Blog
Your analysts are using personal ChatGPT on deal documents. What is the safest way to give them AI?
by Daniel Leong, GTMUpdated
Analysts using personal ChatGPT on deal documents? The safest way to give them AI: what Business plans change, what your NDA says, and a 30-day plan.

Short answer
Our analysts are using their own ChatGPT accounts on deal documents. What is the safest way to give them AI?
Replace personal accounts with a firm-managed AI workspace under single sign-on, pair it with a one-page acceptable use policy rather than an outright ban, and connect it to the folders where your deal documents already live, so the sanctioned tool is faster than pasting files into a personal browser tab. Check that the vendor never trains on your data, keeps each customer's data separate, holds independent security certifications and logs every action.
Is upgrading to ChatGPT Business or Enterprise enough, and can analysts keep personal accounts if they turn off model training?
Analysts should not use personal ChatGPT Free, Plus or Pro accounts on confidential deal documents. Even with "Improve the model for everyone" turned off, OpenAI says that giving feedback on a response (thumbs up or down) may allow that whole conversation to be used for training, and the uploaded files stay in a personal account outside your firm's retention, offboarding and audit controls. ChatGPT Business and Enterprise stop training by default and add admin controls, but a general-purpose chat seat is not built around your deal folders, your CRE systems or an investment committee review trail.
Why analysts use personal ChatGPT on deal documents (and why a blanket ban fails)
Across real estate private equity funds, REITs, debt funds and investment managers, analysts face the same daily bottleneck. One acquisition screen means reading a 90-page offering memorandum (OM), reconciling a rent roll against a trailing 12-month operating statement (T-12), checking recovery caps and co-tenancy clauses across leases and amendments, and drafting an investment committee (IC) memo under a bid deadline.
When a firm provides no approved AI tool, or one that needs clunky workarounds, analysts quietly open personal accounts to:
- Summarise executive summaries, submarket narratives and tenant profiles from NDA-bound offering memorandums.
- Reformat messy rent roll exports or seller T-12 line items into clean tables for Excel or ARGUS.
- Search long leases and amendment chains for renewal options, termination rights, rent steps and recovery caps.
- Draft first-pass sections of IC memos, lender updates and quarterly investor letters.
Ban it outright and usage does not stop; it moves to personal laptops and phones where IT and compliance cannot see it. The durable fix is a sanctioned environment that handles deal documents faster than a personal chat tab, with clear rules on what goes where.
Personal ChatGPT vs ChatGPT Business and Enterprise: what changes for deal documents
Most guides stop at "move your team to ChatGPT Business or Enterprise." Before deciding whether a business chat plan is enough for confidential deal flow, separate four questions: model training, data retention and ownership, access controls, and how the tool fits real estate work.
Personal accounts (Free, Plus and Pro)
- Training is on by default
- On individual plans (Free, Plus and Pro), OpenAI may use conversations and uploaded files to improve its models unless the user turns off "Improve the model for everyone" in Data controls.
- The feedback exception
- Even with that setting off, OpenAI says that if a user gives feedback on a response, the entire conversation tied to that feedback may be used for training.
- Training is not ownership
- Turning off training does not delete saved chats. Temporary Chats are not used for training and may be retained for up to 30 days. A personal account belongs to the employee, so the firm cannot enforce single sign-on, remove access when the analyst leaves, audit which documents were uploaded, or certify their destruction to a seller.
Sources: OpenAI, how your data is used to improve model performance (opens in a new tab) and data controls in ChatGPT (opens in a new tab).
ChatGPT Business and ChatGPT Enterprise
- No training by default
- OpenAI does not use inputs, outputs or files from ChatGPT Business or Enterprise to train its models by default, and feedback on a response does not change that.
- Encryption and certifications
- Data is encrypted at rest (AES-256) and in transit (TLS 1.2 or higher). OpenAI holds SOC 2 Type 2 for both plans; ChatGPT Enterprise is also certified to ISO/IEC 27001, 27017, 27018 and 27701. A data processing addendum is available.
- Firm-owned workspace
- The organisation owns the workspace, and both plans support SAML single sign-on. On Business, admins can view, export and delete members' conversations. OpenAI's own pages differ on Business retention: its enterprise privacy page says Business admins can control how long data is retained, its pricing page lists custom data retention policies only under Enterprise, and its user lifecycle documentation says Business workspaces keep chats, files and canvas documents indefinitely. Confirm the position with OpenAI in writing before relying on Business to meet an NDA destruction request. ChatGPT Enterprise adds custom data retention policies, SCIM provisioning, role-based access, IP allow-listing, a compliance API, enterprise key management and data residency for eligible customers in supported regions.
Sources: OpenAI, enterprise privacy (opens in a new tab), business plans and pricing (opens in a new tab) and user lifecycle (opens in a new tab), checked 20 November 2026.
