NAR, Fair Housing, and AI: Governance for Brokerages in 2026

Every designated broker hits the same uncomfortable moment in 2026. AI is in the brokerage already — agents using ChatGPT for listing copy, the ISA running an AI dialer, the marketing team feeding Claude every market report — and there is no written policy on any of it. The brokerage's E&O carrier has started asking about AI use at renewal. The state real estate commission is circulating draft disclosure rules. NAR has put out advisory guidance. And the designated broker is the one who signs the policy.

This article is the governance scaffold for that signature. It maps brokerage AI use against the NAR Code of Ethics, the Federal Fair Housing Act, RESPA, TCPA, and the state license law layer. It pairs with the residential AI rollout playbook and the workflow-specific compliance notes in the transaction-coordination article and the listing-description automation article.

The 5 regulatory layers the brokerage AI policy has to cover

1. NAR Code of Ethics — Articles 1, 10, and 12

The NAR Code of Ethics is the floor for any brokerage where members hold the REALTOR designation. Three articles bear directly on AI:

• Article 1. REALTORS protect and promote the interests of their client. AI that recommends a lender, title company, or affiliated business must serve the client's interest, not the brokerage's. Any AI-drafted email that suggests a preferred provider needs the same disclosure a human referral would carry.
• Article 10. REALTORS shall offer equal professional service without discrimination on protected classes. AI-generated listing-copy that uses protected-class language is an Article 10 violation in addition to a federal Fair Housing Act violation. Same for AI-targeted advertising that excludes protected classes from the audience.
• Article 12. REALTORS shall be honest and truthful in their advertising. AI-generated content (listing copy, market reports, marketing emails) is advertising; the agent of record is responsible for accuracy. AI-generated property descriptions that overstate condition or amenities are Article 12 problems.

2. Federal Fair Housing Act

The FHA covers seven federally protected classes: race, color, national origin, religion, sex (including sexual orientation and gender identity per 2021 HUD guidance), familial status, and disability. AI-generated content fails the FHA in four predictable ways:

• Familial-status hints. "Perfect for a young family," "ideal for retirees," "great for empty-nesters."
• Religious hints. Naming nearby houses of worship, "vibrant Christian community," "walk to the synagogue."
• Disability hints. "No stairs — great for retirees," "active adult community" outside legitimately licensed 55+ properties.
• Source-of-income hints. "No Section 8," "voucher friendly" in jurisdictions that protect source of income.

Real estate-trained AI (Lofty, kvCORE, Listing AI) ships with these guardrails. Consumer-tier AI (free ChatGPT, free Claude) does not. The brokerage policy has to specify which tools are approved for listing-content generation and which are not.

3. RESPA Section 8

Section 8 of the Real Estate Settlement Procedures Act prohibits kickbacks and unearned fees for referrals on federally related mortgages. AI nurture and transaction emails that route the buyer or seller to a preferred lender, title company, or affiliated business must comply.

• Affiliated business arrangement (AffBA) disclosures. If the brokerage owns or partners with the recommended lender or title company, the AffBA disclosure goes in the AI-drafted communication. AI templates have to embed the disclosure language at the template level, not rely on the TC or agent to add it manually.
• Marketing services agreement compliance. Co-marketing arrangements with lenders must reflect actual marketing services exchanged. AI cannot manufacture content credit; the brokerage logs the actual content produced under each MSA.
• No "preferred provider" without disclosure. AI emails that say "we recommend XYZ Title" without the AffBA disclosure are RESPA violations whether a human or AI wrote them.

4. TCPA and state telemarketing rules

The Telephone Consumer Protection Act governs every AI dialer, sms-drip, and outbound voice campaign the brokerage runs. Three rules:

• Express written consent. Inbound web-lead consent on Zillow, Realtor.com, and IDX-form leads is generally treated as express written consent — verify the lead-source contract language. Cold-list SMS requires affirmative opt-in.
• Quiet hours. No SMS or call before 8 a.m. or after 9 p.m. in the recipient's local time zone.
• Opt-out honored. "Reply STOP to opt out" included in every SMS; opt-out removes the contact from all sequences within 24 hours.

Florida's Telephone Solicitation Act, Oklahoma's TCPA-equivalent, and Washington's CEMA all add state-specific rules that are stricter than federal TCPA. Brokerages operating in those states need policy language that respects the strictest applicable rule.

5. State real estate commission rules

Every state real estate commission has license-law rules that apply to AI in addition to NAR and federal law. Three patterns are emerging across 2026:

• Disclosure of AI in communications. Some states (California, Colorado, draft rules in Florida and Texas) are heading toward mandatory disclosure when AI handles client-facing communications. The brokerage policy should require a one-line addendum in the buyer/seller representation agreement disclosing AI use in communications and content.
• Unauthorized practice of law (UPL) limits. AI that drafts contract addenda or interprets contract clauses crosses into UPL territory. The policy must specify that AI parses dates and drafts party-update communications; it does not interpret clauses. The agent or attorney handles interpretation.
• Supervision requirements. Designated brokers are responsible for supervising agent conduct. Brokerages with 25+ agents need a documented audit process — pull 10 random AI-generated listings and 10 random AI-fired SMS conversations monthly and review for compliance.

