AI Policy
Version 1.0, 4 September 2026. BrokerDIY is an AI platform for a regulated industry, sold to licensed professionals who owe duties to their clients. This page says what our AI does, what it must never be trusted to do on its own, and what we hand you so you can meet your own obligations. It forms part of our Terms of Use.
1. The rule everything else follows from
BrokerDIY does not make decisions about people. You do.
Our AI drafts, summarises, ranks, transcribes, estimates and suggests. Every output is a proposal to a licensed human who reviews it and decides. Nothing in the platform is designed or permitted to approve, reject, price, screen or select a person automatically, and anything touching money, contract terms, or whether someone gets a home is yours to approve before it reaches them.
That is not a disclaimer bolted on afterwards. It is the design, and the rest of this page is what we do to keep it true.
2. Where AI is used, and what it does
| Where | What it produces | Who decides |
|---|---|---|
| Assistants and drafting | Email, listing copy, summaries, task suggestions | You, before anything sends |
| Voice agents | Answers and places calls, transcribes, books, routes | You set what it may say and do; a human takes anything consequential |
| Document intelligence | Extracted fields, checklists, missing-item flags | You, before a transaction relies on it |
| Valuations and market analysis | Estimates, not appraisals | You, and a licensed appraiser where one is required |
| Lead and pipeline ranking | An ordering suggestion for your own follow-up | You. It orders your work, it does not qualify a person |
| Marketing and campaigns | Creative, audiences, budget suggestions | You, subject to section 5 |
3. Documentation for you as a deployer
Automated-decision laws are arriving that put duties on the business using an AI system, not only on the company that built it. Colorado’s SB 26-189 takes effect 1 January 2027 and covers technology that produces predictions, recommendations, classifications, rankings or scores used to make, guide or assist decisions about a person, including in housing. Several other states are moving the same way.
Where that applies to you, you will need documentation from the developer. This section is that documentation, and you do not have to ask for it.
- Intended uses. Drafting and summarising communications; transcribing and handling calls; extracting fields from transaction documents; producing market estimates; ordering your own follow-up queue; generating marketing creative. All as assistance to a licensed human who decides.
- Uses we do not support, and which are outside the intended use. Automatically accepting or rejecting an offer, applicant or client; scoring a person’s tenancy, creditworthiness or suitability; setting or varying price or terms per person without human review; selecting or excluding audiences by a protected characteristic or a proxy for one.
- Categories of training data. We do not train foundation models. Generation runs on third-party models supplied by the providers on our sub-processor list, and on open-weight models we host ourselves. Where we tune or evaluate our own components, the only material used is Aggregate Data as defined in section 8 of the Terms — statistics carrying no names, addresses, document contents or client information, unattributed to any brokerage and not reversible. One brokerage’s data never trains a system serving another, and our providers are contractually barred from training their general models on your content.
- Known limitations. Generative output can be confidently wrong, incomplete or out of date. Extraction misreads unusual layouts, handwriting and poor scans. Valuations are estimates and are not appraisals; they vary by market and by data availability. Speech recognition degrades with accents, background noise and poor lines, and a voice agent can mishear a number. Any generative system can reproduce bias present in its training data or in a prompt. None of it is legal, tax, appraisal or compliance advice.
- Human review. Review before an output reaches a client, a counterparty or a public audience. Check any figure, date, address or contract term against the source document. Treat an extracted field as a draft until you have seen the page it came from. Keep a person in the loop on anything that affects whether someone gets a home or on what terms.
- Material changes. If we change a model or a feature in a way that materially affects output or these limitations, we will tell you before it takes effect for you.
- Records. We retain the logs and documentation needed to support your record-keeping for at least three years.
4. Telling people they are talking to an AI
Our voice agents identify themselves as an automated assistant, and the platform is configured so that they do so by default. Do not turn that off, and do not configure an agent to claim to be a named human. A growing number of states require disclosure when a consumer interacts with generative AI, some specifically in regulated occupations, and several require notice or consent before a call is recorded.
You are responsible for the disclosures and consents your jurisdictions require, as set out in section 4 of the Terms. We give you the controls; we do not determine what the law requires of you, and we do not review your scripts.
5. Fair housing and algorithmic discrimination
Fair housing law applies to a machine exactly as it applies to a person, and “the tool did it” is not a defence for either of us.
- Do not use our tools to target, filter, exclude or rank people by race, colour, religion, sex, disability, familial status or national origin, or by any other protected characteristic, or by a proxy for one — and postcode, school rating, language and neighbourhood description can all act as proxies.
- Audience selection built with our tools must comply with fair housing law and with the advertising platforms’ own housing-category rules, which restrict targeting for a reason.
- We test our own generated marketing copy against fair-housing patterns and block or flag output that trips them. That is a safety net, not a substitute for your review.
- We may suspend an account we reasonably believe is being used otherwise, as section 11 of the Terms says.
6. What we do with your data in AI features
When a feature calls a model, the content of that request is processed to answer it and is not retained by the provider to train on. Your data stays isolated to your brokerage. The full position, including the narrow licence we take and the Aggregate Data right that survives your account, is in our Privacy Policy and section 8 of the Terms. Nothing in this page widens it.
7. If the AI gets something wrong
Tell us through our contact page. Include what you asked for and what came back, and we will look at it — this is the fastest way we learn about a failure mode, and we would rather hear it than not.
If a person believes something in your brokerage was decided about them unfairly, that request goes to you, because you made the decision and you hold the relationship. We will give you what we hold so you can answer them, including what the platform produced and when.
8. Changes
AI regulation is moving quickly and this page will move with it. The version and date at the top tell you what you are reading, and we will tell you before a material change takes effect for you.