A brokerage owner in the Pacific Northwest recently set up an outbound AI voice agent to work a list of 300 expired listings. The workflow made sense on paper: the agent would call each number, introduce the brokerage, ask whether the seller was still interested in listing, and route warm responses to agents. The campaign was ready to launch on a Monday morning. Nobody had reviewed the script for Fair Housing compliance.
That gap is more common than it should be. Most brokerages think of Fair Housing as a listing-description problem — watch the adjectives, stay away from protected-class language, done. Outbound voice scripts carry the same risk, and in some ways a harder one: the words are spoken at scale, often to homeowners in specific zip codes, and the call logs may be the only record of what was said.
What Fair Housing risk looks like in a voice script
Fair Housing law prohibits representations, statements, or conduct that discriminate based on race, color, religion, national origin, sex, familial status, or disability. An AI voice script can trip one of those wires without ever using an obvious slur or prohibited phrase.
Consider how an AI drafts an introduction. If the prompt used to generate the script references the neighborhood name, the agent building the workflow may have inadvertently trained the AI to use language that signals who it expects to be calling. If the script uses different qualifying questions depending on which list segment the lead came from — and those segments were cut by geography in a way that tracks demographic lines — the pattern of questions across calls may constitute steering, even if no individual call sounds discriminatory.
A subtler risk: tone and persistence. A voice script that follows up more aggressively on some calls than others, without a neutral, documented rationale, can draw a complaint if it correlates with a protected class. The AI does not know it is doing this. The brokerage is still responsible.
The audit before the launch
Before any outbound voice campaign goes live, the script needs a structured review. This is not a lengthy process, but it has to be deliberate.
Start with the source list. How was it segmented? If the list was filtered by neighborhood, zip code, or any variable that could be a proxy for national origin or race, document why that filter exists and whether it is neutral on its face. Expired listings filtered by property type and price range are defensible. Expired listings filtered by the name of a specific subdivision that happens to be predominantly one national origin is a different matter.
Next, read the script out loud — all variations of it, including how the AI handles a no-answer, a callback, and an objection. Look for any language that describes the neighborhood, references community characteristics, or makes assumptions about the seller's situation. Remove all of it. The script should be about the property and the transaction, not the location's character or the seller's circumstances.
Then look at the qualifying questions. Are they the same for every call? If the agent asks different questions based on lead source, price range, or any other variable, confirm that the branching logic is documented and neutral. A broker or attorney should review any branching that could correlate with protected class status.
What belongs in the call log
Once the campaign runs, the logs need to be kept in a retrievable format. That means not just whether the call was answered, but the script version used, the call date and time, and any branch the AI took during the conversation. If a complaint arrives six months later, the ability to show exactly what the AI said on a specific call — and prove it matched an approved script — is the difference between a quick resolution and a protracted investigation.
This is the practical argument for treating AI voice scripts like any other advertising material: document the version, track the approval, and keep the record. Say a team makes 1,000 outbound calls a month. Without versioned scripts tied to call logs, there is no way to demonstrate that a call made in March used the approved language and not an earlier draft.
Real Estate AI Group's approach on every voice agent build is to wire the call log directly to the compliance record: each call references the script version that was active at the time. That connection should be built in before the first call goes out, not reconstructed afterward.
Where to start
Pull the script your current or planned outbound voice agent uses and read it as if you were a compliance officer seeing it for the first time. Flag any language that describes the neighborhood, makes assumptions about the seller, or varies based on lead segment. Then have your broker or a real estate attorney review those flagged sections before the campaign runs. That single review, done before launch, is the most cost-effective Fair Housing step available to any brokerage deploying AI outbound calls.