ChatGPT and Claude compared
For individual plans, from OpenAI's and Anthropic's help pages, read November 6, 2026:
| ChatGPT | Claude | |
|---|---|---|
| Used for training? | May be, unless you opt out | Only if you allow it, or the chat is flagged for safety review |
| Where to change it | Settings, Data controls, Improve the model for everyone | Settings, Privacy, Help Improve our AI models |
| Private chat | Temporary chat: no training, out of history, a copy kept up to 30 days for safety | Incognito chat: not used to improve Claude |
| When you delete a chat | Permanent deletion scheduled within 30 days | Removed from back-end storage within 30 days |
| What outlasts deletion | Copies saved separately in Library or Space; security or legal holds | De-identified training data up to 5 years if training is on; flagged chats up to 2 years; legal holds |
On both, the thumbs up and down buttons change things. OpenAI says that feedback lets it use the entire conversation for training even after you opt out, and Anthropic stores the related conversation for up to 5 years. Skip those buttons in any chat that holds a statement.
Before you upload
- Change the training setting. It applies to new chats, so do it first.
- Use a temporary or incognito chat. Don't save a temporary ChatGPT chat afterward: saving turns it into a regular chat under your normal settings.
- Upload a transactions-only CSV. It usually leaves out your name, address and full account number. ChatGPT and Claude both read CSV files, and they tend to read them more reliably than PDFs.
- Send the shortest period that answers the question. One month is less to expose than a year.
- Protect the AI account. Turn on two-factor authentication, and stay out of accounts you share with other people.
After you're done
- Delete the chat. Content from a file can stay in the conversation until you do.
- In ChatGPT, also delete any saved copy in Library or Space. OpenAI says deleting a chat doesn't remove those.
- Don't leave statements in a project or custom GPT, where files stay until you remove them or delete the project or GPT.
- Don't create a share link for that chat.
Uploading vs a read-only connector
A connector such as BankBridge shares your data differently. Neither option keeps your data away from the AI company, since what the assistant reads still enters the conversation. The difference is how much it reads.
| Statement upload | BankBridge | |
|---|---|---|
| Identity details | Name, address, full account number | Account name and last 4 digits |
| Amount of data | The whole file | What each question requests |
| Companies involved | The AI company | The AI company, BankBridge and its bank-connection provider |
| To stop | Delete each chat and file | Remove the connector, and delete past chats you don't want kept |
Anthropic's training page adds a point for Claude: training data doesn't include raw content from connectors, including remote MCP servers, unless it's copied directly into your conversation. A file you upload has no such exception.
BankBridge is read-only, doesn't keep a copy of your transactions, and never sees your bank password. It costs $5 a month per connected bank, so uploads remain the better fit for a one-time question or if you'd rather not connect a bank at all. If you ask every month, the connector safety checklist covers what to look for, and you can connect your bank in about five minutes.
Statements that belong to a business or a client
OpenAI says it doesn't use ChatGPT Business, Enterprise or Edu content to improve models by default, and Anthropic says the same of its commercial products, apart from feedback you choose to send. Use one of those workspaces for company statements. If you hold a client's statements, check your engagement terms and any rules that apply to your profession before uploading them to any AI tool.