Key takeaways
- Your agents are contractors, not employees, which changes how you deploy anything. Provide it, do not mandate it.
- Fair housing language is the sharpest risk. Models reproduce decades of property copy including the phrasing that is no longer acceptable.
- The biggest measurable gain is enquiry response time, not listing copy.
- Keep it away from valuation and contract interpretation. Both carry licensing consequences.
Most articles about AI in real estate start with listing descriptions, because listing descriptions are the most obvious thing an agent writes. They are worth doing and they are not where the money is. The money is in the twenty minutes between an enquiry arriving and somebody responding to it, and in the administrative load that sits between a signed contract and a closing.
There is also a structural problem specific to this industry that nobody else has to solve, and it is worth addressing before anything else.
You cannot mandate a tool to people you do not employ
A brokerage of eighty agents does not have eighty employees. It has a small operations and transaction team, and eighty independent contractors who choose their own tools, run their own businesses, and are deliberately not directed in how they perform their work. That structure exists for good reasons and interfering with it has consequences well beyond software adoption.
So the deployment model that works in a law firm or an agency does not transfer. You cannot require agents to use a tool, and you should be careful about instructing them in how to do their work generally.
What brokerages do successfully instead is treat it as a recruiting and retention benefit rather than an operational mandate. The brokerage provides the account, provides the training, and provides the templates. Agents use it or do not. Meanwhile the brokerage's own staff, who are employees, use it fully in transaction coordination, marketing and administration.
The rules that do apply attach to output rather than method. Anything published under the brokerage name, on the brokerage's listings, or in the brokerage's advertising is subject to the brokerage's standards regardless of how it was produced. That is a normal supervision position and it does not touch classification.
The framing that gets adoption in a brokerage: agents respond to anything that helps them win listings and close faster, and ignore anything that sounds like head office adding process. Present it as a tool that answers leads faster than the agent down the road, not as a brokerage initiative.
Fair housing is the risk nobody plans for
This is the one to take seriously. Generative models have been trained on an enormous quantity of historical property marketing, and historical property marketing contains exactly the language that fair housing law exists to prevent. Ask for a warm, appealing description of a three bedroom house near a school and you will frequently get copy describing who would enjoy living there.
The problematic output is rarely crude. It is phrasing that characterises the likely occupant or the character of a neighbourhood rather than the features of the property: references to family suitability, to a quiet area for certain residents, to walkability framed around a demographic, to religious or community facilities as selling points, to safety in a way that carries implication. Each of these reads as ordinary marketing copy and each of them is the category of statement fair housing guidance addresses.
The control is simple and non-negotiable. Every generated description gets a fair housing read before publication, by a person who has been trained on what to look for. Give agents a short list of what to strike:
- Any description of who the property suits rather than what it has.
- Any characterisation of the neighbourhood's people rather than its amenities.
- Any reference to schools framed as suitability rather than as a factual proximity.
- Any language about safety, quietness or community that implies a group.
- Any reference to religious institutions, ethnic character or family composition.
Then instruct the tool to describe the property only. That single instruction removes most of the problem at the source, and the human read catches the rest.
Where the time actually comes back
| Task | Fit | Note |
|---|---|---|
| Drafting responses to inbound enquiries | Strong | Speed of first response is the highest value variable in the business |
| Transaction coordination correspondence and chasing | Strong | Employee staff, high volume, repetitive, immediately checkable |
| Summarising long client email threads into next actions | Strong | Removes the reading, keeps the judgement |
| Listing and marketing copy first drafts | Good | Requires a mandatory fair housing read |
| Turning market statistics you already hold into client commentary | Good | You supply the data, it supplies the narrative |
| Listing presentation and proposal preparation | Good | Structure and wording, not the pricing argument |
| Social and newsletter content from your own listings | Good | Watch for the same flattening every other industry sees |
| Explaining contract terms to a client | Unsuitable | Unauthorised practice of law exposure in most states |
| Valuation or pricing recommendations | Unsuitable | Licensing and appraisal rules, and the model has no local data |
| Disclosure documents | Unsuitable | Statutory content with no margin for a plausible approximation |
Response time is the whole argument
Every brokerage owner already knows that the agent who replies first has an enormous advantage, and that enquiries arriving at nine on a Sunday evening frequently sit until Monday. That is the gap worth closing, and it is the one AI closes most straightforwardly.
The version that works is not an autonomous bot answering leads unsupervised, which carries fair housing exposure and sounds exactly like what it is. It is a drafted reply, specific to the property and the question asked, sitting ready for the agent to read, adjust and send from their phone in forty seconds instead of composing it in six minutes.
That difference converts a Monday morning reply into a Sunday evening one. Owners who track lead conversion find that this single change moves a number that nothing else in the technology stack has moved for years.
What not to do: let it send. An unsupervised reply to a housing enquiry is a fair housing statement made by your brokerage with no human involved, and the potential downside is not proportional to the six minutes saved. Draft and review is the right setting here and probably always will be.
The transaction team is the quiet win
Transaction coordinators are employees, work in high volume, and handle correspondence that is almost entirely repetitive: chasing documents, confirming dates, updating parties, following up lenders and title, summarising where each file stands. This is the least glamorous part of the brokerage and it responds better to AI than anything else in it.
A coordinator handling forty files can realistically recover several hours a week on correspondence drafting and file summarisation alone, and unlike agent adoption it is fully within the brokerage's control to implement, train and measure. If you want a defensible number for an owner, this is where to get it.
What to configure
- A brokerage business account covering employees fully, with agent seats offered as a benefit.
- Training on inputs disabled and retention set, since transaction files contain financial and identity information.
- A spending ceiling, particularly if image generation gets used in marketing.
- Administrator-managed access, so agent departures remove access the same day.
- A one-page fair housing standard for anything generated and published.
- Separate task guidance for agents and for transaction and marketing staff, because the useful tasks barely overlap.
Frequently asked questions
Can AI write our listing descriptions?
As a first draft that a person edits, yes. The specific risk is fair housing language, because models trained on historical property copy reproduce phrasing that describes the likely occupant rather than the property. Instruct it to describe features only, and require a fair housing read before anything is published.
How do we roll this out to independent contractor agents?
Provide it rather than mandate it. You cannot direct how a contractor performs their work without affecting classification, so brokerages offer the account and the training as a benefit and let adoption happen voluntarily. Apply your standards to published output instead, which is ordinary supervision and does not touch the classification question.
Where is the biggest return?
Enquiry response time and transaction coordination. Faster first responses move lead conversion more than anything else available, and transaction coordinators are employees doing high-volume repetitive correspondence, which makes that the easiest place to implement, train and measure.
Can we let AI answer leads automatically?
Not unsupervised. An automatic reply to a housing enquiry is a statement made by your brokerage with no human involved, which puts fair housing exposure on an unattended process. Draft and review keeps almost all the speed advantage with none of that risk.
Answer enquiries faster than the brokerage down the road
We choose the provider, configure retention and access, cap the spend, and write separate task guidance for your agents and your transaction and marketing staff, including a fair housing standard for anything published. Fixed price, live in 30 days or less.