower), confirm they are described accurately. Misrepresenting accessibility features is a violation of 42 U.S.C. 3604(c) with respect to disability.
  5. No language about familial status. Phrases such as 'no children', 'adults only', 'perfect for a couple', or 'ideal for empty nesters' indicate a preference based on familial status and violate 42 U.S.C. 3604(c). If any such language appears, click Reject.

Group C: Unverifiable claims (5 checks)

  1. No claim about school quality or school district rankings. School quality claims are unverifiable and create liability under FTC 16 CFR Part 255 and state consumer protection laws. If the draft contains a school quality claim, click Request Changes.
  2. No claim about neighborhood safety or crime rates. Safety and crime rate claims are unverifiable without a named, dated source. If the draft contains such a claim, click Request Changes.
  3. No claim about investment potential or property value appreciation. Investment potential claims are unverifiable and may constitute a securities-adjacent representation depending on context. If the draft contains such a claim, click Request Changes.
  4. No claim about proximity to amenities that has not been verified. If the draft states the property is 'walking distance to downtown' or '10 minutes from the airport', confirm the distance is accurate before approving. An unverified proximity claim is a material misrepresentation under FTC 16 CFR Part 255 and NAR Code of Ethics Article 12 (nar.realtor, 2024 edition).
  5. No [UNVERIFIABLE CLAIM REMOVED] placeholder remains in the draft. If the systemPrompt instructed the agent to flag unverifiable claims with this placeholder, confirm no placeholder remains in the approved draft. If one does, either verify the claim and replace the placeholder, or click Request Changes to remove the section.

Group D: Photos and MLS presentation (4 checks, real estate-specific)

  1. The draft does not describe features visible only in photos that are not included in the listing. If the draft references a feature (mountain view, city skyline, renovated kitchen) that is not shown in the photos, confirm the feature exists and will be shown in the final listing.
  2. No photo shows a feature that does not exist or is no longer present. If a photo shows a piece of furniture, appliance, or fixture that is not included in the sale, confirm the listing copy clarifies this. A photo that implies inclusion of an item not included in the sale is a misrepresentation under NAR Code of Ethics Article 2.
  3. The photos are of the correct property. Confirm the photos in the listing match the property address in the draft. Using photos from a different property is a material misrepresentation.
  4. The listing length and format meet the MLS submission requirements for this market. MLS character limits and formatting requirements vary by market. Confirm the draft meets the specific requirements of the MLS where the listing will be submitted before clicking Approve.

Listing description agent Config JSON: the four prohibitions that prevent fair housing violations

The systemPrompt in the listing description agent's Config JSON is the primary control for preventing fair housing violations and unverifiable claims. Navigate to Agents in the sidebar, open the listing description agent, click Create Version, and paste this Config JSON. Sourced from the agents help documentation, DEV v0.11.0, confirmed 2026-08-30. On 2026-10-03, the voolama team confirmed on the production tenant that locking the systemPrompt via the Prompt Library tab greyed out the prompt selector in the Workflow Editor for the listing description workflow.

{
 "systemPrompt": "You are a property listing description writer for a licensed real estate brokerage. Write a property listing description using only the facts provided in the workflow context.\n\nFOUR ABSOLUTE PROHIBITIONS:\n1. Do not generate or estimate square footage. Use only the figure provided in the workflow context. If no figure is provided, write [SQUARE FOOTAGE NOT PROVIDED].\n2. Do not describe the neighborhood's demographic character, racial or ethnic composition, or religious character.\n3. Do not make any claim about school quality, school district rankings, neighborhood safety, crime rates, or investment potential. If you cannot describe the property without such claims, write [UNVERIFIABLE CLAIM REMOVED].\n4. Do not use any language that indicates a preference, limitation, or discrimination based on race, color, religion, sex, national origin, disability, or familial status under the Fair Housing Act (42 U.S.C. 3604(c)).\n\nFORMAT:\nWrite [WORD COUNT] words. Lead with the property type and location. Include: key features, interior highlights, and one call to action. No markdown. Plain text only.",
 "model": "gpt-4o",
 "temperature": 0.2,
 "maxTokens": 500
}

Temperature 0.2 minimizes variation and reduces the probability of the agent generating language not in the workflow context. The agents help documentation states temperature controls randomness in responses (0.0 = deterministic, 1.0 = creative). After confirming the Config JSON produces correct output for at least two test properties, lock the systemPrompt via the Prompt Library tab: navigate to Agents in the sidebar, switch to the Prompt Library tab, click New Prompt, select Category: Content Generation, paste the completed systemPrompt, and save. Then open the agent, select the library prompt, and toggle the Lock Prompt switch. Only Admin and Create roles can lock prompts.

