AI and Fair Housing: What Real Estate Agents Need to Change Now

AI is already part of many real estate workflows, and the Fair Housing question is shifting from “Should agents use it?” to “Can they use it without creating liability?”

That was the focus of a recent Wisconsin REALTORS® Association webinar featuring Dr. Michael Akinwumi, chief AI officer at the National Fair Housing Alliance (NFHA). His central warning was straightforward: an AI vendor does not inherit an agent’s license or Fair Housing responsibilities.

The agent remains accountable for the recommendation, message, or marketing material that reaches a client.

For working agents, this does not mean abandoning AI. It means treating AI output as draft material that requires professional review—especially when the task involves neighborhoods, schools, safety, crime, demographics, or buyer preferences.

The risk is often in interpretation, not the software name

Akinwumi’s point is important because many agents are looking for a “Fair Housing-safe” AI tool. That may be the wrong question.

Large language models generally do not need an explicit instruction to discriminate. They can infer meaning from seemingly ordinary phrases. “Good schools,” “safe area,” and “low crime” may sound practical, but they can become proxies for protected characteristics when used to rank or describe communities.

The risk increases when an agent adds personal information about a client, such as race, ethnicity, religion, family composition, disability, lifestyle, or current location. The system may use that information to personalize results in ways the agent did not intend—and may not notice.

This is the difference between personalization and steering. A recommendation can feel helpful while still directing a buyer or renter toward—or away from—certain communities.

How common AI assistants compare

No general-purpose assistant should be treated as a compliance officer. Their usefulness depends more on the workflow around them than on the brand.

| Tool category | Typical cost | Useful for agents | Main Fair Housing concern | |---|---:|---|---| | ChatGPT Plus | $20/month | Drafting emails, listing copy, checklists, role-play | May infer sensitive meaning from prompts and uploaded client details | | Claude Pro | $20/month in the U.S. | Rewriting, summarizing, reviewing language | A polished explanation can still encode unsupported neighborhood assumptions | | Google AI Pro | $19.99/month in the U.S. | Drafting, research assistance, Google Workspace workflows | Connected data and personalization may add context an agent did not intend to use | | MLS or brokerage AI | Varies; often bundled or priced per user | Listing fields, lead follow-up, search workflows | The brokerage may control the system, but the agent still controls many inputs and outputs |

The practical cost is not just the subscription. If an assistant saves 30 minutes a week, $20 per month may be easy to justify. If an agent uses it for neighborhood recommendations, the cost of a mistaken output includes review time, retraining, reputational damage, and potentially legal exposure.

ChatGPT and Claude may be reasonable drafting tools, but neither should be positioned as a Fair Housing safeguard.

Skip this if...

Skip a paid general-purpose AI subscription if you only need occasional email cleanup and already have an approved brokerage tool. The extra monthly fee will not solve a compliance problem by itself.

What agents should change immediately

First, separate objective property facts from subjective neighborhood labels.

Instead of asking an AI system to “find a safe neighborhood with good schools for this family,” use a narrower request: “Summarize the verified property features, commute information, taxes, and publicly available school-district data. Do not rank communities by safety, desirability, or demographic characteristics.”

Even then, verify the underlying information independently. AI can fabricate sources, confuse jurisdictions, or present outdated data with confidence.

Second, stop putting unnecessary client identity information into prompts. The model usually does not need a buyer’s race, religion, family status, disability, or lifestyle to draft a showing itinerary or explain a transaction step.

Third, ask for the basis of a recommendation—but do not confuse an explanation with proof that the output is lawful. An AI-generated rationale can simply be a more detailed version of the same unsupported assumption.

A better review asks:

• What factual input produced this recommendation?
• Is the information current and verifiable?
• Does the language rank one community against another?
• Could “safe,” “family-friendly,” “exclusive,” or similar wording act as a proxy?
• Would I give the same answer to a different client profile?
• Can I replace the subjective claim with neutral, public information?

The NFHA testing idea is useful for brokerages

The NFHA has been testing AI responses with different buyer and renter profiles, looking for changes in language or recommendations when race, ethnicity, location, or transaction type changes.

That testing approach is more useful for a brokerage than simply asking an AI vendor whether its product is “compliant.” A small internal audit could use identical housing questions with neutral test profiles, then compare the outputs.

Brokerages should document:

• Which AI products agents are allowed to use
• What client data may be entered
• Which prompts are prohibited
• Who reviews AI-generated marketing and recommendations
• How questionable outputs are reported and retained

This is also where a brokerage can spend money more intelligently. At $20 per user per month, a team with 20 agents might spend roughly $400 per month on individual subscriptions, but should budget separately for compliance review and staff training.

Who needs to pay attention—and who can mostly ignore it

Agents using AI only to correct grammar, turn notes into a follow-up email, or create a transaction checklist face a narrower risk profile, provided they remove confidential information and check the result.

The issue is much more urgent for agents and teams using AI to:

• Recommend neighborhoods
• Target housing advertisements
• Score or prioritize leads
• Describe schools, crime, safety, or community character
• Personalize property searches
• Generate replies to questions about protected characteristics

The bottom line is not “never use AI.” It is “do not outsource judgment.”

Use AI for administrative acceleration and first drafts. Keep housing recommendations tied to objective, verifiable criteria, offer consistent information to clients, and make the final decision as a trained real estate professional. That is the change this news should prompt in your business today.