Pasted Personal Data Into ChatGPT? What to Do Next
Pasted personal data into ChatGPT, Gemini, or Claude? See what deletion and training opt-outs actually do, plus a five-point check before your next prompt.
For: People who use ChatGPT, Gemini, or Claude at work and worry about pasting personal or confidential data
Key points
- Deleting a chat removes it from history, but it does not guarantee removal from training data, safety logs, or backups.
- ChatGPT, Gemini, and Claude each have a training opt-out, but feedback, safety review, and legal retention sit outside it.
- Before your next prompt, check five things: personal data, sensitive data, third-party data, contract rules, and whether the service trains on your input.
Editor's note
Verified on November 13, 2026 against OpenAI, Google, and Anthropic help pages and privacy policies, plus the Japanese Personal Information Protection Commission's June 2023 guidance. Settings and retention periods change, and this is not legal advice. Check the current UI and your contract before relying on a setting.
Editorial policy and how we use AI →Contents
- What should you do first?
- What does deletion not remove?
- Where are the "don't train" settings?
- ChatGPT
- Gemini
- Claude
- Five checks before your next prompt
- Personal data versus sensitive personal data
- Third-party data is not yours to decide
- How do you mask names and details?
- If you use AI for work, what should you choose?
- What is the legal position?
- FAQ
On Friday evening, you paste a client's contract draft into ChatGPT and ask it to flag risky clauses. On Monday morning, you realize the client's name, address, and phone number were still in the document. You delete the chat, but you are not sure what that actually did.
Deleting a chat and removing data from training, safety logs, and backups are different operations. The practical order is: delete the conversation, check the training opt-out, then read the provider's retention policy and exceptions. No consumer AI service can promise that a pasted detail is gone from every system.
This guide covers what to do after a paste, where the training settings live, the five checks to run before your next prompt, and how to mask names without losing the point of the document.
This is general information, not legal advice. Whether a specific input is lawful depends on the data, purpose, contract, and provider settings. Ask your legal or privacy team before making a decision.
What should you do first?#
Do three things in order: delete the conversation, confirm the training setting, and read the retention exceptions. Do not assume deletion is enough.
- Delete the conversation. In ChatGPT, use the chat menu. In Gemini, delete the activity in Gemini Apps Activity. In Claude, delete the conversation. It disappears from your history immediately.
- Check the training opt-out. On consumer plans, training may be on by default. Turn it off in the data controls or the provider's privacy portal so new conversations are not used.
- Read the retention exceptions. Feedback, safety review, backups, and legal retention can outlive the deleted chat. Check the provider's help page for the exact scope.
| Action | What it fixes | What it does not fix |
|---|---|---|
| Delete the chat | Your visible history | Training data, safety logs, backups |
| Turn off training | New conversations | Past conversations, feedback, safety review |
| Shorten retention | Future storage | Data already stored, legal retention |
If the paste involved a client's confidential data, a lawsuit, or an audit, talk to your company's security or legal team before deleting anything. A deletion can interfere with evidence preservation.
What does deletion not remove?#
A delete button usually removes the conversation from your history and the provider's live system. It does not automatically remove data used for training, feedback, safety review, backups, or legal compliance.
| Service | What deletion does | What sits outside it |
|---|---|---|
| ChatGPT (consumer) | The chat disappears from history | If you send thumbs up or down, the whole conversation may be used for training even after you opt out. Business and API data are not used for training by default |
| Gemini Apps | You can delete activity; auto-delete defaults to 18 months and can be set to 3, 36, or never | Human-reviewed chats are kept for up to 3 years even if you delete activity. With activity off, Google still uses chats to provide responses, for safety, and for human review |
| Claude | A deleted conversation leaves history immediately and is removed from the backend within 30 days | Even after opting out, flagged content and explicitly reported content can be used for safety and model improvement |
These are the providers' published descriptions as of November 13, 2026. The scope varies by plan and region. Check the current help page and the actual UI.
Sources: OpenAI data controls, Gemini Apps privacy hub, Anthropic Privacy Policy.
Where are the "don't train" settings?#
Consumer plans let you change the training setting, but the opt-out does not cover feedback or safety review. The exact labels move, so verify them in the current UI.
ChatGPT#
On Free, Plus, and Pro, open Settings > Data controls and turn off Improve the model for everyone. Alternatively, use the OpenAI Privacy Portal and choose Do not train on my content. Either is enough for ChatGPT conversations.
After you opt out, new conversations are not used to improve models. If you send thumbs up or down, however, the entire conversation associated with that feedback may be used. Temporary chats are not used while they remain temporary, but saving one turns it into a regular chat that follows your account settings.
ChatGPT Business, Enterprise, Edu, and the API do not use inputs or outputs for training by default. API organization owners can enable data sharing, and members can then choose to share feedback and the current Playground chat.
Gemini#
Gemini Apps uses Gemini Apps Activity in your Google Account to manage storage and deletion. Auto-delete defaults to 18 months and can be set to 3 months, 36 months, or never.
Turning activity off does not stop all use. Google says it still uses chats to provide responses, protect users, and conduct human review. Human-reviewed chats are kept for up to 3 years even if you delete activity. Google's help page tells users not to enter confidential information they do not want reviewers to see.
Work and school Google accounts may have separate data processing terms. Do not assume they match your personal account.
Claude#
Claude lets you opt out of training on your inputs and outputs in account settings. A deleted conversation leaves your history immediately and is removed from the backend within 30 days.
Even after you opt out, Anthropic says flagged content and content you explicitly report can be used for safety and model improvement. Deletion and the training opt-out are separate actions.
Five checks before your next prompt#
Run these five checks before you paste. If any answer is yes, mask the detail or use a different workflow.