The brokerage AI policy: 8 sections

A minimum-viable brokerage AI policy covers eight sections. Each section is 1 to 3 paragraphs; the full policy runs 6 to 10 pages.

1. Approved tools list. Specific tools authorized for client-facing work (Lofty, kvCORE, Follow Up Boss + FUB AI, ChatGPT Enterprise, Claude Team). Specific tools prohibited or restricted (free-tier AI for client PII).
2. Listing-content review. Every AI-generated listing description gets agent review plus an automated Fair Housing scan before publishing.
3. Outbound communications consent. Every contact in the database must have documented consent for SMS and voice contact. The CRM enforces; the team lead audits monthly.
4. AffBA and RESPA compliance. AI templates that touch lender, title, or affiliated business include the AffBA disclosure language at the template level.
5. Contract interpretation boundary. AI parses dates. Agents and brokers interpret clauses. AI does not draft addenda.
6. Client disclosure. The buyer/seller representation agreement includes a one-line AI-use disclosure. Clients ask, AI tools answer honestly when asked if they are speaking with a person.
7. Data handling. No client PII enters consumer-tier AI tools. Enterprise-tier tools with appropriate DPA in place for any client data.
8. Audit and review. Monthly review of 10 random AI-generated listings, 10 random AI-fired SMS conversations, and the brokerage Fair Housing scan log.

The 30-day rollout

Days 1 to 7. Designated broker drafts the policy using the eight sections above plus state-specific addenda. Brokerage attorney reviews.

Days 8 to 14. Agents and staff sign the policy. The buyer/seller representation agreement gets the AI-use addendum added.

Days 15 to 21. Approved tools list goes live. Any tool not on the list gets pulled or restricted. Fair Housing scanner goes live on listing-content workflows.

Days 22 to 30. First monthly audit runs. Designated broker reviews findings; tunes policy based on what surfaces.

The full governance scaffold ships as part of every AI enablement engagement.

Insurance and E&O implications

Three patterns are emerging in 2026 real estate E&O underwriting:

• AI use is not yet a coverage exclusion. Carriers are asking about AI on renewal but have not denied coverage. Document the policy and disclose AI use accurately.
• TCPA exclusions are tightening. Most policies already exclude TCPA violations; AI dialers and SMS make this exclusion more relevant. Verify TCPA compliance with documented consent records.
• Fair Housing claims trigger underwriter review. A single Fair Housing claim — AI-generated or not — triggers underwriter scrutiny on renewal. The Fair Housing scanner plus human review documentation is the strongest defense at renewal.

Pitfalls to avoid

Do not paste client PII into free-tier AI. Free ChatGPT, free Claude, and personal Gemini tiers do not have brokerage-acceptable data handling. Enterprise-tier tools only for any client-specific work.

Do not skip the AffBA disclosure on AI lender or title referrals. RESPA Section 8 violations are expensive. Template-level disclosure language is the only durable fix.

Do not assume the AI is right on contract interpretation. AI parses dates accurately. AI interprets clauses inconsistently. The agent or broker owns interpretation.

Do not let agents pick their own tools without review. Shadow IT inside a brokerage is a Fair Housing risk. Approved tools list plus monthly audit closes the gap.

Do not skip the disclosure conversation with clients. Clients accept AI use readily when disclosed; they react badly when surprised. The representation agreement addendum is a 30-second conversation that prevents 30-minute conversations later.

FAQ

Q: Is AI use disclosure mandatory yet? A: As of mid-2026, no federal mandate and only emerging state mandates. NAR has issued advisory guidance but not required disclosure. Best practice is the one-line representation-agreement addendum regardless.

Q: Does the Fair Housing scanner catch everything? A: No. Automated scanners catch 75 to 90% of protected-class language patterns. The remaining 10 to 25% requires human review. Both layers are mandatory.

Q: What if an agent generates AI content that violates Fair Housing? A: The agent of record is accountable; the brokerage is supervisorily accountable. The policy plus audit log is the brokerage's defense. The fix is retraining and tighter tool configuration.

Q: How do I handle AI use at recruiting time? A: Disclose AI use during the recruiting conversation. Agents joining the brokerage need to know the approved tools list, the listing-content review process, and the audit cadence on day one.

Q: What's the right cadence for policy review? A: Annual minimum. We recommend a quarterly check-in for the first year as state rules tighten, then annual review afterward.

Q: Do I need different policies for different states? A: One core policy plus state-specific addenda. The eight sections above apply everywhere; addenda cover state-specific disclosure, telemarketing, and source-of-income rules.

---

If you want a brokerage AI policy scoped against your specific states, agent count, and existing tools — reach out. We will draft the policy, configure the audit cadence, and ship the rollout in 30 days. The full engagement model is on the AI for real estate overview.

SOURCES

Cited and consulted.

1. 01NAR REALTOR Magazine — Practice and Technologynar.realtor · accessed May 8, 2026
2. 02Inman — Real Estate Technology Coverageinman.com · accessed May 8, 2026
3. 03RISMedia — Best Practices for Real Estate Brokeragesrismedia.com · accessed May 8, 2026
4. 04HousingWire — Brokerage Operations Coveragehousingwire.com · accessed May 8, 2026