Where client data goes: personal AI, business plans, APIs and a local model
Same model, very different terms. What happens to a client file on a personal plan, a business plan, the API and a model on your own hardware, checked on the vendors’ pages.
In short
Personal ChatGPT and Claude plans can train on chats unless the user opts out, and the personal Gemini app uses chats to improve Google’s services. Business plans and paid APIs do not train by default and come with a data processing agreement; business plans add SSO and admin controls. Only a local model keeps the prompt on your hardware.
The same model can sit behind very different terms. A client file pasted into a personal account, a business workspace, the API or a model on your own server ends up in different places, for different lengths of time, under different contracts.
Every cell below comes from the vendor’s own page, read on 5 November 2026. Where a page says nothing, the cell says so.
The short answer
| Setup | Trains on your chats? | Contract for client data |
|---|---|---|
| Personal plan (ChatGPT, Claude, Gemini app) | Yes by default, or until the user opts out | None listed |
| Business plan (ChatGPT Business/Enterprise, Claude Team/Enterprise, Gemini in Workspace) | No, by default | Data processing agreement |
| Paid API (OpenAI, Anthropic, Gemini paid tier) | No, by default | Data processing agreement |
| Local model on your hardware | Only if you set it up to | Not needed for the model; needed with your hosting provider |
OpenAI: ChatGPT and the API
| Personal (Free, Plus, Pro) | Business, Enterprise | API | |
|---|---|---|---|
| Training default | On; “Improve the model for everyone” can be turned off | Not by default | Not by default |
| Retention | Saved chats stay in history; temporary chats up to 30 days | Admins set it; deleted chats removed within 30 days | Up to 30 days for abuse monitoring; zero retention for eligible use cases |
| Admin controls | None (personal account) | Admins can view, export and delete workspace chats | Not stated on the pages checked |
| DPA | Not listed | Yes | Yes |
| SSO | Not listed | SAML SSO | Not stated on the pages checked |
| Data location | Not stated on the pages checked | At-rest residency for eligible new Enterprise and Edu workspaces | EU processing for eligible endpoints |
Anthropic: Claude and the API
| Personal (Free, Pro, Max) | Team, Enterprise | API | |
|---|---|---|---|
| Training default | Opt-out: used unless the user turns it off | None by default | Not by default |
| Retention | Up to 5 years if training is allowed, otherwise 30 days | Kept in the product until deleted, then removed within 30 days; custom retention on Enterprise | Deleted within 30 days by default; zero retention by agreement |
| Admin controls | None (personal account) | Central admin; Enterprise adds SCIM, audit logs and role-based access | Not stated on the pages checked |
| DPA | Not listed | Yes (commercial products) | Yes (commercial products) |
| SSO | No | Yes | Not stated on the pages checked |
| Data location | Not stated on the pages checked | Not stated on the pages checked | Not stated on the pages checked |
Google: Gemini
| Gemini app (personal) | Gemini in Google Workspace | Gemini API | |
|---|---|---|---|
| Training default | Used to improve Google services; human reviewers read some chats | Not used for model training outside your domain without permission; no human review | Free tier: used to improve products, may be human-reviewed. Paid: not used |
| Retention | Auto-delete after 18 months by default (3, 36 or never); reviewed chats up to 3 years | Gemini app chats up to 36 months, as admins set | Paid: logged for a limited period to detect abuse |
| Admin controls | None (personal account) | Admins decide whether chats are saved and for how long | Not stated on the pages checked |
| DPA | Not listed | Cloud Data Processing Addendum | Paid: Data Processing Addendum |
| SSO | Not listed | Not stated on the AI privacy page | Not stated on the pages checked |
| Data location | Not stated on the pages checked | Not stated on the pages checked | Not stated on the pages checked |
A local model
A model running on your hardware, or on a server you rent and control, has no vendor terms for the prompt. It trains on nothing unless you set it up to, keeps what your configuration keeps and sits where you put it. You still need a processing agreement with whoever hosts the server.
What this means for a professional office
- Personal plans: never for client data. The terms are written for individuals, the defaults favour training, and there is no processing agreement.
- Business plans and paid APIs: acceptable for work that is not confidential, under the DPA. The data still leaves the firm, and retention runs on the vendor’s clock.
- Confidential matters: a local or private model behind a gateway that redacts, routes and logs.
The table assumes staff use the setup you chose. The most common leak is an employee on a personal plan anyway. Our guide to the API-wrapper myth explains why, and the tutorial builds the gateway layers in code.
