Imagine you’re applying for a job and want a chatbot to polish your CV. You upload the file, ask for a stronger introduction and hit send. Only afterwards do you notice what else you included: your home address, phone number and your references’ contact details. None of that was needed to improve the wording.
It’s easy to share too much when you’re focused on getting something done. Understanding the things you should never share with an AI chatbot helps you decide what to keep private without losing a useful answer.
This guide covers 10 types of information to keep private, practical ways to ask questions with fewer personal details, and steps to take if you’ve already shared something sensitive. The focus is everyday use of general-purpose chatbots. Authorised work in an organisation’s approved AI system is a different situation, governed by its own rules about what you can upload.
Table of Contents
- Before You Paste: Does the Chatbot Need That Information?
- 10 Things You Should Never Share With an AI Chatbot
- Sensitive Information to Remove Before Uploading Files to AI
- What Happens to Information You Share With a Chatbot?
- Safer Prompts: Three Practical Examples
- What to Do If You Have Already Shared Sensitive Information
- AI Chatbot Privacy Checklist: What Not to Share
- Frequently Asked Questions
- Make Sharing Less Information Your Default
Before You Paste: Does the Chatbot Need That Information?
To recognise things you should never share with an AI chatbot, ask yourself these questions before uploading anything:
- Could this identify me or someone else?
- Could it give someone access to an account or system?
- Am I authorised to share it?
- Could a placeholder or fictional example do the job?
In our hypothetical CV example, the chatbot needs the person’s experience and the role they want. Their street address adds nothing to the editing task.
The UK’s National Cyber Security Centre advises against including sensitive information in queries to public language models. That is a useful starting point whenever you feel unsure about an upload. ncsc.gov.uk
Privacy concerns how information is collected, used and shared. Security concerns protecting it from unauthorised access or misuse. A secure account still needs sensible privacy choices.
10 Things You Should Never Share With an AI Chatbot
1. Passwords, One-Time Codes and Account Recovery Details
Passwords are among the things you should never share with an AI chatbot, even when you need help with a login problem. Keep passwords, security PINs, one-time passcodes, recovery codes and password-reset links out of conversations.
These details can grant access to an account. Even a code that expires quickly should stay within the login or recovery process that requested it.
Instead, describe the problem: “I receive an incorrect-password message after resetting my password.” If you attach a screenshot, remove visible credentials and recovery links first.
2. Bank, Card and Payment Account Information
Avoid uploading complete card numbers, CVV codes, banking credentials or unredacted statements. Statements can reveal your address, account identifiers and a detailed record of where you spend money.
You can still ask for help organising a budget. Use broad categories such as rent, groceries and transport, with rounded amounts where exact figures are unnecessary.
For example: “My monthly income is approximately £2,500, and these are my main expenses.” The chatbot can help structure those figures without seeing the account behind them.
3. Government ID Numbers and Identity Documents
Passport scans, driving licences and tax documents contain more information than most chatbot tasks require. The same caution applies to US Social Security numbers, UK National Insurance numbers and Australian tax file numbers.
An identity document may combine a photograph, signature, date of birth and identifying number in one file. This combination puts unredacted ID documents among the things you should never share with an AI chatbot.
If a form confuses you, share the blank question instead. “What does ‘country of tax residence’ mean?” does not require uploading your completed application. Use official guidance or qualified advice when the answer affects an important decision.
4. Your Exact Address, Live Location and Daily Routine
Location details deserve particular care when combined with your name, workplace or family information. Travel dates and daily routines can reveal when you are away or where someone can find you.
For most local questions, a city or neighbourhood is enough.
Ask for “weekend activities in Manchester” rather than supplying your full address and explaining that your home will be empty. If a service genuinely requires an address, provide it through that service’s appropriate process.
5. Identifiable Medical and Mental Health Records
Medical reports, prescriptions and therapy notes may contain names, patient numbers, birth dates and deeply personal histories. Uploading the complete record can disclose much more than the question requires.
You can ask what a medical term means without attaching an identifiable report. Where context is useful, keep it limited and avoid unnecessary identifying details.
Discussing a general health concern is different from handing over your medical history. A chatbot’s explanation also should not become the sole basis for diagnosis, treatment or an urgent health decision.
6. Confidential Work Documents and Business Plans
Internal contracts, unpublished results, client proposals, proprietary code and meeting transcripts may belong to your employer or its customers. Having access to a document does not automatically authorise you to upload it elsewhere.
