AI Privacy for Realtors: What to Turn Off Before You Paste

Hi folks,
You’ve got a client email to answer, three showings to organize, and an AI tool offering to tidy everything up in seconds.
Then your mouse hovers over Paste.
“Hang on. Where does this information actually go?”
Realtors have been asking me about this, and I’m glad they are. A client’s budget, personal circumstances, and negotiating position deserve more thought than “the chatbot seemed friendly.”
Let’s walk through the settings worth checking, what those settings actually do, and how to get useful help from AI without handing over the whole client file.
First, “turn off sharing” can mean several things
There usually isn’t one switch that stops every use of your information. You may be dealing with:
Model training: whether your content helps improve the provider’s AI models.
Memory and personalization: what the assistant remembers or uses to tailor replies.
Storage and review: how long information stays in the service and who may access it.
Connected apps and shared links: whether information can move to another service or be seen by someone else.
Turning off training addresses one of those questions. The provider still processes what you send to deliver the service, and other uses or retention may apply.
My starting rule: keep personal, financial, and confidential client information out of public or unapproved AI tools, even after changing the settings.
Where to find the privacy controls
These steps focus on personal accounts. If your brokerage provides your account, ask the administrator which rules apply to that workspace. Paying for an individual subscription does not, by itself, establish that it is approved for client information.
ChatGPT: turn off model improvement
Open your account menu and choose Settings.
Select Data controls.
Switch off Improve the model for everyone and confirm if prompted.
On mobile, open the sidebar, tap your profile, then find Data controls. When signed in, this choice applies across devices. It stops new conversations from being used for model improvement; it doesn’t delete existing chats.
Temporary Chat provides another option: while a chat remains temporary, it isn’t used for model improvement or saved in your history, and doesn’t create or update memories. A copy may remain for up to 30 days for safety. Check personalization separately; temporary does not necessarily mean no existing context is used.
One easy detail to miss: submitting response feedback on a personal account can allow the associated conversation to be used for training, even after opting out. Check voice and memory controls separately, too.
OpenAI says Business, Enterprise, and Edu workspace content is not used for training by default. Your organization’s access and retention rules still matter. OpenAI’s data controls guide · Training and feedback details
Gemini: turn off Keep Activity
On the Gemini website, open Settings & help.
Select Activity.
Find Keep Activity and choose Turn off, or Turn off and delete activity if appropriate.
With Keep Activity off, future chats aren’t used to train Google’s AI models unless you submit feedback. Google still retains those chats for 72 hours to provide and protect the service; safety-related human review can still occur. Previously reviewed material may be retained separately for up to three years.
Some connected features become unavailable. Work and school accounts have different arrangements, so check with the administrator. Google’s activity instructions · Gemini Privacy Hub
Claude: check the model-improvement setting
Open your profile menu and choose Settings.
Go to Privacy.
Turn off Help Improve our AI models.
For consumer plans, Anthropic says this stops new chats and coding sessions being used for future model training. It cannot undo training already underway or completed. Safety-flagged conversations may still be used for trust and safety purposes. Commercial accounts have separate terms. Anthropic’s current instructions
Using Copilot or another tool?
Check the exact product and account. Microsoft’s newer personal Copilot experience, for example, places controls such as Saved memories and One shared experience under Settings → Personalization. Turning them off limits those features; it should not be treated as a model-training opt-out. Older Copilot versions have different controls.
Use Microsoft’s current privacy guide for your version. For any tool, look for its own explanations of training, retention, feedback, connected apps, and business-account protections. A familiar brand name doesn’t tell you which terms your account uses.
Your client’s name is only one piece of the puzzle
Removing “Jane Smith” from a prompt helps very little if you leave her exact address, employer, purchase budget, and the reason she urgently needs to move.
For everyday drafting, start with a made-up scenario or placeholders. Add the real details afterward in your brokerage’s approved system.
Here are two prompts you can adapt:
Showing follow-up: “Draft a warm follow-up email after a property showing. Use [CLIENT NAME], [PROPERTY], and [NEXT STEP] as placeholders. Ask whether they have questions. Keep it under 100 words. Do not invent property facts, deadlines, or commitments.”
General buyer education: “Create a plain-English checklist of questions a first-time buyer could discuss with their real estate professional. Use a fictional example. Do not provide legal advice or assume local contract rules. Flag anything that needs local verification.”
AI can help with structure and wording without knowing a buyer’s bank balance. And “Client McClientface” is perfectly qualified to star in your next training exercise.
