AI Safety
Teen AI privacy
4 min read · Updated
AI chat feels private and is not. What a teenager types is a message sent to a company, stored on its systems, and in some cases used to improve models. Most services let you turn off memory and training, and those two settings are the ones worth changing.
The privacy problem with AI chat is not that these companies are sinister. It is that the interface feels like a diary and behaves like a form submission, and almost nobody thinks about the difference while typing.
What actually happens to what they type
In broad terms, and this varies by provider:
- The message is sent to the provider and processed on their systems, not on the device.
- It is generally stored as conversation history attached to the account.
- It may be used to improve models, depending on the service and the account settings.
- Some systems keep a memory of details across conversations so later answers can use them.
- Staff may review some conversations for safety or quality purposes.
None of that is unusual for an internet service. What makes it worth attention here is the sheer specificity of what people type into these tools compared with, say, a search box.
What teenagers actually put in
More than they mean to, and usually without a decision to disclose anything:
- Full names, school names and year groups, usually as context for a question.
- Photographs of homework, which often carry a name and a school in the header.
- Screenshots of group chats, containing other people’s messages.
- Details of a friend’s situation, in order to ask for advice about it.
- Health, family and relationship details, when asking something personal.
- Occasionally, things that belong nowhere online at all: addresses, passwords, card details.
The pattern is that the personal information arrives as context rather than as the subject. Nobody sets out to disclose their address; they set out to ask whether a delivery will arrive.
The settings that actually matter
Provision varies, but two settings are common and worth changing on a teenager’s account:
Memory
Some services retain details across conversations so later answers can use them. Convenient, and it also means a detail mentioned once persists. ChatGPT allows a linked parent to turn memory off.
Training on conversations
Many services use conversations to improve models unless you opt out. ChatGPT allows a linked parent to opt the teen account out. Other providers put the equivalent somewhere in data or privacy settings.
It is also worth knowing that history can usually be deleted, and that deleting a conversation is not the same as it never having been processed.
How to explain it so it lands
Abstract privacy warnings do not work on teenagers, and it is not because they do not care. It is because the consequence is invisible and far away, which makes it unreal.
Concrete framings do better:
Things that tend to actually change behaviour:
- It is a message to a company, not a thought. Would you put it in an email to a stranger there?
- Would you be fine if a screenshot of this went round your year group?
- If you are asking about a friend, is that story yours to type in?
- Would you write this on a form at a shop?
The friend one is the most useful, because it reframes privacy as something you owe other people rather than a rule imposed on you.
A household baseline
- Turn off memory and opt out of training on the accounts your teenager uses.
- Agree a short list of things that never go in: full name, address, school, passwords, financial details, photographs with identifying information.
- Extend it to other people: no screenshots of someone else’s messages, no details of a friend’s situation.
- Check history settings and how to delete a conversation, together, once.
- Revisit when they start using a new service, because the settings will be different.
TeenCue’s own privacy model
A guide about privacy should be straight about the product it sits on, so here is exactly how TeenCue works.
When a prompt is submitted on a supported AI service, the TeenCue browser extension sends the text to be classified. The classification produces a category, a short topic label and a safety signal. That result is stored. The prompt text itself is processed in transit and is not kept as a transcript archive.
What a parent sees is the classification: which service, what kind of topic, when, and whether anything looks worth a conversation. What a parent does not get is a conversation view, because there is no stored conversation to show.
What TeenCue does not do is worth stating as plainly:
- It does not store conversation transcripts.
- It does not record general browsing history. It covers supported AI chat services only.
- It does not block or filter anything.
- It is not designed to be hidden. The extension is visible and can be removed.
The full detail, including what is retained and how to request deletion, is in the privacy policy.
Common questions
Are AI chat conversations private?
Not in the way the interface implies. Messages are sent to the provider, generally stored with the account, and in some cases used to improve models unless you opt out.
Can I delete my teenager’s AI chat history?
Most services allow conversation history to be deleted from account settings. Deleting it afterwards is not the same as it never having been sent, which is why the settings are worth changing up front.
Does TeenCue store what my teenager types?
No. Prompt text is processed to produce a category and a safety signal, and is not kept as a transcript archive. What is stored is the classification, not the conversation.