What a general-purpose chat seat still does not solve
- Apps, not deal-room governance
- ChatGPT offers apps (formerly connectors) for file stores such as SharePoint, Box and Google Drive. For SharePoint, workspace admins can allow or block whole site collections and exclude files with chosen sensitivity labels, once OpenAI enables those controls for the workspace; the rules cannot target a single folder or file. None of this is set up around your deal folders, broker data rooms or deal walls, and files added to a shared project can blur deal walls unless someone polices them.
- No CRE layer
- A chat seat is not built around your deal pipeline or the underwriting models your team keeps in Excel and ARGUS.
- No IC review trail
- When an investment committee member asks where a figure came from, a chat thread gives prose, not a reviewable trail of sources, assumptions and sign-off.
Comparison: personal ChatGPT vs ChatGPT Business or Enterprise vs Fifth Dimension
Model training on deal documents
- Personal ChatGPT (Free, Plus, Pro)
- On by default; even if opted out, feedback on a response may allow training on that conversation
- ChatGPT Business or Enterprise
- Off by default, including on feedback
- Fifth Dimension
- Customer data remains private and is never used to train the models that power Fifth Dimension
Environment and isolation
- Personal ChatGPT (Free, Plus, Pro)
- Personal consumer account owned by the analyst
- ChatGPT Business or Enterprise
- Organisation workspace; OpenAI's pages differ on Business retention; Enterprise adds custom data retention policies, data residency and enterprise key management
- Fifth Dimension
- Each customer operates in a separate product environment, with no data cross-pollination; on-premise or air-gapped deployment available
Certifications and residency
- Personal ChatGPT (Free, Plus, Pro)
- Consumer terms of use
- ChatGPT Business or Enterprise
- SOC 2 Type 2 on both; ISO 27001, 27017, 27018, 27701 on Enterprise only; data processing addendum
- Fifth Dimension
- SOC 2 Type II compliant, ISO 27001 certified and GDPR compliant; EU data stays in the EU and US data stays in the US
Access and single sign-on
- Personal ChatGPT (Free, Plus, Pro)
- Personal email and password; the firm cannot revoke or audit
- ChatGPT Business or Enterprise
- SAML SSO on both; SCIM, role-based access and IP allow-listing on Enterprise
- Fifth Dimension
- SAML SSO and role-based access, IP allow-listing, and data lifecycle management
How deal documents get in
- Personal ChatGPT (Free, Plus, Pro)
- Manual upload, or a personal connection to Box, SharePoint or Google Drive that the firm does not manage
- ChatGPT Business or Enterprise
- Upload into chats or general-purpose apps
- Fifth Dimension
- Workspaces synced with your underlying data rooms; connects Excel, SharePoint, Argus, Dealpath and more
Review for IC and investors
- Personal ChatGPT (Free, Plus, Pro)
- Chat answers with no firm review trail
- ChatGPT Business or Enterprise
- Can cite sources; no CRE review and sign-off trail
- Fifth Dimension
- Evidence & Controls keeps sources, assumptions, review steps and changes visible
Audit trail
- Personal ChatGPT (Free, Plus, Pro)
- None available to the employer
- ChatGPT Business or Enterprise
- Admin controls; compliance API on Enterprise
- Fifth Dimension
- Every operation is logged and traceable
What your NDA means for AI on deal documents
What follows is general information, not legal advice. Your NDA and your counsel govern what you may do with deal materials.
Before receiving a confidential offering memorandum or a data room login, a buyer signs a confidentiality agreement with the seller or broker. Common provisions that a personal AI account can conflict with:
- Representatives
- NDAs typically let the buyer share evaluation material only with directors, officers, employees, advisers and financing sources (often only with the broker's consent) who need to know it and are bound by confidentiality, and make the buyer liable for its representatives' breaches. A personal consumer AI account is unlikely to qualify as one of the firm's representatives.
- Permitted purpose
- Evaluation material may be used only to evaluate the potential acquisition. A vendor using uploaded documents to improve its models is unlikely to fit within that purpose.
- No copies without consent
- Many agreements say materials may not be copied without the owner's and broker's written consent. Uploading a file to a personal AI account creates a copy outside the firm.
- Return or destroy
- When a firm passes, the NDA typically requires returning or destroying all copies and analyses on request, often with written certification. Many agreements carve out archived backups and copies kept for legal or compliance reasons; a personal AI account is neither. If an analyst pasted the rent roll into a personal account, the firm cannot verify or certify that destruction.
These clauses apply to any AI vendor, including an approved one, and including us. Ask counsel whether each agreement treats a firm-contracted provider, bound by confidentiality and under the firm's control, like your other IT systems, and confirm the provider can remove a deal's data on request so you can certify destruction.