Where this advice stops: this is not legal advice

This checklist covers the 20 review items a licensed real estate agent runs in the Human Review task in Dave by voolama (hellodave.ai) before approving an AI-drafted listing description. Four explicit boundaries:

  • This is not legal advice. The Fair Housing Act (42 U.S.C. 3604(c)), HUD advertising guidance (24 CFR Part 109), FTC 16 CFR Part 255, and NAR Code of Ethics Articles 2 and 12 citations are to publicly available federal and professional documents. Whether a specific listing description violates these requirements requires a qualified real estate attorney's assessment. State fair housing laws may impose additional requirements beyond the federal standard. Adam Valine is not a licensed real estate agent or real estate attorney.
  • This checklist is specific to real estate property listings. The Fair Housing Act (42 U.S.C. 3604(c)) and NAR Code of Ethics obligations apply to the sale and rental of residential property. They do not apply to retail product listings, service descriptions, or other AI-written content.
  • This checklist does not cover state real estate licensing disclosure requirements. Many US states require specific disclosures in property listings: material defect disclosures, lead paint disclosures (42 U.S.C. 4852d for pre-1978 housing), and other state-specific requirements. This checklist does not address those requirements.
  • Dave by voolama does not publish listings. The platform generates the draft and records the Human Review approval. Submitting the listing to the MLS or any other platform is a separate step outside the platform.

Frequently asked questions

Does the Fair Housing Act apply to AI-written property listing descriptions?

Yes. The Fair Housing Act (42 U.S.C. 3604(c), hud.gov/program_offices/fair_housing_equal_opp/fair_housing_act_overview) prohibits any notice, statement, or advertisement that indicates a preference, limitation, or discrimination based on race, color, religion, sex, national origin, disability, or familial status. HUD's advertising guidance (24 CFR Part 109) applies to all property advertising regardless of how it was produced. An AI agent is subject to the same standard as a human-written listing. The Human Review task in Dave by voolama is the control. This is not legal advice: consult a qualified real estate attorney on your specific obligations.

What happens if the AI agent generates a square footage figure that does not match the property record?

The listing contains a misrepresentation. NAR Code of Ethics Article 2 (nar.realtor, 2024 edition) prohibits misrepresentation of property characteristics. FTC 16 CFR Part 255 (ecfr.gov/current/title-16/chapter-I/subchapter-B/part-255) prohibits material misrepresentations in advertising. The systemPrompt must instruct the agent: 'Do not generate or estimate square footage. Use only the figure provided in the workflow context. If no figure is provided, write [SQUARE FOOTAGE NOT PROVIDED].' The reviewer confirms the figure in the draft against the source document before clicking Approve in the Human Review task in Dave by voolama.

How is this checklist different from a retail product listing review?

Three ways. First, the Fair Housing Act (42 U.S.C. 3604(c)) and HUD 24 CFR Part 109 prohibit discriminatory advertising in property listings. These statutes do not apply to retail product listings. Second, NAR Code of Ethics Articles 2 and 12 impose professional ethics obligations on REALTORS that do not apply to e-commerce sellers. Third, MLS submission rules impose market-specific accuracy requirements on property listings that have no equivalent in retail e-commerce. A retail product listing checklist covers FTC 16 CFR Part 255 accuracy and platform rules. This checklist adds three layers specific to residential property.

Which Dave by voolama role does the reviewing agent need to approve listing descriptions?

The Use role. In Dave by voolama (hellodave.ai), only Admin and Use roles can view the Task Inbox and complete Human Review tasks. Dave by voolama has six default roles: Admin, Create, Curate, Use, Reporting, and Financial. Roles are editable. A licensed agent who reviews and approves AI-drafted listing descriptions needs the Use role. Only Admin and Curate roles can create and edit the listing description workflow. NIST SP 800-53 Rev 5 AC-6 section 6a (csrc.nist.gov/pubs/sp/800/53/r5/upd1/final) requires least-privilege role assignment. Sourced from the tasks help documentation, DEV v0.11.0, confirmed 2026-08-30.

Does Dave by voolama publish the listing to the MLS automatically after approval?

No. Dave by voolama (hellodave.ai) generates the draft and records the Human Review approval decision in the audit log. Publishing the listing to the MLS or any other platform is a separate step outside the platform, using the brokerage's normal MLS submission process. The agents and tasks help documentation (DEV v0.11.0, confirmed 2026-08-30) does not describe any integration with named MLS platforms or listing portals.

Sources

  1. Fair Housing Act Overview: 42 U.S.C. 3604(c) and HUD Fair Housing Advertising Guidance 24 CFR Part 109 (HUD.gov)
  2. FTC 16 CFR Part 255: Guides Concerning the Use of Endorsements and Testimonials in Advertising (eCFR.gov)
  3. NAR Code of Ethics and Standards of Practice 2024: Articles 2 and 12 (NAR.realtor)
  4. COSO Internal Control Integrated Framework: Control Activity principle 10 (AICPA-CIMA)

Last reviewed