AI chatbot for real estate leads

Keep the first release administrative. Give every visitor the same intake path, use approved listing data, record what the system said, and transfer judgment calls to a licensed professional.

Put Fair Housing rules ahead of the prompt

HUD lists seven federally protected classes under the Fair Housing Act: race, color, national origin, religion, sex, familial status, and disability. State and local law may protect additional classes. The brokerage's counsel and fair-housing officer should define the applicable rules before anyone writes chatbot instructions.

Store the rules outside a free-form prompt. Identify allowed questions, prohibited uses, approved sources, required disclosures, handoff triggers, retention, complaint handling, and who can change the system.

The chatbot should disclose that it is automated. It should give a direct path to a person and offer an accessible alternative when the interface does not work for someone. HUD notes that federal civil-rights laws can require effective communication with people with disabilities through suitable aids and services.

Use one administrative intake path

Ask about the transaction, location the person names, property type, price or rent range, bedrooms, timing, requested tour, and preferred contact method. Route accommodation or communication requests to staff.

Do not ask about race, religion, national origin, sex, disability, family composition, or another protected characteristic to qualify, score, route, or suppress a lead. A visitor may disclose personal information without being asked. Avoid using it for ranking and route accommodation or legal questions to trained staff.

Use the same required fields and response rules for comparable enquiries. Do not change inventory, access, or lead score because a model infers a protected characteristic from a name, language, ZIP code, device, writing style, or prior behavior.

Answer neighborhood questions consistently

Questions about schools, crime, safety, and “good neighborhoods” need an approved response policy. The chatbot should not translate “good for families” into a demographic profile or select areas by a protected class.

HUD's April 24, 2026 letter says real-estate professionals may share crime-rate and school-quality information when it is provided equally and consistently and there is no intentional discrimination based on protected characteristics. That clarification does not authorize steering.

Name the neutral source and let the consumer decide. For schools, link to the district or state data. For crime, use a public agency source where available. Apply the same source policy to every user. A licensed professional should handle subjective follow-up.

Keep listing data under brokerage control

Answer from current brokerage-approved or MLS-authorized data. Separate verified listing fields from generated explanation. If price, availability, property features, or a showing time cannot be confirmed, say so and offer human follow-up.

NAR's 2025 statement on AI listing displays says MLSs should consider data authorization, participant control, and local IDX disclosure and display requirements. Local MLS rules vary. Verify that data flow, caching, display, and third-party model access comply with the brokerage's license.

Do not let the model construct its own inventory from memory. Save listing identifiers, filters applied, retrieval time, and the text shown.

Separate service from decision-making

The chatbot can schedule a tour, collect neutral criteria, explain the brokerage's process, and route a request. A licensed person should handle representation decisions, legal or financing advice, accommodations, complaints, subjective neighborhood recommendations, exceptions, and actions based on sensitive information.

If the brokerage uses automated scoring, document inputs, purpose, thresholds, owner, and appeal path. Exclude protected characteristics and inspect proxies.

The 2024 HUD digital-advertising guidance, now in HUD's archive, described risks such as denying information, steering, different conditions, and discriminatory delivery. It remains a useful risk checklist, but an archived document is not a substitute for current legal advice.

Test equal treatment

Use paired tests before launch. Submit materially equivalent enquiries that vary names, pronouns, language, assistive-technology needs, family references, and ZIP codes while property criteria stay constant. Compare listings, questions, lead score, route, response time, appointment access, refusals, handoffs, and errors.

Do not use real people's personal data for testing. Counsel should review the design because synthetic profiles can still encode protected-class scenarios.

NIST's Generative AI Profile recommends monitoring whether outputs are equitable across sub-populations, defining groups relevant to the use case, and using structured human feedback. Record failed tests, corrections, system version, approver, and release date. Repeat tests after model, data, routing, or material prompt changes.

Preserve the record and complaint path

Retain the input, approved data retrieved, response, listing identifiers, route, assigned person, timestamps, opt-out state, and system version under the brokerage's policy. Limit access and do not keep sensitive details merely because the model captured them.

Give users a visible way to correct information, request a person, ask for an accommodation, and report a concern. Complaints should stop ordinary automation and reach the designated compliance owner with the transcript intact.

DATIK's Lead Chatbot covers approved knowledge, neutral intake, routing, and human takeover. The pricing page lists installation and operating fees. A free diagnostic can map one buyer or renter intake path with counsel and licensed staff before live listings are connected.

Sources and review limits

• HUD: Fair Housing Rights and Obligations
• HUD: April 24, 2026 Dear Colleague letter summary
• HUD archive: Digital-platform advertising guidance
• NAR statement on Zillow's app for ChatGPT
• NIST AI 600-1: Generative Artificial Intelligence Profile

Sources were checked on October 6, 2026. This is an operational guide, not legal advice. Federal, state, local, licensing, MLS, privacy, retention, and advertising rules can differ. Counsel and the compliance owner should approve the actual intake, data sources, tests, and release.