Deal teams also handle tenant financial statements, investor information and other non-public material beyond the broker package. If your firm has a broker-dealer affiliate or is, or advises, an SEC-registered adviser, ask your chief compliance officer and counsel how supervision, recordkeeping and privacy rules apply to AI use.
Enterprise AI security for deal documents: what to require
Before approving any AI platform for acquisition packages, lease files or portfolio reporting, require these controls. For the full diligence framework, see the AI vendor security checklist for real estate CFOs.
- 1. No training on your data
- A plain-language commitment that your inputs, documents and outputs are never used to train or improve models.
- 2. Isolated customer environment
- Logical or physical separation, so one client's files never mix with another's, and on-premise options if you need them.
- 3. Independent certifications and residency
- Current SOC 2 Type II and ISO 27001, GDPR compliance, and data kept in the region you agree.
- 4. Central access control
- SAML single sign-on, role-based access and IP allow-listing, so IT governs access in one place.
- 5. Connections instead of downloads
- A direct connection to where deal documents already live, so analysts stop downloading confidential files to local folders to feed a chatbot.
- 6. Source citations
- Every extracted figure and term should link back to its source document, so a reviewer can check it before it reaches IC or investors.
- 7. Audit logs and lifecycle controls
- Every action logged, and agreed controls on how data is retained and removed.
- 8. Published policies and subprocessors
- A trust centre with the vendor's security policy, acceptable use policy and subprocessor list, so you can see who else touches your data.
A 30-day plan to move analysts off personal ChatGPT
A compliance memo alone rarely ends unapproved AI use. Use four weeks to move acquisitions, asset management and finance teams onto a governed setup. This plan starts once security review and the contract for the sanctioned tool are complete.
Week 1: map what analysts actually do, without blame
- Run a short anonymous survey: which three weekly tasks take the most document or spreadsheet work, and where have you tried AI?
- Identify the highest-risk documents leaving the data room: usually NDA-bound OMs, rent rolls, T-12s, lease packs and debt term sheets.
- Align IT, legal and compliance, and deal leadership on one message: the firm wants AI on every deal, inside its governed workspace.
- Ask analysts to disclose which deal files and chats sit in personal AI accounts, and tell them in advance how the firm will treat good-faith disclosure. Ask them not to delete anything yet.
- Have counsel and compliance decide, before anything is deleted, whether each item is preserved, exported or deleted, and whether any NDA notice is owed. Record what was disclosed and what was done under your records retention policy.
Week 2: publish a one-page policy and stand up the workspace
- Publish the acceptable use policy below, naming the approved tools, the never-paste list and the verification rule.
- Turn on single sign-on, role-based access, IP allow-listing (with a path for remote and mobile access) and your agreed data residency on the sanctioned platform.
- Connect your deal libraries so analysts do not upload files by hand.
Week 3: move the three highest-friction tasks first
- First-pass screening of an inbound OM, rent roll and T-12.
- Lease and amendment review.
- First-draft IC screening memos, with sources attached for the reviewer.
Week 4: enforce and review
- Block consumer AI file uploads on corporate devices once the sanctioned workspace is live and faster than the old habit, paired with data loss prevention on paste and a personal-device rule in the policy.
- Restrict user consent to third-party apps in your identity and workspace settings (for example Microsoft Entra ID and Google Workspace), so personal AI accounts cannot connect to firm file stores.
- Review aggregate adoption with deal team leads; compliance or IT reviews audit logs under the policy, with any monitoring notice your jurisdiction requires.
- Update your investor due diligence questionnaire answers on AI use, reviewed by compliance, citing the vendor's current certifications and trust centre documents.
Risky habit vs safe alternative: a table for deal teams
Share this with your acquisitions, asset management and portfolio finance teams:
Offering memorandum under NDA
- Risky habit (personal ChatGPT)
- Uploading the OM to a personal account to draft a summary
- Why it creates risk
- Risks breaching the NDA's representative, no-copy and permitted purpose terms; the firm cannot certify destruction; feedback may allow training
- Safe alternative
- Work from the deal folder inside the firm's governed workspace
Rent roll
- Risky habit (personal ChatGPT)
- Pasting tenant names, suites, dates and rents into a chat to calculate rollover or WALE
- Why it creates risk
- Exposes confidential tenant data; long tables can lose rows with no trail back to the source
- Safe alternative
- Use the governed workspace, with figures linked to their source
T-12 operating statement
- Risky habit (personal ChatGPT)
- Uploading the seller's T-12 to categorise expenses
- Why it creates risk
- Moves property-level financials outside the firm's controls; no benchmark against your own operating history
- Safe alternative
- Work from the seller's T-12 inside the governed workspace
Leases, amendments and side letters
- Risky habit (personal ChatGPT)
- Uploading one lease to ask about recovery caps or termination options
- Why it creates risk
- A single-file prompt can miss how a later amendment changed the clause
- Safe alternative
- Review the lease together with its amendments inside the governed workspace
Debt term sheets and covenants
- Risky habit (personal ChatGPT)
- Pasting lender quotes and covenants into a personal chat
- Why it creates risk
- Exposes lender pricing and deal structure outside the firm's controls
- Safe alternative
- Compare term sheets inside a permissioned deal workspace
IC memos and investor materials
- Risky habit (personal ChatGPT)
- Drafting in a personal chat with no source links
- Why it creates risk
- Falls outside the firm's controls and records; IC and investors cannot see where numbers came from
- Safe alternative
- Draft inside the governed workspace, with sources linked and actions logged
Template: one-page AI acceptable use policy for deal teams
Copy and adapt this for your handbook, compliance manual and deal-team onboarding. Have your counsel and chief compliance officer review it before you adopt it.
