The short answer: it depends on which ChatGPT you are using, and for a lawyer, that answer is not good enough on its own. On consumer plans, your conversations may be used to train OpenAI's models unless you opt out. On business plans and the API, they are not used for training by default, but they still leave your control, transit US infrastructure, and are retained under policies you do not set. Confidential in the everyday sense is not the same thing as compatible with professional secrecy.
This article walks through what actually happens to the text you type into ChatGPT, why "is it confidential" is the wrong question for legal work, and how to draw a defensible line for your firm.
What ChatGPT actually does with your data
OpenAI's data handling differs sharply by product tier, and most of the confusion in legal circles comes from talking about "ChatGPT" as if it were one thing. As of this writing, the broad picture is:
- Free and Plus (consumer). Conversations may be used to improve OpenAI's models unless you disable the setting or use a temporary chat. Chats are stored on your account, and flagged content can be reviewed by humans under OpenAI's policies.
- Team, Enterprise, and Edu. Business tiers are not used for training by default, add admin controls and encryption commitments, and offer a data processing agreement. This is a materially better posture, and also a contract your firm has to actually sign and configure, not a default.
- The API. Data sent through the API is not used for training. It may still be retained for a limited period for abuse monitoring unless stricter retention terms apply to your account.
Every one of those statements is a snapshot of a vendor's current policy. Policies change, and none of them were written with professional secrecy statutes in mind. If your confidentiality analysis rests on a settings toggle, it rests on something the vendor can move.
Deleted does not always mean deleted
Two facts belong in any lawyer's mental model of consumer AI tools.
First, deletion is a policy, not a physical guarantee. Providers typically describe deletion as completing within a window, commonly around thirty days, and carve out backups and legal obligations.
Second, those carve-outs are real. In 2025, a United States court ordered OpenAI to preserve ChatGPT conversation logs in ongoing copyright litigation, including chats users believed they had deleted. Whatever your view of that dispute, the lesson for lawyers is uncomfortable: a provider's retention promises operate inside the US legal system, not above it. Data held by a US company is reachable by US courts and US disclosure regimes, wherever the user sits.
For a profession whose duty of confidentiality is enforced by criminal law in some jurisdictions, including Article 321 of the Swiss Criminal Code, that jurisdictional point is not a technicality.
Why "confidential" is the wrong question for legal work
For most users, confidentiality means "will strangers see my chats". For a lawyer, the bar is higher and the questions are different:
- Professional secrecy. Client identity, facts of the matter, and legal strategy are protected by bar rules and, in Switzerland, by criminal law. Sending them to a third party without a proper legal basis and safeguards is a problem even if that third party never leaks them.
- Privilege. Commentators and bar opinions have flagged the risk that routing privileged material through a consumer service, under terms that permit the provider to use it, sits badly with the confidentiality expectations privilege depends on.
- Data protection. Under the GDPR and the Swiss nFADP, client data in a prompt is personal data being disclosed to a processor. That requires a data processing agreement, a lawful basis, and a transfer analysis when the processing happens in the United States. Consumer ChatGPT offers none of that; business tiers offer some of it.
- Jurisdiction. Even with training disabled, the data is processed on US infrastructure by a US company, within reach of the US CLOUD Act. For Swiss and EU firms advising on sensitive matters, where the data physically lives and which courts can compel its production is part of the analysis, not an afterthought.
What bar associations say
Guidance has converged on a consistent theme: lawyers may use generative AI, and remain fully responsible for what goes into it and what comes out of it.
The American Bar Association's Formal Opinion 512 on generative AI puts the duty of confidentiality front and centre: lawyers must evaluate a tool's terms and data practices before inputting client information, and in some circumstances need informed client consent. European and Swiss professional bodies have issued guidance in the same direction, with particular attention to secrecy obligations and data protection law.
None of this guidance bans the tools. All of it puts the burden of the confidentiality analysis on the lawyer, which is exactly where clients and regulators will put it too.
A practical policy your firm can defend
The workable line is not "never use AI". It is a written policy that separates what may be pasted into a general-purpose chatbot from what may not:
- Never paste client names or identifying facts, privileged communications, draft filings in live matters, opposing-party documents, or anything a client handed you in confidence.
- Safe uses include anonymised drafting, generic legal writing, plain-language explanations, brainstorming structures and counter-arguments, and working with material that is already public.
- If your firm uses ChatGPT deliberately, use a business tier, sign the DPA, disable training, restrict it to defined use cases, and record the analysis. An AI governance policy turns individual judgement calls into a firm-level standard.
- Verify everything. Confidentiality aside, general-purpose chatbots fabricate citations. Every authority must be independently checked.
The purpose-built alternative
The reason this analysis is hard with general-purpose chatbots is that they were not built for it. Legal AI platforms invert the defaults: instead of asking whether the vendor's terms can be bent into compliance, the infrastructure is designed around the obligations.
That is the approach we take at Whisperit. Client data is stored on Swiss infrastructure, AI providers are bound by data processing agreements that prohibit training on client data, and firms choose a data residency tier, including a Swiss-only tier where AI processing never leaves Switzerland and never falls back to another jurisdiction. The full picture, provider by provider, is on our security page, and security teams can request the technical data flow documentation that we hand to IT and compliance reviewers.
For a side-by-side look at how the general-purpose assistants compare with a purpose-built platform, see our Whisperit vs Claude vs ChatGPT comparison.
The bottom line
Is ChatGPT confidential? On consumer plans, not in any sense a lawyer can rely on. On business tiers, it can be made defensible for a defined set of non-privileged uses, if your firm does the contractual and configuration work and writes down the boundary. For client-identifying and privileged material, the honest answer is that a general-purpose chatbot processed in the United States is the wrong tool, whatever the settings say.
The firms getting this right are not the ones avoiding AI. They are the ones who can show a client, in writing, exactly where every piece of the file went.