Short answer: It depends on the plan, and on your confidentiality agreement. A personal ChatGPT account is the wrong place for diligence files: on Free, Plus and Pro, OpenAI says "data sharing is enabled by default." ChatGPT Business and Enterprise do not train on your files by default, but they keep your chats until they are deleted, and some OpenAI staff can see stored chats for set reasons. Before using any plan for AI due diligence, check that your agreement lets you share Evaluation Material with an AI vendor, use the project code name instead of the target's name, and decide how you will delete what the tool keeps when the deal ends.
What each ChatGPT plan does with an uploaded document
These terms come from OpenAI's own pages, read on November 16, 2026: data controls, enterprise privacy, business data and pricing.
| Free, Plus, Pro (personal) | ChatGPT Business | ChatGPT Enterprise | |
|---|---|---|---|
| Trains on your content? | "data sharing is enabled by default" (you can turn it off) | "By default, OpenAI does not use content from ChatGPT Business, Enterprise, or Edu workspaces to train its models." | Same as Business |
| How long chats are kept | Not covered on the pages we read | "Your workspace admins can control how long your data is retained." Deleted or unsaved chats "are removed from our systems within 30 days", with exceptions | "Your workspace admins control how long your data is retained." Deleted chats removed "within 30 days, unless we are legally required to retain them" |
| Who at OpenAI can see stored chats | Not covered on the pages we read | "authorized employees that require access for engineering support, investigating potential platform abuse, and legal compliance" and "specialized third-party contractors" who review "for abuse and misuse" | Same as Business |
| Who in your firm can see them | Only the account holder | Set by your workspace | "Workspace admins ... can view, access, export, and delete end user conversations in the workspace." |
| Price | Personal plans, priced separately | Standard seat "$20/month if billed annually. $25/month if billed monthly." | "Please contact sales to learn more about our Enterprise plan." |
So the business plans answer the training question well. The deal question is the next one: a file you upload, and every answer about it, sits in OpenAI's systems for as long as your workspace keeps it. For ChatGPT privacy outside deal work, see Is ChatGPT private? on Zero Data Retention AI.
Why "no training" is not enough for a deal
A confidentiality agreement usually cares about three things, and training is only one way to break them:
- Who has the information. Agreements typically limit sharing to your "Representatives" who need it to evaluate the deal. A vendor holding your chats is someone else holding it.
- What it is used for. Evaluation Material may be used "solely for the purpose of evaluating the Transaction", in the words of one agreement filed with the SEC (TransDigm and Servotronics, 2025). No-training terms help here.
- What happens at the end. The same agreement says that on request "you will return or destroy at your cost all Evaluation Material", and "Any destruction of materials shall be confirmed by you in writing." Copies in a chat history count.
The CIM and NDA page goes through these clauses in more detail, including the backup carve-outs some agreements add.
A safer way to run AI due diligence
- One approved tool, never a personal account. Write down which tool your team may use for deal files.
- Read the agreement before the first upload. Look at the definition of Representatives, the use restriction and the return-or-destroy clause. If AI use is unclear, ask the counterparty or your counsel, and get the answer in writing.
- Use the code name. Agreements and teasers already use one ("Project Rise" in the TransDigm agreement). Keep it in your prompts too.
- Separate each deal. Use a separate project, workspace or account for each target, so a question about one target can never pull in another target's files.
- Check every number. FINRA's 2026 oversight report warns that generative AI can produce "information that is inaccurate or misleading, yet is presented as factual information" (FINRA). Tie every figure back to the source page before it reaches a model or an investment committee memo.
- Plan the end of the deal. Delete the chats and files, or reset the workspace, when the process ends or the seller asks for destruction. Keep a note of the date, because the agreement may ask you to confirm it in writing.
What other tools do differently
ChatGPT is one choice. Deal teams also use:
- AI inside the data room. Datasite, Ansarada, DealRoom and Intralinks now offer AI that works on documents inside the room, under the room's own permissions. The data room page compares them.
- Finance-specific platforms. Hebbia says "Your documents, prompts, and outputs are never used to train Hebbia's models or any model provider we work with" (Hebbia). Rogo says "We never use your private data to train or update our models" (Rogo). Both are sold through a sales process.
- A private AI workspace. Faradex gives each firm its own Workspace on dedicated infrastructure in the United States. Files and conversations are held in memory and are gone when the Workspace is reset, and only your team can see your files. It does not connect to data rooms, so files are uploaded by hand. It accepts PDFs, Word documents and spreadsheets; no page or file limits are published in the material we checked, so test your own documents first.
ChatGPT Enterprise and Claude Enterprise both say they do not train on your business data unless you opt in.