- 1. Purpose
- [Firm name] encourages AI to speed up deal screening, lease analysis, underwriting and portfolio reporting, provided all confidential deal, property, tenant and investor data stays inside our governed environment.
- 2. Approved tools
- For deal documents, underwriting, lease review and IC or investor materials: [approved platform], through company single sign-on. For general non-deal work: [approved general tool, if any]. Personal or consumer AI accounts, unvetted browser extensions and unapproved meeting note-takers may never be used on firm business; approved note-takers need the consent of everyone recorded. AI features inside software we already use (for example Copilot, Acrobat or Zoom) count as AI tools and need the same approval.
- 3. The never-paste rule
- Never upload, paste or type into a personal or unapproved AI tool: offering memorandums and data room files under NDA; rent rolls, T-12s, budgets, appraisals or underwriting models; leases, amendments, side letters, estoppels or tenant financials; debt term sheets, loan agreements or joint venture agreements; IC memos, investor letters, questionnaire answers or any non-public information.
- 4. Verify before you use
- AI speeds up work but does not replace judgement. Before any AI-assisted figure, term or passage goes into a model, IC memo, lender package or investor communication, check it against its source document. You remain accountable for every number you submit. Label AI-drafted content internally, and disclose it where an agreement or investor requires.
- 5. Deal walls and non-public information
- Respect deal walls: never combine information across walled deals, and never use AI on material non-public information outside approved tools.
- 6. Personal devices
- Do not access deal documents or approved AI tools on personal devices except through [approved access method].
- 7. Records and NDA destruction
- AI-assisted outputs that become firm records are kept under [records retention policy]. Work from connected deal libraries, not local downloads. When a deal passes and a destruction notice applies, tell compliance or IT so the deal's data is handled under our retention policy.
- 8. Report mistakes quickly
- If you pasted deal or investor information into an unapproved tool, report it to compliance within 24 hours, and do not delete it until compliance tells you how it will be handled. Prompt, good-faith reporting will be treated as a mitigating factor under [disciplinary policy].
- 9. Exceptions, questions and new tools
- To request an exception, test a new AI use or connect a new data source, contact [compliance officer or CTO] at [email].
- 10. Monitoring and enforcement
- Use of approved tools is logged and may be reviewed by compliance or IT. Breaches are handled under [disciplinary policy]. Staff confirm they have read this policy at onboarding and each year. Owner: [name]; next review: [date].
Where Fifth Dimension fits
Fifth Dimension is AI for commercial real estate, from underwriting to lease abstraction. On security, from Fifth Dimension security:
- SOC 2 Type II compliant, ISO 27001 certified and GDPR compliant; EU data stays in the EU and US data stays in the US.
- Customer data remains private and is never used to train the models that power Fifth Dimension.
- Each customer operates in a separate product environment, with no data cross-pollination between clients; on-premise or air-gapped deployment is available.
- SAML SSO and role-based access, IP allow-listing, and agreed controls on how customer data is retained and removed.
- Every operation is logged and traceable, giving compliance teams visibility into how work was completed.
For deal teams, Workspaces give a shared place for the context, activity, review and outputs behind an investment decision, synced with your underlying data rooms. Evidence & Controls keeps sources, assumptions, review steps and changes visible, and Fifth Dimension's AI agent applies your firm's terminology, processes and priorities consistently across each task. Review the policies in the Fifth Dimension Trust Center.
Together, we identify the manual processes that are slowing you down across the full lifecycle of how you invest in and manage your assets. Start with the people and work that matter most, then expand as Fifth Dimension connects more of your organisation. Fifth Dimension includes high-quality service to help your team identify valuable workflows, configure the platform around how your firm works and keep improving what you have built. Request a demo, and bring the workflow, not the files, until IT and risk have agreed what can be shared.




