ReddyGet a free demo
Menu

Is ChatGPT safe for client information? Seven steps to keep client data safe when you use AI

Updated Oct 30, 2026

Agents ask whether ChatGPT is safe for confidential information. The honest answer is that it depends on three things: what you paste, which plan you're on, and one setting most people never open. On a personal account, what you type can be used to train the model unless you turn that off, and on some tools, human reviewers can read a sample of chats. This guide gives you seven steps that cut how much of a client's private details reach an AI tool, while you still use AI for the drafts and the busywork. Each step says what you keep and what you hand off.

Step 1: Sort what you know about a client into three piles

Most of what you'd ask AI about is already public: the address, the listing facts, the sale history, the neighborhood. That can go into any tool. Private is the client's name, phone, and email, why they're moving, their budget, and the lowest price they'd take. That goes only into a tool your brokerage approved, and only when the task needs it. The last pile never goes into a chat tool: Social Security numbers, bank and account numbers, photo IDs, loan documents, passwords, and wire instructions. The NAR Code of Ethics (Standard of Practice 1-9) bars a REALTOR® from knowingly revealing a client's confidential information, and that duty doesn't end when the deal does.

You do

  • Deciding which pile each detail goes in

Hand off

  • Pulling the public facts, such as listing details and nearby sales, from live data instead of from the client's file

Step 2: Turn off training, or use a plan that doesn't train

In ChatGPT, the setting is "Improve the model for everyone", under Data controls in Settings, and it's on by default on Free, Plus, and Pro. A Temporary Chat isn't used for training, but OpenAI can keep it for up to 30 days. In Claude, the setting is "Model Improvement" in the privacy settings; with it on, Anthropic can keep your chats, de-identified, for up to five years. In Gemini, it's "Keep Activity": with it on, human reviewers can read some chats, and Google asks you not to enter anything confidential you wouldn't want a reviewer to see. The business plans, ChatGPT Business and Enterprise, Claude Team and Enterprise, and Gemini in Google Workspace, don't train on your work by default. Setting names change, so look for them in each tool's privacy settings.

You do

  • Checking the setting on every AI account you use

Hand off

  • Client work, on a business plan or a tool your brokerage approved

Step 3: Take the names out before you paste

Most tasks don't need to know who the client is. To answer an inspection email or plan a counteroffer, use a tool your brokerage approved, and paste the email with the names, phone numbers, and email addresses taken out. Write "the buyer" and "the seller" instead. Keep only the facts the task needs, such as the price, the dates, or the repair list. An address alone is public, but next to the price and the repairs it points to one client's deal, so leave it out when the task works without it. Put the names back in when you copy the draft into your own email.

You do

  • Taking out the names, the contact details, and anything else that points to the client

Hand off

  • The draft, written from the facts you left in

Step 4: Keep wire instructions out of every chat and every email

Wire fraud is where leaked deal details turn into lost money. The FBI says chat tools can quickly write official-sounding emails with directions to wire funds, and voice cloning can make a call sound like someone the buyer knows. In one case in the FBI's 2025 Internet Crime Report, buyers closing on a home got an email that pretended to be from their attorneys and wired more than $449,000. Tell every buyer, at the first meeting and again in writing before closing, that wire instructions don't change by email, and to call the title company at a number they already have, not one from the email, before they send money. If a wire goes wrong, call the bank right away and file a report at ic3.gov. In that case, the full amount was still in the account and the bank put it on hold.

You do

  • Warning every buyer at the first meeting and again before closing

Hand off

  • The written warning, drafted once and ready to send to every buyer

Step 5: Connect only what the work needs

AI tools can now connect to your email, your files, and your calendar. A connected tool can read what's in them: that's the point, and it's also the risk. Connect an account when a task needs it, sign in with your own login, and turn on two-step sign-in for your email. Every few months, open the list of connected apps in your Google or Microsoft account and remove the ones you don't use anymore.

You do

  • Deciding which accounts a tool can reach
  • Removing the apps you no longer use

Hand off

  • Reading and sorting the mailbox you did connect

Step 6: Read every draft before it goes out under your name

A leak doesn't need a hacker. A message sent to the wrong person, or a draft that carries one client's details into another client's email, does the same damage. Before you send anything AI wrote, check who it goes to, every name, and every number. A reply in a long thread can carry text the next person shouldn't see, so read what's quoted below your message too.

You do

  • Checking the recipient, the names, and the numbers
  • Sending it

Hand off

  • The first draft, waiting in your drafts folder

Step 7: Use the tools your brokerage approved

Ask your broker which AI tools are approved for client work, and on which plan. NAR's AI policy template for brokers says that if a tool has not been approved, you should assume it is off-limits for client or brokerage data, and that you enter only the minimum the task needs. If your brokerage has no list yet, ask for one, and keep client work on a business plan until it does.

You do

  • Asking your broker for the approved list

Hand off

  • Client work, in the tools on that list

Where Reddy fits

Reddy works from the email and records you connect, in the chat you already use. It writes email as a draft in your own mailbox, and the email goes out when you send it. Reddy doesn't use your transaction data to train general-purpose AI models.

Wire instructions, and the call to check them, stay with you, your buyer, and the title company.

Read next

Guides

Hand off the work

Get a free demo, and we'll map these steps to your business.

Put Reddy to workTry for free
See pricingPut Reddy to workTry for free