Leading Your Office Into AI: what to automate, what to protect, and how to find tomorrow's sellers.

AI is not a tech project. It is a margin project.

In 1998 the edge was a pager, a fax machine and a thick MLS book. The agents who won were the ones who called the most people and remembered them. That has not changed. What changed is the cost of remembering, following up and showing up prepared. It just dropped close to zero.

So the question for a broker is not which AI tool to buy. It is which hours in your office produce money, which hours only support the money, and which hours are pure admin that a machine should carry. Start there and every tool decision gets easy.

Have every agent write down what they actually do all week.

Not what they think they do. A real list, with hours. Then sort each task into one of three buckets. The list is the AI plan. Nothing gets automated that is not on it.

Worksheet

Fill it in here or print it. It saves in your own browser only. Nothing is sent anywhere.

The big tools are converging. Pick by the task, not the logo.

Claude, ChatGPT and Gemini now do most of the same things: write, read documents, search the web, remember instructions, and work inside projects. The differences are real but small, and they change every few months. Do not make your office policy depend on one brand.

• Writing in your voice and reading long documents: test all three on the same disclosure packet or listing and keep the one your agents like.
• Already living in Google Workspace or Microsoft 365: the built-in assistant (Gemini or Copilot) may be the cheapest win, because it is already where your files are.
• Client data of any kind: only on a business or team plan where training on your data is off and an admin controls the accounts. Free personal accounts are for public information only.

Six questions to ask any vendor, in writing, before client files go in: Is data encrypted? Can we set retention and delete on demand? Is our data used for training? Can an admin manage and remove users? Are actions logged? Where is the data stored?

You would not hand a new hire your Platinum card on day one. Do not hand it to an AI either.

Access is earned in steps. A new tool, or an AI that can connect to email, calendar, CRM or files, starts at the bottom and moves up only after a clean month.

Ten minutes of setup that makes every answer better.

Every major assistant lets you save standing instructions so you stop repeating yourself. Set it once per agent, and the AI shows up knowing who it works for, what market it is in, and what it is never allowed to do.

Tomorrow's sellers are already in your database. AI helps you find them first.

The average agent has a few hundred names. A busy one has 2,000. Most of those people have not heard from the agent in a year. That is the cheapest lead source in the business, and it is the one AI is best at working.

1. Clean the list. Export your CRM. Have AI find duplicates, missing fields, and past clients with no contact in 12 months. Do this inside an approved business workspace, not a free account.
2. Add public records. County assessor and recorder data is public: years owned, owner-occupied or not, property type. Long ownership plus life change is where listings come from.
3. Add market context. Use your MLS's market statistics and your own sold history to tell each owner what their neighborhood did this year. That is a reason to call, not a pitch.
4. Score by property, never by person. Rank by years owned, estimated equity, property type and past relationship. Never by age, family, ethnicity, religion or any protected characteristic, and never by guesses about those.
5. Run a resurrection campaign. A real, personal note to people who have not heard from you: what their home did, what you are seeing, an offer to help. AI drafts it. The agent edits it and sends it.

Post when your town is actually looking.

• Timing by area: your own social insights show when your followers in each area are online. Ask AI to read that export and build a posting calendar around it.
• Where people talk: local Facebook groups, Nextdoor, school and sports parent groups, HOA newsletters. Follow each group's rules. Be useful first.
• Local vendors and events: AI can build a monthly list of farmers markets, fundraisers and grand openings in your zip codes. Show up, sponsor, or feature local businesses. Relationships beat ads.

Reviews: the safe way to show your best

Featuring great reviews in your marketing is fine. Hiding bad ones is not. Under the FTC's 2024 rule on consumer reviews, a reviews page cannot suppress negative reviews while implying it shows them all, and civil penalties run above $50,000 per violation. Google bans review gating outright: you may not discourage negative reviews or selectively ask for positive ones.

Title and escrow partners: keep it clean

Title companies can still teach a class, sponsor an event, or give you standard property profiles. What creates RESPA Section 8 risk is a title company doing work an agent would otherwise pay for, like building your database, running your marketing, or setting up your AI, in exchange for referrals. Paying an expense the agent would otherwise carry is a thing of value under Regulation X, and violations can bring fines, jail time and triple damages. If it saves you money and comes with an expectation of business, ask your broker or counsel first.

The rules that do not move.

• Altered listing photos: since January 1, 2026 (AB 723), a digitally altered listing photo needs a disclosure on or next to the image and a link or access to the original. Virtual staging and AI edits included.
• Bots tell people they are bots: California law requires disclosure when a bot is used to influence a sale online.
• Fair housing: HUD withdrew its 2024 guidance on digital ads in 2025, but the Fair Housing Act and California's broader FEHA did not change. No neighborhood descriptions based on who lives there, and no ad targeting by protected class, even when an AI suggested it.
• Confidentiality: client information goes only into an approved business workspace with training off.
• Supervision: what an AI produces under an agent's name is that agent's work and the broker's responsibility. A human reads it before it goes out.

This page is general education, not legal advice. Check current rules with your broker, your MLS and your counsel before you act.

Some work should never leave your building.

Free, open AI models now run on a good desktop or a small office server. Tools like Ollama and LM Studio install them in minutes. Nothing you type leaves the machine, and there is no monthly bill.

• Good fits: cleaning up a database export, summarizing internal training notes, sorting old files, first drafts of internal documents.
• Still use the big business plans for: the best writing quality, web research, and anything your agents need on their phones.
• The office pattern: public information can go anywhere. Client information goes to the business plan with training off, or stays on the local machine.

A one-page policy your agents will actually read.

Copy it, put your office name on it, have counsel review it, and have every agent sign it. Revisit it every quarter. It is a starting point, not legal advice.