You can use ChatGPT for most of the writing and reading in a commercial real estate job — lease summaries, LOI drafts, market write-ups, tenant email — without leaking client data, and it comes down to one setting and one rule. The setting: put your firm on a paid business-tier account, where OpenAI does not train its models on what you type. The rule: decide, before anything goes in the box, which of your documents are public, which are merely internal, and which are restricted by an NDA — and keep the restricted ones out until that business account is in place. Almost everything principals fear about AI and confidentiality dissolves once those two decisions are made at the firm level instead of left to each broker’s guess. This is a practical guide to making them, and to the tasks where ChatGPT earns its keep.
What “Leaking” Actually Means (and the Two Fears That Aren’t Real)
Before you can use ChatGPT safely, you have to be precise about what you are actually afraid of, because most principals worry about the wrong thing.
The real risk is narrow: on a personal consumer account with default settings left on, the text you paste can be retained and used to help train future versions of the model. If a broker drops an NDA-bound offering memorandum into a personal ChatGPT account, the confidential contents have gone somewhere your firm did not authorize and cannot pull back. That is the thing to prevent, and it is entirely preventable.
Now the two fears that are not real. First, ChatGPT is not handing your deal to a competitor. Training a model on text does not store that text in a searchable form another user can retrieve; nobody is going to type “show me the seller financials Arthur pasted last Tuesday” and get them. The exposure is worth closing, but it is not a leak in the way brokers picture it. Second, using AI does not automatically breach your NDA. Most confidentiality agreements restrict disclosure to third parties, and a properly configured no-training account is, for routine work, no different from using Outlook or Dropbox — third-party services already holding your data. Confidentiality is a solvable configuration problem, not a reason to keep your team’s usage underground, which is exactly where it goes when left unaddressed.
The One Setting That Matters: Consumer vs. Business ChatGPT
Everything turns on which version of ChatGPT your firm is on, and the difference is not about capability. The models are the same. The difference is what happens to your data.
On a personal consumer account — Free, Go, Plus, or Pro — OpenAI trains on your conversations by default. You can turn that off (Settings → Data Controls → “Improve the model for everyone”), and a Temporary Chat is excluded from training and deleted within about thirty days. But those are per-user toggles that depend on every broker remembering to set them — not a control you can rely on across a team.
On a business-tier account — ChatGPT Team or ChatGPT Enterprise — OpenAI does not train on your inputs or outputs at all, by default, as a contractual commitment rather than a checkbox. That single fact is what makes confidential work defensible. The gap between Team and Enterprise is not training — neither one trains on your data — it is governance: Enterprise adds a formal data-processing agreement, audit logs, single sign-on, and custom data-retention controls, which most 4-to-20-person firms do not need on day one and can add later. Team, at a modest per-seat monthly cost, closes the actual exposure for a small firm.
The same pattern holds across the tools your firm might already pay for: the business tiers of Claude, the Google Workspace and Business versions of Gemini, and Microsoft Copilot with commercial data protection all carry equivalent no-training commitments. Because AI terms change every few quarters, confirm the current language on the vendor’s own business-data page before you standardize — but the principle is stable: the paid business tier is the safe one, the free personal account is the exposure. If your team is already using AI informally, and at most small firms it is, that quiet usage is happening on exactly the accounts you do not control. The fix is not to ban it; it is to move everyone onto one account you do. That shift, from accidental use to a deliberate firm capability, is the whole subject of the state of AI adoption in small commercial real estate firms.
A Simple Rule for What Can Go in the Box
A business-tier account handles the vendor side. You still need a rule your team can follow in three seconds, because “use good judgment” is not a policy. Sort every document into three tiers.
| Tier | What it is | CRE examples | Rule |
|---|---|---|---|
| Public | Already published or freely shareable | Marketing flyers, public listings, published market reports, generic email | Use freely on any account |
| Internal | Not secret, but not for outside eyes | Internal notes, draft copy with names removed, general strategy questions | Use on a business-tier account |
| Restricted | Bound by an NDA or clearly confidential | Offering memoranda, seller financials, rent rolls with tenant names, LOIs mid-negotiation | Business-tier account only, and only once you have confirmed the terms permit it |
The tiering converts a vague anxiety into a two-second decision. Most of the daily work a broker wants help with — polishing a market paragraph, drafting a listing, cleaning up an email — is public or internal and carries no meaningful risk on a business account. The restricted tier is small, and its rule is simple: business account, plus a quick check that the specific NDA does not prohibit third-party processing. When in doubt, strip the identifying details — swap “the Henderson portfolio at 400 Main” for “a three-property retail portfolio” — and ask about the anonymized version. Most of the useful prompt survives with none of the exposure.
