You wouldn't hand a stranger your diary. Here's why some people accidentally do the equivalent every day — and how to stop.
There's a moment most people don't think about. You're stuck on a problem — a work email you can't figure out how to phrase, a medical question you're embarrassed to Google, a legal clause in a contract you don't understand. You open ChatGPT or Claude or Gemini, type out the full situation with all the details, hit send, and get exactly the help you needed.
The problem isn't that you used AI. The problem is what you typed before you hit send.
Public AI tools have become the place people go when they want honest, non-judgmental help with something sensitive. That's genuinely one of the most valuable things about them. But the openness and convenience that makes them feel safe is exactly what makes certain habits risky. Most people using AI tools every day have no real idea where their conversations go, how long they're stored, whether they can be used to train future models, or what happens if the company behind the tool is breached.
This isn't a reason to stop using AI. It's a reason to use it more carefully. These five rules take almost no extra effort, and they protect you in every conversation, with every tool, from now on.
Rule One: Never Type Real Personal Identifiers Into a Public AI Chat
This is the most important rule, and the most commonly broken one.
Real personal identifiers include your full name combined with other details, your home address, your date of birth, your Social Security number, passport number, driver's license number, bank account or credit card numbers, login credentials, and medical record numbers. Any of these, alone or in combination, is the kind of data that can be used to identify, locate, or impersonate you.
The problem isn't that AI tools are malicious. Most are built by companies that take data privacy seriously and have policies around how conversations are stored and used. The problem is that once you type something into a chat interface, you have no control over what happens to it. Data can be stored in server logs. It can be used for training future model versions unless you specifically opt out. It can be exposed in a data breach. It can be reviewed by human employees for safety or quality assurance purposes. Company policies change, and what's private today may be handled differently after an acquisition or a terms-of-service update.
None of that is hidden. It's in the privacy policies that almost nobody reads.
The fix is simple: anonymize before you type.
If you want help with a legal document, describe the situation in general terms without including the actual document, the real names of people involved, or identifying location details. If you want help drafting a sensitive email, replace the real names with placeholders. "My manager David at [Company Name] sent me this email..." becomes "My manager sent me this email..." You get the same quality of help. You've given away nothing identifiable.
If you need to share something specific — like actual contract language — consider using a secure, private deployment of an AI tool through your company's IT-approved systems rather than a consumer-facing public chat interface. The privacy implications are completely different.
Rule Two: Treat AI Conversations Like Emails, Not Like Thoughts
Here's a mental model that reframes everything: when you type into an AI chat, imagine you're sending an email to a company's customer service department. You'd still get helpful responses. You might even get very good ones. But you wouldn't type your bank PIN into that email. You wouldn't include your medical history. You wouldn't write things you'd be mortified to see appear in a data breach news story next year.
This isn't paranoia — it's the same basic hygiene you apply to any digital communication. AI chat interfaces are not private journals. They are networked software services with server logs, privacy policies, and all the data-handling realities that any software-as-a-service product involves.
The feeling of privacy in an AI chat comes from talking to something rather than someone. It's a one-on-one interface that feels intimate and responsive. That feeling is real in a social sense — the AI isn't going to judge you or gossip about you. But it's not real in a data sense. The conversation lives somewhere after you close the tab.
Practical application of this rule:
Before you type something sensitive, ask yourself: "Would I be comfortable if this appeared in a news article about a data breach at this company?" If the answer is no, rephrase it. If you genuinely need to include that level of detail to get useful help, that's a signal to use a more private channel.
Rule Three: Check Whether Your Conversations Are Being Used to Train the Model — and Opt Out If You Can
Most major AI tools offer some form of control over whether your conversations are used to improve or train future model versions. Most people have never looked for this setting.
This matters because training data usage means your conversations — including any personal details you included — may be reviewed, stored long-term, and incorporated into future model development in ways you can't predict or reverse.
How to find and use this setting:
ChatGPT (OpenAI): Go to Settings → Data Controls → toggle off "Improve the model for everyone." With this off, your conversations in ChatGPT are not used to train OpenAI's models. You can also choose not to save chat history at all, which means conversations aren't stored after the session.
Claude (Anthropic): Anthropic does not use conversations from Claude.ai to train its models by default for paid users. Free-tier users should check the current privacy policy and settings, as these can change. For sensitive work, the paid tier provides stronger privacy defaults.
