Last Updated on August 11, 2026 by Jade Artry
Quick points
At a glance
Never share: Passwords, financial credentials, government ID numbers and identifiable medical records.
Redact first: Client names, case details, and identifying numbers on documents you upload.
Generally fine: General questions, public information, and non-identifying context.
If you've overshared: Change any exposed credentials immediately, then delete the conversation.
How Much Personal Information Should You Share With AI?
The simplest rule is to give an AI only the information it actually needs to answer your question, and remove identifying detail wherever you can. I find this more useful than trying to memorise every provider's privacy policy, because it still works when the tool, model or settings change. Context is often genuinely useful. Identifying detail usually isn't.
What Personal Data Can AI Collect From Files, Photos and Other Uploads?
Files, photos and screenshots can contain far more personal information than the text you deliberately type, including metadata and identifying details you may not notice before uploading them. It's not only what you type that matters; most people think about AI privacy purely in terms of the words they type into the box.
- Uploaded documents often carry more than the visible text, including author names in file properties and comments left by other people.
- Photo metadata can include the time, device and sometimes the location a photo was taken, depending on your phone's settings.
- Screenshots frequently capture names, email addresses or account details sitting in the corner of a window you didn't mean to share.
- Browser-connected context, where a tool has access to your browsing, can pull in far more than the specific page you meant to reference.
- Pasted email threads usually carry the full sender and recipient chain, not just the message you're asking about.
- Track changes and comments in a document can reveal earlier drafts, other authors, or internal discussion never meant to leave the document.
- Location clues buried in casual descriptions, like naming a specific local landmark, can narrow down where you live faster than you'd expect.
- Other people's data folded into your own question, such as a colleague's number in a spreadsheet you're asking about.
As a woman, location and routine details are the ones I'm personally most careful with, more than most other categories on this list. A specific gym, a regular walking route, or which train I get isn't sensitive information in the abstract, but it's exactly the kind of pattern that matters if it ends up somewhere I didn't intend it to. That's the lens I bring to this list, and it's worth thinking about which categories matter most to your own situation, not just treating the list as generic.
Why Do People Share Personal Information With AI Chatbots?
People often overshare with AI because a conversational interface lowers our guard in a way a blank form doesn't. AI feels responsive, interested and private, even when none of those things are technically true.
There's a growing body of research into privacy and self-disclosure with anthropomorphic and companion-style AI that treats this as a genuine human-computer interaction pattern rather than simply users being careless. The more a system feels like it's listening, the more naturally people disclose, in the same way people say more to a warm, attentive stranger than they would fill out on a form asking the identical questions. Knowing that about yourself is useful. It means the moment to be deliberate about what you share is before you start typing, not after you notice you've said more than you meant to.
How to Redact Personal Information Before Sharing It With AI
To redact something before sharing it with AI, remove identities, third-party information, credentials, irrelevant context and hidden information in any attachments. Here's the sequence I actually use.
- Remove identity. Strip out your name, address, date of birth and any other detail that ties the content specifically to you.
- Remove third parties. Take out anyone else's name, contact details or personal information folded into your question.
- Remove credentials. Passwords, account numbers, security codes, anything used to access an account.
- Remove irrelevant context. If the AI doesn't need it to answer the question, it doesn't need to see it.
- Check attachments before upload. A document can carry far more than the paragraph you meant to ask about; skim it for anything on the list above before it leaves your device.
What Information Is Safe to Share With AI?
What's actually safe to share includes public information, anonymised scenarios and non-sensitive drafts, though this page is understandably heavy on what to avoid, so it's worth being equally clear about what doesn't need a second thought.
- Public information that's already freely available, like a company's published opening hours.
- Anonymised scenarios, where you've already stripped out the identifying detail.
- Fictional or example data used to illustrate a point rather than describe a real situation.
- Generic questions that don't reference your specific circumstances at all.
- Non-sensitive drafts, like a first pass at a blog post or a birthday message.
- Properly redacted documents, once you've worked through the steps above.
The goal isn't to be afraid of AI chatbots. It's to know which category something falls into before you hit send.
What Should You Never Share With AI?
You should never share passwords, financial credentials, government ID numbers, confidential work information or anything that identifies a child. Most of this is common sense once it's spelled out, but it's easy to forget in the moment when a chatbot feels like a helpful assistant rather than a company's server.
Passwords and authentication codes
If you're troubleshooting a login, ChatGPT doesn't need the credential to understand the problem. Never paste a password, PIN, security question answer or two-factor code into any AI chat. Describe the problem instead.
Full financial or payment credentials
Full bank account numbers, card details and payment credentials should stay out of a chatbot. If the amount itself matters to the question, provide the figure without the identifying account information.
