If your parents have discovered ChatGPT in the last year — and statistically, at least one of them has — they are probably using it the way most people do: asking medical questions, pasting in emails to “make this sound nicer,” asking it to summarize bank statements, and getting it to draft replies to family group chats. None of this is malicious. All of it is potentially a privacy problem. This guide explains what actually happens to a chat log after your parent hits send, which AI tools are safer than others, and the three settings you need to flip on their account the next time you visit.
The short version: the AI itself is not the threat most people imagine. The threat is what the company behind the AI does with the chat logs — for training, for review by contractors, for advertising attribution, and for the inevitable breach three years from now. The good news is that the major providers now let you opt out. The bad news is that the opt-out is buried three menus deep and your parents will never find it on their own.
What actually happens when your parent sends a message to ChatGPT
When your parent types “I have stage 3 diabetes, what should I eat?” into ChatGPT, that message does not just disappear into a void. It takes a journey:
- Transmission: the message goes to OpenAI’s servers over an encrypted connection. This part is fine — encryption in transit is universal now.
- Storage: the message, and the AI’s reply, are saved to your parent’s account history. They persist for 30 days by default, longer if “memory” or “chat history” is enabled. Many users never delete these logs.
- Training queue: if your parent is on the free tier and has not opted out, the chat is eligible to be sampled into future model training runs. OpenAI has stated they remove direct identifiers, but the conversations themselves — symptoms, family details, financial situations — become part of the model’s training corpus.
- Human review: a small fraction of conversations are reviewed by human contractors for quality and safety purposes. This is rare, but it happens, and the contractors are not always in your country.
- Aggregate analytics: the company tracks topics, usage patterns, and engagement metrics for product improvement and advertising decisions.
Steps 3 and 4 are the ones that should give you pause. A stage 3 diabetes question is one thing; a chat log that includes a parent pasting in a bank statement to “help me understand this” is another. Medical questions, financial documents, family drama, and political opinions are all flowing into training pipelines right now.
The three settings you need to flip
The single most useful thing you can do for your parents’ AI privacy is sit with them for ten minutes and change three settings. These are not hidden in the sense of “secret developer menus” — they are hidden in the sense of “your parent will never click through four layers of account settings on their own.”
Setting 1 — Turn off chat history and training (ChatGPT)
Have your parent log in to chat.openai.com, click their profile picture in the bottom-left, and go to Settings → Data Controls. Toggle off both:
- Chat history & training — this stops new chats from being used for model training. The trade-off is that chats are no longer saved between sessions, so your parent cannot resume an old conversation. For most elderly users, this is fine; they do not return to old threads anyway.
- Improve the model for everyone — turn this off too. It is a separate signal that contributes to evaluation pipelines.
If your parent has a ChatGPT Plus or Pro subscription, the model is not used for training by default, but turning off chat history still prevents long-term storage of sensitive conversations.
Setting 2 — Use Claude with the “no training” guarantee
If your parent is open to switching tools, Claude (claude.ai) from Anthropic has a stronger default privacy posture than ChatGPT for free-tier users: Anthropic does not train on user conversations by default, full stop, on any tier. There is no setting to flip — the protection is built in. For elderly users who do not care which logo is on the page, Claude is currently the more privacy-respecting default choice.
The trade-off: Claude’s free tier has stricter rate limits than ChatGPT’s. Your parent will hit “you have reached your message limit” more often. For someone using AI a few times a week, this is fine. For someone who has integrated it into daily routines, it is annoying.
Setting 3 — Turn off Gemini “Apps Activity” (Google)
If your parent uses Gemini inside their Google account — which they probably do if they have a Gmail address and use Google on Android — the Gemini activity is being saved to their Google Activity log by default, the same place their search history lives. This means their Gemini conversations are visible to anyone with access to their Google account, and they are subject to Google’s broader data retention policies.
To turn it off: have them go to myactivity.google.com → Gemini Apps → turn off. They can also delete prior Gemini activity from the same screen. This is the single biggest Google-specific privacy lever for AI use.
What is actually safe to share with an AI?
The pragmatic answer is not “nothing.” Your parents can get real value from AI — draft letters, summarize long articles, explain confusing medical test results in plain language, plan a trip — without leaking anything dangerous. The skill is knowing what to type and what to keep out. A simple rule of thumb:
Safe to share with any AI tool:
- General questions about health, nutrition, and medications, framed generically (“What are common side effects of metformin?”)