Gemini (Google): Go to your Google Account → Data & Privacy → Web & App Activity. Turning off Gemini Apps Activity stops Google from saving your Gemini conversations. You can also review and delete conversation history directly from the Gemini interface.
The exact locations of these settings change as companies update their interfaces. The habit to build is checking the privacy settings of any AI tool you use regularly — not just once, but periodically, since policies and options do change.
If a tool doesn't offer any control over data usage or conversation storage, that's important information. For anything sensitive, choose a tool that does.
Rule Four: Be Especially Careful With Medical, Legal, and Financial Information
These three categories deserve their own rule because they're the areas where people are most likely to share detailed personal information — and where the consequences of that information being exposed are most serious.
When you're dealing with a health concern, a legal problem, or a financial stress, the appeal of a private, non-judgmental AI conversation is strongest. That's exactly the moment to be most careful.
Medical information is among the most sensitive category of personal data that exists. It can affect insurance eligibility, employment decisions, and relationships. When asking AI tools about health questions, keep it general. "What are the common causes of chest pain in people in their 40s?" is a question you can ask safely. "I'm [full name], 43, and I've been having chest pain since I started [specific medication] for my [specific condition]" is a different matter — and the second version doesn't produce meaningfully better answers.
Legal information often involves details about specific situations, documents, and parties that you have confidentiality obligations around — especially in a professional context. Lawyers, in particular, should be extremely cautious about pasting client-related content into public AI tools. Many bar associations have issued guidance on this, and the professional consequences of inadvertently sharing privileged client information through a third-party AI service can be severe.
Financial information — account numbers, income details, investment positions, tax information — has obvious exposure risks. Use AI to understand financial concepts, draft general questions, or work through strategies in the abstract. Keep the specific numbers and identifiers out of the conversation.
The pattern in all three cases is the same: ask about the concept, the situation, or the strategy. Don't include the identifying details that make the question personal.
Rule Five: Know Which AI Tool You're Actually Using — and Who Owns It
The AI tool landscape in 2026 is crowded. There are dozens of products that look and feel similar, powered by models from a handful of major labs, distributed through interfaces built by hundreds of different companies. The privacy policies of the underlying model provider and the interface you're using may be completely different.
When you use a third-party app, browser extension, or business tool that's "powered by AI," you're often talking to a model provided by OpenAI, Anthropic, or Google — but your data is first handled by the company that built the interface, not the model provider. That company has its own privacy policy, its own data storage practices, and its own retention policies.
Questions to ask before using any AI tool:
- Who made this product, and who provides the underlying AI model?
- Where can I find the privacy policy, and does it clearly explain how my conversations are stored and used?
- Is this tool aimed at businesses or consumers, and does it have enterprise-grade privacy protections?
- Has this company had any reported data breaches or privacy incidents?
A quick web search for "[tool name] privacy policy" and "[tool name] data breach" takes two minutes and can tell you a lot.
Also be alert to unofficial-looking AI tools that appear as browser extensions, Chrome apps, or mobile apps promising free access to premium AI models. Some of these are legitimate third-party integrations. Some are harvesting conversation data with minimal or deceptive privacy practices. The rule of thumb: if you can't find a clear privacy policy from a named company with a real website, don't type anything sensitive into it.
The Bigger Picture: AI Tools Are Getting More Integrated, Not Less
These rules aren't a response to some unusual or extreme risk. They're a response to where things are heading.
AI tools are becoming embedded in email clients, word processors, browsers, operating systems, and workplace software. The number of places where your words can be processed by an AI model — often without you explicitly choosing to use one — is growing every year. The habit of thinking about what you're sharing, with which tool, under what privacy terms, is becoming as essential as knowing not to click suspicious links in emails.
The people who will navigate this well aren't the ones who avoid AI tools entirely — those tools are too useful to abandon. They're the ones who develop a basic reflex around what they share, with whom, and under what conditions.
That reflex is these five rules:
1. Never type real personal identifiers into a public AI chat
2. Treat AI conversations like emails, not like thoughts
3. Check your data training settings and opt out where you can
4. Be especially careful with medical, legal, and financial details
5. Know which tool you're actually using and who owns it
None of these require technical expertise. They require the same instinct you'd apply to any communication tool — knowing that digital convenience and digital privacy are different things, and choosing to protect both.
Do you have questions about a specific AI tool's privacy settings? Drop them in the comments — privacy settings change frequently, and we'll do our best to point you in the right direction.
0 Comments