Government identity numbers
National Insurance numbers, passport numbers and driving licence numbers should stay out of any AI conversation entirely. These are exactly the kind of static, hard-to-change identifiers that cause lasting harm if they end up in the wrong dataset, unlike a password you can simply reset.
Confidential client or work information
If work information is confidential, only use an AI service your organisation has explicitly approved for that type of data, and follow its own handling policy. A business account offers stronger contractual protections, but it doesn't make every confidential document appropriate to upload. See our guide on verifying clients and suppliers in the age of AI for a related business risk worth knowing about.
Other people's private information
Sharing someone else's address, medical details or private messages without their knowledge is a privacy issue independent of AI, since they never had the chance to consent to their information being processed by a third-party service on your behalf.
Identifiable information about children
Full names, schools and routines belonging to children are worth keeping general with any AI tool, for the same reason you'd be cautious sharing this anywhere else online.
Intimate or private imagery
Never upload intimate images of another person without their consent. Sexual images involving children can constitute child sexual abuse material and should never be uploaded to an AI service; if you're dealing with an existing image or incident involving a child, use current NSPCC or CEOP guidance rather than trying to process it through AI.
Identifiable medical records
If you want help understanding a blood result, the useful information is usually the result itself, not your full name, address and NHS number sitting above it. Asking a general health question is different from uploading an identifiable document; the question is fine, the document usually isn't, unless you remove the identifying details first.
What Information Should You Share, Redact or Never Give to AI?
Share general or public information freely, redact personal documents and real-world scenarios, and never share credentials, government ID numbers or identifiable information about children. When you're not sure, it's usually a quick answer.
| Information | DSS recommendation |
|---|---|
| A general question or public information | Fine |
| Your CV | Redact unnecessary identifiers like date of birth and full address |
| A contract you're reviewing | Remove client names and confidential terms first |
| A medical result | Remove your name, NHS number and other identifiers where possible |
| A bank statement | Don't upload the full identifiable document; describe the relevant figures |
| A password or 2FA code | Never |
| A passport or National Insurance number | Never |
| A child's location or daily routine | Never share the identifiable version |
How to Anonymise Information Before Using AI
To anonymise information before using AI, replace real names and identifying details with general labels while keeping only the facts the model needs to answer your question. The trick is asking which details the AI actually needs to produce a useful answer.
- Work example: instead of ‘John Smith at 14 Elm Road owes us £3,200 under Contract 2026-118', try ‘Client A owes £3,200 under a service agreement. How should I phrase a polite follow-up?'
- Medical example: remove your name, NHS number, address and date of birth before pasting in a letter or result, keeping only the clinical detail you actually want explained.
- Family example: replace your child's name, school and specific location with a general description, such as ‘my 10-year-old' rather than identifying detail.
- Financial example: give the relevant figures and the actual problem rather than a full account number or sort code.
Is It Safe to Share Sensitive Data With ChatGPT Plus, Claude Pro or Other Paid AI?
No, not on its own. Paying for a tool can make it feel as though you've moved into a more private version of the service. With consumer AI accounts, that isn't necessarily what you've paid for. A paid consumer account, such as ChatGPT Plus or Claude Pro, isn't the same thing as enterprise or business confidentiality. Genuine contractual protection against training and stronger data handling generally only arrives with a business or API tier under a signed agreement.
What Should You Do if You Shared Sensitive Information With AI?
If you've already shared something sensitive, delete the conversation where appropriate and then deal with the underlying exposure, whether that means changing a password, securing a financial account or following your organisation's incident process. Deleting the chat isn't always the most important action; the severity of the response should match the severity of what you exposed.
- Password or security credential: change it straight away.
- Financial credential: secure the account and contact your provider if needed.
- Confidential work information: follow your organisation's incident procedure.
- Identity information: assess the exposure and take proportionate protective action.
- An ordinary personal detail: delete the chat and review your settings, but there's no need to panic.
For the practical steps to actually remove the data, see how to delete your data from AI chatbots.
Which AI Chatbots Give You Stronger Privacy Controls?
Claude currently gives consumers stronger model-training consent than ChatGPT, Gemini or Perplexity, while privacy-first tools such as Proton Lumo and Duck.ai go further by changing how data is handled by design. See our individual breakdowns of ChatGPT, Claude, Gemini and Perplexity, the full comparison in Are AI Chats Private?, Best Private AI Chatbots if privacy is your main priority, or the AI Safety hub for more on AI safety.
What Information Should You Never Share With AI? The Bottom Line
If you remember one thing from this guide, make it this: AI doesn't need to know everything you know.
I use these tools constantly, and I still strip out names, addresses, account details and other context whenever it isn't necessary to the task. It takes seconds, works across every chatbot, and gives you a layer of protection that doesn't depend on a company's settings staying the same.