Check the organisation’s policy and use approved systems for approved tasks.
Australia’s Office of the Australian Information Commissioner recommends minimising personal information entered into AI products. Its organisational guidance also recommends avoiding personal information, particularly sensitive information, in publicly available generative AI tools. OAIC
For general writing help, a fictional brief may be sufficient.
7. Customer, Employee and Other People’s Personal Data
A customer spreadsheet or support conversation can include names, delivery addresses, phone numbers and order histories. Employee files may reveal pay, performance or health information.
The key question is not just whether you are comfortable sharing it. It is whether you have authority to disclose someone else’s information through that tool.
For a reply draft, use [Customer] and summarise the issue. Keep the real order details in the authorised customer system, then add them to the finished response yourself.
8. API Keys, Private Keys and Technical Secrets
API keys and access tokens allow software to use services. Depending on their permissions, exposed credentials can enable access to data or paid resources. Crypto wallet recovery phrases can provide control over funds.
Keep these secrets out of prompts, configuration files and debugging screenshots.
Replace real values with [API_KEY_REMOVED] before asking for coding help. Check error logs too: they sometimes contain credentials or private URLs. If a secret has already been exposed, replacing it in your next message does not revoke the original.
9. Private Messages, Intimate Content and Identifiable Personal Stories
A private conversation may reveal someone’s relationship problems, sexuality, finances or family circumstances. Removing their name may not remove the identifying context.
You can ask for help wording a difficult message without uploading the entire conversation. Describe the situation broadly and include only the sentences necessary for the task.
For instance: “Help me politely decline a personal request” may be enough. Avoid sharing intimate images or someone else’s private disclosures simply because a chatbot offers to analyse them.
10. Children’s Identifying Details, Photos and Schedules
Children’s school reports and photographs can expose school names, birth dates, uniforms, addresses or regular pickup arrangements. An ordinary-looking image may contain several of these details together.
For learning support, start with an age range, subject and difficulty.
“Suggest reading activities for a child aged seven who struggles with longer words” provides useful context without a full name, school or identifiable report. Ask whether the image or document is necessary before uploading it.
Sensitive Information to Remove Before Uploading Files to AI
Things you should never share with an AI chatbot can also be hidden inside an otherwise ordinary file. Check:
- Screenshots: browser tabs, notifications, email addresses and account numbers.
- Documents: comments, tracked changes and author information.
- Spreadsheets: hidden sheets, extra columns and customer identifiers.
- Photos: visible badges, addresses and location metadata where present.
- Earlier messages: details that become identifying when combined.
A black rectangle placed over text does not necessarily remove the underlying information. Use proper redaction or create a clean extract, then inspect what you plan to upload.
In the CV example, removing the address is only part of the job. The references’ details and document comments need checking too.
What Happens to Information You Share With a Chatbot?
Sending a message, storing it, using it for training and publishing it are different actions. Do not assume that every prompt becomes public—or that a private-looking chat is processed only on your device.
Check the provider’s policy for your product, account and settings.
For example, OpenAI’s data controls documentationTemplates says switching off “Improve the model for everyone” prevents new conversations from being used for training. Chats can still remain in history. Submitting feedback may allow the associated conversation to be used for training. Temporary chats are not used for model improvement but may be retained for up to 30 days for safety. OpenAI Help Center
These controls serve different purposes. A training opt-out is not a deletion request.
Connected Apps, Fake Tools and Misleading Instructions
Before connecting an inbox or cloud drive, review what the tool can access and whether that access is necessary. Start with the smallest useful scope.
Prompt injection happens when instructions inside material such as a webpage or document try to redirect an AI system. Researchers describe how this can lead to unauthorised tool use or disclosure in vulnerable systems. Review unexpected requests to send information or perform actions. arxiv.org
Use the provider’s official website to find its apps. Avoid entering details into an unfamiliar chatbot promoted through an unsolicited message. AI-generated scam messages can sound convincing; verify unusual requests through a separate, trusted channel.
Protect your account with a unique password and multifactor authentication where available.
Safer Prompts: Three Practical Examples
Let’s return to our hypothetical job applicant. Later, they use AI for a customer reply and a work summary. These examples show how to get useful help while leaving out things you should never share with an AI chatbot.
Improve a CV Without Uploading Your Full Profile
Before:
“Rewrite my attached CV.” The file includes an address, phone number and references.