- ✔️ Does the text contain a name, address, phone number, email, customer ID, or another identifier?
- ✔️ Does it contain sensitive personal data such as health, beliefs, criminal history, or sexual orientation?
- ✔️ Does it contain information about a client, colleague, or family member?
- ✔️ Does a contract, NDA, or company policy prohibit sending it to an external service?
- ✔️ Is the service set or contracted not to train on your input?
Personal data versus sensitive personal data#
Personal data is information about a living person that can identify them, such as a name, address, or customer number. Sensitive personal data includes health, beliefs, social status, criminal history, and other categories that can lead to discrimination. The second category usually requires stronger legal grounds and consent.
A contract review can contain personal data even if you never mention a private individual. A client name, address, deal value, and account manager can identify a person or a small company. Removing the name alone may not be enough if the address and company remain.
Third-party data is not yours to decide#
You cannot decide alone whether a client's or colleague's information can go into an external AI service. A confidentiality clause may prohibit it, and your company may have a separate AI policy. Ask before you paste. It is easier to stop before the input than to clean up after it.
How do you mask names and details?#
Replace names, addresses, and account numbers with placeholders, and give the AI only the structure it needs. Then restore the real values outside the AI.
| Do not paste | Paste instead |
|---|---|
| Yamada Trading Co., Mr. Yamada, Shibuya, Tokyo | Company A, Mr. A, a city in Kanto |
| Delivery by November 20, 2026 | Delivery by November 20, 2026 (keep the date if needed) |
| Revenue of 1,234,567,890 yen | Revenue of about 1.2 billion yen |
| Client contact: Sato, 090-... | Client contact (phone number withheld) |
Tell the model that the names are placeholders: "Company A, Mr. A, and Company B are aliases. Keep them as aliases and analyze the clauses." Restore the real names outside the chat. Do not paste the mapping table into the same conversation, or the masking loses its purpose.
For a long contract, extract the structure first, then discuss only the risky clauses. Splitting the work reduces the amount of personal data in one prompt. The same applies to images: check name badges, addresses, customer names, and account names before uploading. See how to ask an AI about an image.
If you use AI for work, what should you choose?#
For work that involves personal data, check the business contract and data processing terms, not just the consumer toggle. Look at five things: whether training is off by default, retention, human review, subprocessors, and data location.
| Option | Best for | Check |
|---|---|---|
| Business plans (ChatGPT Business / Enterprise, Google Workspace, Claude Enterprise) | Internal confidential data | Default no-training, retention, admin controls, contract |
| Consumer plan with training off | Personal research and masked documents | Feedback, safety review, retention exceptions |
| Local model | Data that cannot leave the device | Device security, model license, backups |
Deepent Research says on its site that input is not used for training. The details, including retention, subprocessors, and image deletion, are in the Privacy Policy. It is one option when you want a provider that states a no-training policy.
See also: what to delegate to AI and is uncensored AI legal?.
What is the legal position?#
Japan's Personal Information Protection Commission published guidance on generative AI services on June 2, 2023. It says that a business handling personal information must confirm that the input is within the purpose of use. If personal data is provided to the AI provider for a purpose other than generating the response, and the user has not consented, the Personal Information Protection Act may be violated. The guidance asks businesses to confirm that the provider does not use the data for machine learning.
For individual users, the guidance notes that input personal data can be used for training, combined with other information, and output back to other users. It asks users to check the provider's terms and privacy policy before entering personal data.
Under the EU GDPR, the purpose, legal basis, retention period, and data subject rights matter. Health and other special categories receive stronger protection. US state laws and sector rules add another layer.
There is no single rule that says "pasting data into AI is illegal." The answer depends on the data, purpose, consent, contract, and provider settings. Ask your legal or privacy team about a specific case.
FAQ#
Does deleting the chat remove everything?#
No. Deletion usually removes the conversation from your history and the live system. Training data, feedback, safety review, backups, and legally required records can sit outside that deletion. Check the provider's help page for the exact scope.
Do free and paid plans handle training differently?#
Consumer free and paid plans usually require you to change the training setting yourself. Business plans and APIs often do not train on inputs by default. The same brand can behave differently depending on the plan and contract.
Are temporary chats safe?#
Temporary chats are not used for training while they remain temporary and do not appear in history. If you save one, it becomes a regular chat. Safety review can still apply even to temporary chats.
Is it illegal under Japan's Personal Information Protection Act?#
Not automatically. The issue arises when a business provides personal data to a provider for a purpose beyond generating the response without the required purpose limitation or consent. Individuals should also watch for confidentiality duties and contract violations.
Which AI services do not train on my input?#
Consumer plans usually need an opt-out. Business plans and APIs often default to no training. A local model can keep input on your device. Deepent Research states that input is not used for training; the details are in its Privacy Policy.
Do three things today: delete the conversation you are worried about, check the training setting, and mask names in your next prompt. Those three steps reduce the risk even when you cannot undo the past.
Try it yourself
Deepent Research is an LLM chat service that refuses almost nothing. Everything covered in this note can be tried without downloading anything.
This note is provided for educational and research purposes. It does not encourage illegal activity.
How this note was edited and verified
Our editors check primary sources and organize only what the facts support. What we checked, and what we did not, is listed in the editor's note on each article.
Related notes
What Can You Hand to an AI, and What Should You Keep?
Summaries and background reading are easy to delegate. Personal data and publishing raw output are not. Here is where the line sits, and why, in five checks.
Is Uncensored AI Legal? Risks of Running It for Others (2026)
Is uncensored AI legal? Running it yourself is usually lawful. Running it for other people adds provider duties, privacy and liability.
No-Filter AI: How to Choose and Try One Free
What no-filter AI really means, three checks that separate a usable service from hype, and three prompts that show in ten minutes whether it fits your work.