What to Actually Use ChatGPT For in CRE
With the account and the rule in place, the question becomes where ChatGPT earns its keep. The answer is the language-heavy, repetitive parts of the job — which, at a small firm, is most of the week — and the strongest uses cluster by role.
| Role | High-value ChatGPT tasks | Why it lands here |
|---|---|---|
| Brokerage | Lease abstract first drafts, LOI drafting, listing and email copy, market write-ups | High document volume, low tolerance for blank-page time |
| Property management | Tenant notices and emails, maintenance triage summaries, meeting-note cleanup | Repetitive correspondence across a portfolio |
| Investment / acquisitions | Deal-teaser triage, offering-memorandum reading, first-pass market research | Fast filtering of far more inbound than a lean team can read |
The through-line is that these are all first-draft and summarization tasks, where a person reads the output before it goes anywhere. That is precisely the work a general-purpose tool does well, and the reason a small firm rarely needs specialized software to start — the same ChatGPT subscription covers all of it. A broker who pastes a forty-page lease and asks for the rent schedule, renewal options, and assignment clause gets a usable first pass in seconds, then still reads the lease — but starts from a structured draft instead of a blank page.
The skill that separates a broker who dabbles from one who gets real value is prompting — being specific about the role, the format, and the output you want. It is a learnable skill, not a personality trait, and it is the fastest lever a firm has. The structured version of building that fluency across a team is laid out in the CRE AI training playbook. And because these general assistants are increasingly able to carry out multi-step tasks on their own, it helps to understand what that means in plain terms, which is covered in what an AI agent is for property professionals.
Five Rules for Handling Confidential Deal Data
Once restricted documents are in play, five habits keep you safe. None requires a technologist.
-
One firm account, no personal logins for work. The moment brokers use personal Plus accounts for client work, you have lost the control that makes everything else safe. Standardize on one business-tier account and make it the only one used for firm business.
-
Strip identifiers when you can. For most analytical questions, the tenant name, exact address, and party names are not what you need answered. Anonymize the input and you get the same answer with none of the exposure.
-
Check the specific NDA before restricted data goes in. Not all confidentiality language is the same. For anything genuinely sensitive, spend two minutes confirming the agreement does not bar third-party processing. A no-training business tool is usually fine; the point is to check rather than assume.
-
Never let it touch the closing math unchecked. Confidentiality is only half the safety question. The other half is accuracy, and these tools will hand you a confident, wrong number. Any figure that goes into an offer, a reconciliation, or a client deliverable gets verified by a person.
-
Write it down once. A single page — which account, the three tiers, the “check the NDA” rule — is the difference between a policy and a hope, and it takes an afternoon.
These five habits turn confidentiality from the reason you avoid AI into a settled question you have already answered.
Where ChatGPT Gets Commercial Real Estate Wrong
Using ChatGPT safely is not only about data. The tool has real limits, and knowing them is what separates useful adoption from an embarrassing mistake in front of a client.
It is unreliable at exact math on money. Ask it to compute effective rent across a term with free-rent months and stepped increases and it will produce a clean, authoritative table that is sometimes wrong. Treat every number it gives you as a draft to check, and keep the arithmetic that determines an offer or a reconciliation in a spreadsheet or a purpose-built platform where the calculation is deterministic.
It fabricates specifics when it does not know them. Ask for comparable sales in a submarket and it may invent plausible-looking comps with real-sounding addresses and cap rates that do not exist; ask for a citation and it may manufacture one. This is the single most dangerous failure for CRE, because a fabricated comp in a client presentation is a credibility event. Use it for structure, drafting, and summarizing documents you provide — never as a source of market facts you have not independently confirmed.
These limits do not shrink the tool’s value; they define where a human stays in the loop — the same discipline that separates the small firms getting real gains from the ones stuck at dabbling, a theme that runs through the small-firm AI manifesto.
Setting Your Firm Up in an Afternoon
None of this is a technology project. A firm of ten can be set up correctly before the end of the day.