Safer version:
“Improve this experience section for a customer-support role. Keep the facts accurate and do not invent achievements. I have removed contact details and references.”
Draft a Customer Reply Without Exposing Records
Before:
“Reply to this customer,” followed by an entire support thread.
Safer version:
“Draft a polite reply to [Customer] about a delayed order. Explain that we are checking with the courier. Do not promise a delivery date or refund.”
The applicant can add authorised account details afterwards.
Understand a Document Without Sharing the Entire File
Before:
“Summarise this internal report,” with confidential attachments included.
Safer version:
“Summarise this approved, redacted extract in five bullets. Flag unclear points and do not infer missing figures.”
Before using the result, compare it with the source. Check names, amounts, dates and commitments. Less data reduces unnecessary disclosure; it does not guarantee an accurate answer.
What to Do If You Have Already Shared Sensitive Information
Pause and identify what you shared, where you shared it and whether any external service received it. Accidental sharing does not automatically mean someone has misused the information, but the response should match the potential consequence.
| Information or risk | Practical next step |
|---|---|
| Password or recovery credentials | Change the affected password, review sessions and replace exposed recovery codes where possible. |
| API key or access token | Revoke or rotate it through the issuing service; check usage. |
| Card or banking information | Contact your provider through its official app or published number for advice. |
| Work or customer records | Promptly notify the relevant internal security or privacy contact. |
| Identity documents | Seek official guidance appropriate to your country and the information exposed. |
| Sensitive chats or uploads | Review deletion options, saved files, shared links and connected services. |
If a scam is involved, the US Federal Trade Commission’s recovery guidance explains actions for compromised information and payments. Contacting the payment provider promptly may help, although recovery is not guaranteed. consumer.ftc.gov
Deletion also has limits. OpenAI’s retention policy says deleted chats are scheduled for permanent deletion within 30 days, subject to stated exceptions. Files saved separately in Library must be deleted separately; removing the chat does not remove those files. Other providers have different rules. OpenAI Help Center
AI Chatbot Privacy Checklist: What Not to Share
Use this checklist to spot things you should never share with an AI chatbot before pressing send:
- Remove unnecessary names, addresses and identifiers.
- Check for passwords, payment details and technical secrets.
- Inspect the whole attachment, not just the visible paragraph.
- Confirm authority to share information about others.
- Use the approved tool for workplace tasks.
- Review connected-app permissions.
- Replace real details with placeholders where possible.
- Plan to check the answer before acting on it.
When a chatbot asks for more information, reconsider whether it needs that information. You can narrow the question instead.
Frequently Asked Questions
What Sensitive Information Should You Never Share With ChatGPT?
Start with login credentials, financial account details, government identifiers, identifiable medical records and confidential work information. Other categories matter too, especially information about children or people who have not authorised its disclosure.
Is It Bad to Tell ChatGPT Your Secrets?
You do not need to avoid every personal conversation. Consider how sensitive the details are, who they identify and how you would feel if they were accessed unexpectedly. Broad descriptions can support useful discussion without revealing the full story.
What Information Is Usually Okay to Share, Including Health or Money Questions?
Preferences, fictional examples, broad goals and non-identifying questions are generally better starting points. You can ask what a medical term means or how budgeting works without sharing records. Personal medical and financial decisions may require qualified advice.
Can Other Users See My Chatbot Conversations?
Sending a message does not automatically publish it for other users. However, shared links, shared workspaces, account access and provider practices can affect who sees information. Check the service’s sharing controls rather than assuming every conversation has the same audience.
Does Deleting a Chat Remove Every Copy?
Not necessarily. Retention exceptions, separately stored uploads and information already sent to external services can complicate deletion. Follow the provider’s current instructions for the chat, files and any other relevant storage. Deletion does not invalidate an exposed password.
Is Removing My Name Enough to Make a Document Anonymous?
No. A job title, employer, location, exact dates or unusual event may still identify you. Review combinations of details, not only names. Describe a record as redacted rather than anonymous unless you have good grounds for that stronger claim.
Make Sharing Less Information Your Default
Knowing the things you should never share with an AI chatbot makes it easier to decide what to leave out. Often, a short extract and a clear instruction are enough.
Start with less information, add context carefully and review the answer before using it. If you are unsure whether something belongs in a chat, leave it out while you check.
Explore AI Security Guides for more practical explanations of safer everyday AI use.
This article provides general educational information, not personalised legal, medical, financial or incident-response advice.
Good explain