Is It Safe to Put a Purchase Contract Into an AI Tool?

Short answer: sometimes, and the conditions matter more than the answer does. A purchase contract can go into an AI tool when three things are true — your brokerage policy permits it, the account you are using does not train on your inputs, and you have stripped the handful of fields that should never leave your office regardless. If any one of those is missing, the answer is no.

This is the question I get in every workshop, usually about ten minutes in, usually phrased as a joke by someone who is not joking. It deserves a straight answer rather than either of the two bad ones — “it’s fine, everybody’s doing it” and “never put client data in AI.”

I am not a lawyer and this is not legal advice. Your broker-in-charge and your brokerage’s counsel are the authority here, and their policy overrides anything below.

What is actually in the document you are about to upload

Before the tooling question, be honest about the payload. A South Carolina residential purchase agreement plus its usual companions contains:

  • Full legal names of buyers and sellers
  • The property address, which is also somebody’s home address
  • Purchase price, earnest money, and financing terms
  • Lender name and sometimes loan officer contact details
  • Occupancy and possession terms, which reveal when a house is empty
  • Occasionally, in the addenda, employment or income detail

That is a meaningful package of personal and financial information about two households. Most agents email it around all day without a second thought, which is its own conversation — but the fact that we are casual about it in Outlook does not make us right to be casual about it somewhere new.

The three conditions

1. Your brokerage policy has to permit it

Start here, not last. Brokerages are writing AI policies right now, and they range from “use good judgment” to a named list of approved tools to an outright prohibition. If yours has a policy, that is the answer, full stop. If yours does not have one yet, ask your broker-in-charge directly and in writing — partly because you need to know, and partly because the written reply is the thing that protects you later.

South Carolina license law imposes a duty of confidentiality to your client that does not expire when the transaction closes. A tool choice that leaks client information is a license problem before it is a technology problem.

2. The account has to not train on your inputs

This is the condition people get wrong, because it is not a property of the software — it is a property of which plan you are on. The same brand name can behave differently on a free consumer tier and a paid business tier.

Do not take my word for what any vendor’s current policy is; these change, and a blog post is the wrong place to learn them. Go and check three things yourself, today, in the account you actually use:

  1. Does this plan use my conversations to train models? Find the setting. Turn it off if it is on and can be turned off. If it cannot be turned off, that account is not suitable for client documents.
  2. How long is my data retained, and can I delete it? You want a clear answer and a delete button.
  3. Is there a business or team tier with different terms? Usually there is, and usually that is the one you want if you are handling client files.

The rough rule of thumb, which you should verify rather than trust: free consumer tiers are for your own work; paid business tiers are for client work. The gap between those two is where most of the risk in this whole subject lives.

3. You have to strip the fields that never leave your office

Even on a properly configured account, some things do not go in. Not because the vendor is untrustworthy — because you do not need them there, and the cheapest way to not lose data is to not have it in the first place.

My permanent redact list:

  • Social Security numbers
  • Bank account and routing numbers
  • Driver’s licence and passport numbers
  • Full dates of birth
  • Wire instructions of any kind, ever
  • Anything relating to a client’s health, immigration status, or criminal history

None of those are required for the tasks you actually want done. Extracting deadlines, summarising an inspection, drafting a repair request, building a timeline — none of it needs a Social Security number. If a field is not needed for the job, it does not get uploaded for the job.

In practice this takes about ninety seconds with a PDF editor, and it is the single highest-value habit in this entire subject.

Free guide

The longer version of this is in The Real Estate Agent’s Guide to Claude & Cowork — 26 pages, written for South Carolina agents, with twenty-six prompts you can copy and use. No charge. Download it here.

The test I actually use

Before anything client-related goes into any tool, I ask one question: would I be comfortable explaining this exact upload to this exact client, in plain words, if they asked?

“I used a secure AI assistant on a business account that does not train on our documents, to pull the deadlines out of your contract into a calendar, and I checked every date against the contract myself” is a sentence I can say to a client without flinching.

“I pasted your contract into a free chatbot to see what it would say” is not.

If you cannot say the sentence, do not do the thing. That test is not legally binding but it has never once led me wrong.

What about the argument that you should never upload anything

It is a defensible position and some brokerages have taken it. It is also worth understanding what it costs.

The agent who will not upload a document is not avoiding AI — they are restricted to the shallow half of it, the part that writes listing descriptions from a text prompt. That is the low-value half. Meanwhile they are almost certainly still emailing the same documents unencrypted, storing them in a consumer cloud drive, and texting photographs of them from a phone that syncs to three places.

I am not arguing that two risks make a right. I am arguing that “never upload client documents to a third party” is a rule most agencies have already broken in a dozen quieter ways, and that a considered policy beats a blanket one that nobody actually follows.

A workable default

If you want one paragraph to operate from until your brokerage publishes something better:

Use a paid business-tier account with training turned off. Redact identity and banking fields before upload. Keep uploads to documents relevant to a transaction you are actually working. Verify every fact the tool returns against the source document. Do not upload anything you would not be able to explain to the client it belongs to. And ask your broker-in-charge for the policy in writing.

That is not a bulletproof compliance program. It is a reasonable, defensible practice, which is the realistic standard for a working agent.

Where the detail lives

The free guide I wrote for agents has a full chapter on this — the redact list, the account settings to check, the Fair Housing overlap, and what goes wrong when people skip it. It is 26 pages, there is no charge, and Chapter Seven is the one that covers this subject properly.

Download the guide.


Barry Cunningham runs Lowcountry AI Workshop and holds a South Carolina real estate license. Nothing here is legal advice; your broker-in-charge and your brokerage’s counsel are the authority on your obligations. If you want to talk it through, the fit call is free.