Keep identification documents, mortgage records, signatures, access codes, private negotiations, and confidential agreements out of unapproved tools. Check attachments, screenshots, filenames, and hidden document details before sharing anything. Swapping names alone does not guarantee anonymity.
For Canadian realtors, the obligations still apply
Canada’s privacy regulators say organizations using generative AI need legal authority for handling personal information, appropriate purposes, and safeguards. Where consent is the basis, it must be meaningful. They also encourage synthetic or de-identified information where personal information is unnecessary. Read the Canadian privacy principles.
PIPEDA applies to many commercial activities. Alberta, British Columbia, and Quebec have substantially similar private-sector privacy laws that may apply instead for activity within those provinces; cross-border handling can still bring PIPEDA into play. Have your brokerage confirm the applicable rules. See the federal/provincial overview.
Quebec: Additional privacy requirements may apply before using AI tools that handle personal information, particularly when information is sent outside Quebec. Your brokerage should assess the proposed use and any required privacy impact assessment or written agreement.
In Ontario, TRESA’s confidentiality requirements continue to apply when using AI. RECO’s guidance says not to enter personal, financial, or confidential client or transaction information into public AI platforms. It also says to obtain explicit written client consent before entering personal details or contract terms into AI platforms. Written consent does not override the need to follow applicable law, assess the tool, and comply with brokerage policies.
Read those obligations together: written consent is not a blanket permission slip for any tool. Review brokerage approval, vendor terms, security, retention, and the particular use before proceeding. Verify AI-assisted content before relying on it or sharing it. Read RECO’s AI guidance.
For consent to be meaningful, clients should understand what information will be used, why, which provider will receive or process it, and the relevant risks and consequences. Depending on the circumstances, information about processing outside Canada may also be relevant. Consent does not make an otherwise inappropriate use lawful.
An uploaded document may contain information about other people, not just your client. Your client’s consent alone does not necessarily resolve obligations owed to those individuals.
Your brokerage remains responsible for protecting personal information under its control, including when an outside provider processes it. Approval should include a review of the provider’s terms, safeguards, access, retention, and incident procedures.
Outside Ontario, check your own provincial regulator, privacy authority, brokerage policies, and board or MLS agreements. Publicly viewable listing information is not automatically yours to upload or reuse without restriction.
Reading this outside Canada?
United States: check applicable federal and state privacy requirements, real estate licensing duties, brokerage policy, and MLS licensing terms. NAR’s practical guidance recommends approved tools, minimum necessary data, and careful review of privacy settings. Review marketing for accuracy and fair housing concerns, too. Describe property features and avoid language that suggests who should live there. NAR’s AI guidance.
EU/EEA and UK: GDPR or UK GDPR duties may apply. Establish an appropriate lawful basis before processing personal data, explain the use, minimize the data, and assess relevant vendor, security, and international-transfer requirements. Consent is not the only possible lawful basis, and a training opt-out cannot replace this assessment. EDPB guidance · ICO guidance.
Elsewhere, start with your local privacy regulator and professional rules. There is no universal “compliant everywhere” setting.
A quick check before the next copy and paste
Account: Am I in the correct, brokerage-approved tool and workspace?
Information: Can I do this with fictional details or placeholders?
Settings: Have I checked training, memory, sharing, and retention separately?
Connections: Does this task need access to email, cloud files, a CRM, or a meeting recording? Limit access to what is authorized and necessary.
Review: Who checks the output, and who do I contact if something goes wrong?
Already pasted something sensitive? Stop sharing more and promptly tell your broker or privacy lead. Record what went where and when, then follow the incident process for containment, deletion requests, and any required notifications. Deleting the visible chat alone does not establish that every retained copy is gone. The brokerage must assess whether applicable law requires notification to affected individuals or a regulator, and whether records must be maintained. Not every mistaken upload automatically triggers a reportable breach; the assessment depends on the facts and applicable law.
Let’s make this easier for your team
Start with one tool your team actually uses. Check the settings together, agree on what stays out, and practice with a fictional client. Put the decisions somewhere people will find them before their next busy afternoon.
I help real estate teams work through these questions through practical AI training and safety consulting. Bring the tools you use and the tasks you want help with. We can work through useful examples and identify what needs your broker’s or legal adviser’s input.
This article provides general education, not legal advice or a compliance certification. Requirements depend on your jurisdiction, role, contracts, and use case. Confirm current provider terms and consult your broker, privacy officer, or qualified legal adviser before using AI with client information.
Images are AI-generated illustrations of fictional office scenarios.
.png)



Comments