AI tools can help draft emails, summarise documents and answer everyday questions. Knowing how to protect your personal data while using AI tools starts with treating each prompt or upload as information disclosed to a service—not as a private note. What happens to submitted information depends on the provider, product, account and settings. Some services may save conversations, use them to improve services or allow human review. A few practical habits can lower your exposure: share only what is needed, check privacy controls before you begin, and avoid entering information you are not authorised to disclose.
1. Check the provider’s privacy terms and settings
Before using an AI service, review its privacy policy, help centre or data-controls page. Check what happens to prompts, uploaded files and conversation history. Find out whether chats are saved, used to improve models or other services, reviewed by people, or covered by deletion and temporary-chat controls.
Do not assume a setting such as “training off” means that nothing is stored or processed. It may affect only future conversations or model-improvement use, without changing chat history, safety review or retention. Read the provider’s explanation and confirm whether the setting applies to your account type.
For example, OpenAI says ChatGPT users can turn off “Improve the model for everyone” in Data Controls. Its FAQ says new conversations then remain in chat history but are not used to train its models. OpenAI also says Temporary Chats do not appear in history, create memories or train models, but a copy may be retained for up to 30 days for safety purposes. These details are product-specific, so check the current guidance before relying on them: OpenAI Data Controls FAQ and Temporary Chat FAQ.
Google’s Gemini Apps Privacy Hub explains that Activity settings affect whether chats are saved and used to improve Google services, and that some conversations may be reviewed by human reviewers. Google advises users not to enter confidential information they would not want a reviewer to see or Google to use to improve its services. See the current Gemini Apps Privacy Hub for details.
2. Share the minimum information needed
Data minimisation means providing only what is necessary to get useful help. Often, you can remove identifying details and still ask the same question. Instead of pasting an entire email thread, briefly describe the situation. Instead of uploading a document with names and contact details, share only the relevant, anonymised passage.
Before submitting a prompt or file, remove or replace:
- Names, email addresses, phone numbers, and home or work addresses.
- Account numbers, customer IDs, reference numbers, and other identifiers.
- Precise locations, dates, or details that could identify someone when combined.
- Document metadata or visible details that are not needed for the task.
Personal data includes information that identifies someone or can be linked to them. A name is an obvious example, but several indirect details may also point to a person. Replacing a name with “Person A” may not be enough if the rest of the prompt still reveals who they are. The FTC recommends limiting personal information shared online and reviewing privacy settings—useful baseline habits when deciding what to share with an AI service. See the FTC’s privacy guidance.
3. Keep secrets and sensitive records out of prompts
Do not enter passwords, one-time authentication codes, private access keys, or security answers into an AI prompt. Treat financial details, medical records, identity documents, and confidential business information with similar care. Do not submit another person’s private information unless you have permission and a valid reason to use that particular service.
A tool may be appropriate for sensitive information only if the provider, account, and specific use have been explicitly approved. A privacy setting on a personal account is not a substitute for approval from an employer, school, client, or other organisation. If you are unsure whether you may share a document, ask the responsible person or use an approved alternative.
If a task can be completed with a fictional example, summary, or redacted extract, use that instead of the original record. For example, you can ask for help improving a message’s tone without including the recipient’s real name, contact details, or confidential background. The less sensitive information you submit, the less you need to rely on settings to protect it.
4. Use chat controls carefully and secure your account
When available, use controls that limit model-improvement use or offer a temporary chat for work that does not need to be saved. First check what each control changes. Temporary chat is a product-specific mode, not a universal promise that a conversation is never retained. Turning off model improvement may not disable all storage, safety processes, or human review. Provider policies and settings can change, so recheck them before using a service for a sensitive task.
Protect your account as well. Use a unique, strong password and enable multifactor authentication where available. Avoid signing in to sensitive AI work on a shared device, and do not leave your account open where someone else can access conversation history. These steps protect account access; they do not determine how the provider handles information after you submit it.
After using a service, review saved chats or activity and delete conversations when appropriate. Do not assume deletion is immediate or universal: rules vary by product, account, and setting. Check the provider’s current documentation to understand what its controls do and what may remain after a chat is removed.
5. Follow work, school, and client policies
If you handle information for an employer, school, client, or community organisation, use only tools and accounts approved for that information. A personal account may have different settings and protections from an organisation-managed account. Do not upload work files simply because a tool accepts them.
Organisations should establish clear rules about which information can be entered into which tools, who approves exceptions, and how saved conversations are handled. NIST describes its Privacy Framework as a voluntary tool to help organisations identify and manage privacy risks. It can support broader information-handling practices, but it does not certify that a particular AI service is suitable for every organisation or use. See the NIST Privacy Framework.
Pause before sharing information that is confidential, regulated, client-owned, or otherwise restricted. Confirm the approved service and account first. If no approved option is available, ask whether a redacted example or a non-AI workflow can meet the need.
6. Remember that privacy and accuracy are separate
Privacy safeguards do not make an AI-generated answer accurate, fair, or appropriate. Check important facts against reliable sources, especially before making financial, health, legal, employment, or safety decisions. Check whether personal information in a response belongs in another service or document before copying it there.
Consider what your prompt reveals through context, too. A question about a rare condition, a small team, or a specific incident may identify someone even without a name. When possible, rephrase the request at a more general level. This can protect privacy while still giving the tool enough context to produce a useful response.
Quick pre-submit checklist
Before you press send or upload, ask:
- Is every detail necessary for this task?
- Could the prompt or file identify me or someone else, directly or indirectly?
- Have I checked the provider’s current privacy settings, retention information, and review practices?
- Am I authorised to share this information with this tool and account?
- Could I use a redacted, shorter, or fictional example instead?
Conclusion
Protecting personal data while using AI tools is mainly about making deliberate choices before submitting information. Check the provider’s current terms and controls, share only what the task requires, and keep passwords, sensitive records, and confidential material out unless the tool and use are explicitly approved. Secure your account and review saved conversations, but do not treat settings or deletion buttons as guarantees of immediate, complete removal. For work or school information, follow the organisation’s policy. A short pause before sending can help you benefit from an AI tool without disclosing more than you intended.
Frequently asked questions
Is it safe to put personal information into an AI tool?
It depends on the information, provider, account, and settings. Treat prompts and uploads as disclosures, check the current privacy terms, and avoid sharing unnecessary or sensitive details.
Does turning off model training stop an AI service from saving my chats?
Not necessarily. A model-improvement setting may not change chat history, retention, safety review, or other processing. Read the provider’s explanation of the specific control.
Are temporary chats completely private?
Not as a general rule. Temporary-chat protections and retention periods vary by product. OpenAI says a copy of its Temporary Chats may be retained for up to 30 days for safety purposes; check the current product guidance.
What information should I never enter into an AI prompt?
Avoid passwords, authentication codes, private access keys, and sensitive records. Do not submit confidential work, medical or financial information, or another person’s private data unless the service and use have been explicitly approved.
Can I use an AI tool for work documents?
Use only a tool and account approved by your organisation for that kind of information. If approval is unclear, ask first or use a redacted example without confidential details.
Should I delete my AI conversations after using the tool?
Review and delete saved conversations when appropriate, but do not assume deletion is immediate or removes every retained copy. The effect depends on the provider’s policies and product controls.