Pick one business-tier account for the whole team — the version that fits your existing setup, so a Microsoft 365 office looks first at Copilot, a Google Workspace office at Gemini, and a firm on neither defaults to ChatGPT Team. Move everyone onto it and retire the personal accounts for firm work. Write the one-page rule: the three data tiers, the restricted-data check, and the “verify the math” line. Then run a short session on the three or four tasks that fill your team’s week, so people start from good prompts instead of trial and error. That deliberate effort is what turns a subscription into a capability — the difference between a firm where AI is a private habit a few people have and one where it is a shared tool the whole office uses safely.
Where to Start
You do not need a technology strategy to get this right. You need to pick the account, set the rule, and point your team at the two or three tasks where fluency pays off fastest. A free AI-readiness assessment is a short working session that looks at your actual mix of brokerage, management, and acquisitions work, finds where your team is already using AI and where confidentiality fear is holding it back, and points you at the right tool and tier for your existing Microsoft or Google setup. Book a free AI-readiness assessment and you will leave knowing exactly how to use ChatGPT across your firm without a single restricted document going anywhere it should not.
Frequently Asked Questions
Is it safe to use ChatGPT for commercial real estate work?
Yes, when two conditions are met. Put your firm on a business-tier account — ChatGPT Team or Enterprise — where OpenAI does not train its models on what you type, and set a simple rule for which documents can go in. Most CRE writing and reading tasks — lease summaries, LOI drafts, market write-ups, tenant email — are safe on a business account. The exposure comes almost entirely from using a personal consumer account with default training settings on for confidential material.
Does ChatGPT train on the data I paste in?
It depends on the plan. On personal consumer plans (Free, Go, Plus, Pro), OpenAI trains on your conversations by default, though you can opt out in Data Controls or use a Temporary Chat. On the business tiers — ChatGPT Team and ChatGPT Enterprise — and on the API, OpenAI does not train on your inputs or outputs at all by default, as a contractual commitment. For a firm handling confidential deal data, the business tier is the version to use.
Will using ChatGPT leak my deal to a competitor?
No. Training a model on text does not store that text in a form another user can retrieve; no one can query ChatGPT to pull back a document you pasted. The genuine risk is narrower: on a consumer account with training left on, confidential contents have gone somewhere your firm did not authorize. A business-tier account, which does not train on your data, closes that gap. The “competitor reads my deal” scenario is not how these tools work.
Does using AI break my NDA?
Usually not, but check the specific agreement. Most NDAs restrict disclosure to third parties, and a no-training business tool holding your data under a data-processing agreement is, for routine work, comparable to Outlook or Dropbox — third-party services you already trust with confidential material. For anything genuinely sensitive, confirm the NDA does not bar third-party processing, and when in doubt, anonymize the input first.
Which ChatGPT plan should a small CRE firm buy?
For most 4-to-20-person firms, ChatGPT Team is the right starting point: it carries the no-training commitment at a modest per-seat cost without the enterprise controls a small firm does not yet need. Enterprise adds a formal data-processing agreement, audit logs, single sign-on, and custom retention — worth it once compliance or scale demands it. If your office already runs on Microsoft 365 or Google Workspace, look first at Microsoft Copilot or Gemini, since they may be bundled with what you pay for.
What should I never use ChatGPT for in CRE?
Two things. Never trust it for exact financial math — effective rent, CAM reconciliation, anything that determines an offer — because it produces confident, sometimes-wrong numbers; keep that arithmetic in a spreadsheet or purpose-built platform and verify by hand. And never treat it as a source of market facts, like comparable sales or cap rates, that you have not independently confirmed, because it will fabricate plausible-looking specifics when it does not know the answer.
How do I stop my brokers from pasting confidential data into personal accounts?
Give them one firm-controlled business account and make it the only one used for firm work, so the safe option is also the default. Pair it with a one-page rule that sorts documents into public, internal, and restricted. Brokers use personal accounts because no one gave them a sanctioned one; remove that reason and the shadow usage moves onto the account you control.
Do I need to hire technical staff to set this up?
No. Choosing a business-tier account, writing a one-page data rule, and running a short prompting session require no coding and no IT department — the whole setup fits in an afternoon. The only real investment is training existing staff to use the tool well, a skills step your own people can take directly rather than a hiring or engineering project.
Arthur Wandzel