How Paxa Media sets it up
We map which data goes where, then split the work: confidential matters on a local model, low-risk drafting on a business plan with a DPA, both behind one gateway with an audit log. It starts with a demo on synthetic data. What a leak costs in law is on our AI data-leakage page.
Key takeaways
- Personal plans of ChatGPT and Claude train unless the user opts out; the personal Gemini app uses chats to improve Google’s services.
- Business plans and paid APIs do not train by default and offer a data processing agreement.
- The free Gemini API tier is used to improve Google’s products; the paid tier is not.
- Only a local model keeps the prompt on your own hardware, and only if egress is checked.
Frequently asked questions
Is ChatGPT Plus safe for client data if I turn off training?
Turning off “Improve the model for everyone” stops training on new chats. It does not give you a processing agreement, admin controls or SSO, and chats still sit in a personal account the firm cannot see. Use a business plan or a local model.
Does a business plan mean the vendor never sees our data?
No. The vendor processes every prompt to answer it and keeps it for a period: admins set retention on ChatGPT Business and Enterprise, and deleted chats are removed within 30 days. The difference is the contract and the training default.
Is the Gemini API free tier fine for testing with real files?
No. Google’s terms say unpaid use improves its products and human reviewers may read inputs and outputs. Test with invented data, or on the paid tier under the data processing addendum.
Where is our data stored on these plans?
OpenAI offers at-rest residency for eligible new Enterprise and Edu workspaces and EU processing for eligible API endpoints. The other pages we checked do not state a location. Ask the vendor in writing before relying on it.
Sources
Pages we checked for the facts above, with the date we last read them.
- Data controls in ChatGPT — OpenAI Help Center, checked 2026-11-05
Personal ChatGPT: training on by default, can be turned off; temporary chats kept up to 30 days. - Enterprise privacy at OpenAI — OpenAI, checked 2026-11-05
Business, Enterprise and API: no training by default, admin retention, 30-day deletion, API 30-day retention, DPA, SAML SSO, admin access. - Introducing data residency in Europe — OpenAI, checked 2026-11-05
Data residency for eligible Enterprise/Edu workspaces and API endpoints. - Updates to Consumer Terms and Privacy Policy — Anthropic, checked 2026-11-05
Free, Pro and Max: user choice on training; 5-year retention if allowed, 30 days otherwise. - Claude plans and pricing — Anthropic, checked 2026-11-05
Training opt-out on personal plans, none by default on Team/Enterprise; SSO on Team/Enterprise; SCIM, audit logs, custom retention on Enterprise. - Is my data used for model training? (commercial) — Anthropic Privacy Center, checked 2026-11-05
Commercial products not used for training by default. - How long do you store my organization’s data? — Anthropic Privacy Center, checked 2026-11-05
API deletion within 30 days; commercial chats deleted within 30 days of deletion; DPA for commercial products. - Gemini Apps Privacy Hub — Google, checked 2026-11-05
Gemini app data used to improve services; human review; 18-month default auto-delete; reviewed chats up to 3 years. - Generative AI in Google Workspace Privacy Hub — Google Workspace, checked 2026-11-05
Workspace: no training outside the domain without permission, no human review, admin retention up to 36 months, Cloud DPA. - Gemini API Additional Terms of Service — Google, checked 2026-11-05
Free tier used to improve products with possible human review; paid tier not used, DPA applies, limited logging.
See it on one of your own workflows
We start with a demo on synthetic data: one office workflow, a local model and the gateway rules from these articles. Real client data comes in only after a written scope.
Book a demo on synthetic data
Tell us which workflow you want to speed up. The demo needs none of your real data.
- Fixed-scope quote — no obligation
- Reply within 1 business day
About the author
Paxa Media
Written by Paxa Media, a technology company based in Rijeka, Croatia. Price ranges in our guides are indicative; your quote depends on scope.
Why you can trust this
- Builds the kinds of systems described here — websites, web and mobile apps, automation and integrations
- Technology company based in Rijeka, Croatia
- Price ranges in these guides are indicative, not quotes; scope sets the final price
Keep reading
All comparisons →Free ChatGPT at Work: What It Costs You in Data, and When to Pay
The price is €0. The data terms are the same ones you get on Go, Plus and Pro. What changes when you pay, and the point where a company needs a different plan.
AI & automationChatGPT behind your own API does not keep client data private
An internal wrapper changes the interface, not where the data goes. Eight layers of defence, the leak each one misses, and why the last gap is a personal chatbot.
AI & automationBuild a leak-resistant AI gateway for a professional office
Six small TypeScript modules that sit between your staff and any model, each covered by a test. Plus the one gap that code cannot close.