- Drafts and revisions of personal letters, emails, and messages, with names and addresses removed
- Educational questions about history, science, technology, and current events
- Recipe ideas, travel planning, book recommendations, hobby advice
- Long articles pasted in for summary, as long as they are not internal company documents
Never safe to share with a free-tier AI tool:
- Full bank statements, credit card numbers, Social Security numbers, or passport scans
- Medical records that include your parent’s full name, date of birth, and provider details
- Legal documents — wills, power-of-attorney papers, contracts — that contain identifying information
- Anything your parent would not be comfortable having read aloud in a public meeting
If your parent absolutely needs help with a sensitive document — a confusing hospital bill, a complex insurance letter — they have two good options. They can paste the text with all identifying details redacted (replace names with “[NAME]”, dates with “[DATE]”, account numbers with “[ACCT]”). Or they can use a paid tier of ChatGPT or Claude, where the no-training guarantees are stronger, and accept the residual risk. What they should not do is paste a raw hospital bill into a free ChatGPT window and trust that everything will be fine.
Why the “AI is reading your data” panic is overblown — but the training risk is real
The viral framing — “ChatGPT is reading everything you type!” — is misleading. ChatGPT is not a person reading your chats in real time. It is a language model that processes text and returns a response. There is no human employee of OpenAI watching your parent ask about their blood pressure medication.
But the viral framing points at a real problem: the chat logs are saved, they are used to train future models, and a small fraction are reviewed by humans. The risk is not “an AI is judging me right now.” The risk is “my parent’s medical question is now in a training corpus that may leak in a breach two years from now, may surface in another user’s chat response, or may be used to target advertising indirectly through aggregate signals.” That risk is low per individual chat but real across thousands of chats over years of use.
This is why the three settings above matter. They do not make AI perfectly private — nothing connected to the internet is perfectly private. They reduce the surface area from “every conversation your parent ever has with the AI is permanently in the training pipeline” to “conversations are ephemeral and not used for training.” That is a meaningful, pragmatic improvement.
FAQ
Does ChatGPT use my parents’ conversations to train its models?
Yes, by default, on the free tier. OpenAI’s terms allow free-tier conversations to be sampled into training data. ChatGPT Plus and Pro subscriptions have training disabled by default. The free-tier training can be turned off by disabling “Chat history & training” in Settings → Data Controls.
Is Claude more private than ChatGPT?
For free-tier users in 2026, yes. Anthropic does not train on Claude conversations by default on any tier. ChatGPT free-tier training is on by default. The privacy posture can change, so verify the current policy before relying on it for sensitive use.
Can I delete my parents’ old ChatGPT conversations?
Yes. In ChatGPT, go to Settings → Data Controls → “Delete all” or navigate to the chat list and delete individual conversations. Deleted chats are removed from active systems within 30 days, though OpenAI may retain them in backups for up to 90 days.
What about Apple Intelligence? Is that safer for my parents?
Apple Intelligence, on-device on iPhone 15 Pro and later, processes many AI requests locally without sending data to Apple’s servers at all. For the requests that do go to Apple’s cloud, Apple uses “Private Cloud Compute” with a no-storage, no-training guarantee that is cryptographically attested. For iPhone-using parents, Apple Intelligence is currently the strongest privacy posture available for everyday AI tasks like email summaries and notification grouping.
Should my parents just stop using AI tools?
No. The realistic threat from AI tools is lower than the threat from reused passwords, unpatched software, and phishing emails. If you have done the work in the password manager guide, your parents are already protected from the bigger risks. Adding the three AI privacy settings above is a smaller, complementary upgrade. The goal is not to make them afraid of AI — it is to let them use it safely.
What to do next
After you flip the three settings above, the next highest-value 10-minute visit is setting your parents up with a privacy-first AI default — which usually means either Apple Intelligence (if they are on a recent iPhone), Claude (if they are on Android or a computer), or a locally-hosted option for the more technical household. The companion guide walks through exactly which to pick and how to install it.
If you have not yet set up a password manager for your parents, do that first. It is the single biggest security upgrade available and a prerequisite for everything else — AI privacy included.