# What Data Does ChatGPT Collect About My Child? The Two Settings That Change It

**Author:** Dan Fitzpatrick
**Canonical URL:** https://parents.theaieducator.io/posts/what-data-does-chatgpt-collect-about-my-child
**Published:** 2026-09-28T07:32:49.000Z
**Updated:** 2026-09-28T07:32:51.088Z
**Category:** Safety and Wellbeing

## TL;DR

On a personal account ChatGPT collects what your child types and uploads, the date of birth and contact details on the account, their device, IP address and approximate location, and a running memory summary of them. Whether those chats also train the model comes down to one setting, "Improve the model for everyone" (OpenAI's data controls page, checked September 28, 2026), and switching it off is not retroactive: OpenAI says to "delete a chat or your account separately when you want to remove content." Do this first: open Settings, then Personalization, then Memory, and read the summary of what the tool thinks it knows about your child. The caveat that changes everything is the account. On a school-managed ChatGPT account OpenAI does not train on the content, but an administrator may be able to access, export, audit, retain and delete prompts, uploaded files, outputs and conversation history, and your home settings do not reach it. If what worries you is what your child is telling the machine rather than where the data goes, that needs a person: your doctor or the school's safeguarding lead.

## Key takeaways

- Open Settings, then Personalization, then Memory tonight and read the summary of what ChatGPT believes it knows about your child; you can correct or delete entries.
- One setting decides training on a personal account: "Improve the model for everyone", in Settings under Data controls. Turning it off stops future chats being used, but does not delete anything already stored.
- A temporary chat is not a vanishing chat. OpenAI says temporary chats are not used to improve its models but may be retained for up to 30 days for safety purposes.
- The trade reverses on a school-managed account: OpenAI does not train on ChatGPT Edu or Enterprise content, but an administrator may be able to read, export and retain prompts, files, outputs and conversation history.
- Ask the school four questions: which AI account your child uses, who can access those conversations and on what grounds, whether training is off, and how long they are kept.

## Article

On a personal account, ChatGPT collects what your child types and uploads, the age and contact details on the account, their device, IP address and approximate location, and whatever its memory decides is worth keeping. Whether those chats also train the model depends on one setting. The bigger caveat: if your child uses a school-managed account, the trade reverses, and your home settings do not reach it.

Start there, because it changes what you are actually deciding. The question is not how much data ChatGPT collects. It collects a lot, as every large service does, including [the My AI chatbot already sitting inside Snapchat](https://parents.theaieducator.io/posts/is-snapchat-my-ai-safe-for-kids). The question is who gets to read it and who gets to learn from it, and on a child's account those two answers come from different places.

## What data does ChatGPT collect?

[OpenAI's privacy policy](https://openai.com/policies/row-privacy-policy/), last updated on February 6, 2026, lists the categories in its own words. Account information: "name, contact information, account credentials, date of birth, payment information, and transaction history." User content: the prompts, files, images, audio and video your child submits. Log data: "Internet Protocol address, browser type and settings, the date and time of your request." Usage data covering what was viewed and which features were used. Device information. And location, which is the general area worked out from the IP address, plus precise location only where someone has switched that on.

Read that list again with a fourteen-year-old in mind. The date of birth on the account. The homework they pasted in. The photo of a worksheet. The question about a rash. The 11pm timestamp.

[The Common Sense Media census of AI use by tweens and teens](https://www.commonsensemedia.org/sites/default/files/research/report/2026-ai-use-by-tweens-and-teens-1.pdf), published in March 2026 from a survey of 1,204 US children aged 9 to 17, found that 57 percent of children who use AI had used it to get information or advice about their health or body, and 37 percent to discuss their feelings or personal problems. The same survey found 71 percent correctly recognized that AI chatbots might use what they tell them to train or improve AI, but only 24 percent were certain that was definitely true. My reading of those two numbers together: children are putting genuinely personal things into these tools while being only vaguely aware of where it goes. That gap is the thing to close, and a conversation closes it faster than a setting does.

## Do my child's chats train the model?

On a personal account, they can, and there is one setting that decides it. [OpenAI's data controls page](https://help.openai.com/en/articles/7730893-data-controls-faq), checked on September 28, 2026, calls the setting "Improve the model for everyone," and says that with it off, "your new conversations won't be used to train OpenAI models."

Two things qualify that, and both matter. The first is that switching it off is not retroactive. OpenAI's own answer to whether turning off training removes what is already there is blunt: "Delete a chat or your account separately when you want to remove content." The second is that temporary chats are treated differently again. OpenAI says they "do not appear in your chat history," "are not used to improve OpenAI models," and "may be retained for up to 30 days for safety purposes." So a temporary chat is not a vanishing chat. It is a chat that leaves no trace where you would look, which is a different thing, and a point worth holding on to if you were planning to check the history.

If your teenager's account is linked to yours, that training setting is one of the ten you can manage. [OpenAI's parental controls page](https://help.openai.com/en/articles/12315553), checked the same day, lists "Improve the model for everyone" among them and describes it as controlling "whether your teen's conversations can be used to help improve ChatGPT models." The same page is equally clear about what linking does not give you: "Parental controls do not provide access to a teen's conversations, chat history, or real-time activity." That is the same wall parents hit when they go looking for [the chat history itself](https://parents.theaieducator.io/posts/can-parents-see-chatgpt-history). If your teen unlinks, you are notified, and your control ends there.

## What does ChatGPT's memory keep?

Memory is the part parents tend to miss, because it is not a log of chats. It is a running summary of your child. [OpenAI's memory page](https://help.openai.com/en/articles/8590148-memory-faq), checked on September 28, 2026, says it retains "relevant preferences and details from your chats and other available sources," and that "Memory does not retain every detail from every conversation. ChatGPT decides which available information is relevant to a response."

So a child who mentioned once that they are struggling in math, or that their parents argue, or which school they go to, may find the tool carrying that forward into conversations weeks later. You can look at this. In Settings, under Personalization, then Memory, there is a summary of what the tool believes it knows, and you can correct an entry, tell it not to mention something again, or delete entries outright.

I would open that page before I changed any setting. It is the single most informative two minutes available to you, and on a teenager's linked account a parent can manage supported memory controls through parental controls.

## Does the school's account change the answer?

Completely, and in the direction most parents do not expect. It is a different question from [whether the school can read the chats](https://parents.theaieducator.io/posts/can-school-see-chatgpt-history), and the answers pull against each other.

OpenAI states that it "does not use content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces to train its models." On the school's account, then, the training question largely goes away. But its [page on data access for managed accounts](https://help.openai.com/en/articles/20001067-data-access-for-your-managed-chatgpt-account), checked on September 28, 2026, says an administrator "may be able to access, export, audit, retain, delete" three things: "Content you submit to ChatGPT (prompts, files you upload, and outputs you generate)," "Conversation history and shared workspace content," and "Usage and activity metadata (for example: sign-in activity, device/session info, feature usage)." OpenAI adds that this applies where it is "enabled by your organization's configuration and applicable law," which means the school's own choices decide how much of it is switched on.

There is a smaller detail with a sharp edge. The new Privacy Center, the place OpenAI is gathering all of these controls into, is described on [its help page](https://help.openai.com/en/articles/20001488-privacy-center-in-chatgpt) as rolling out to "signed-in ChatGPT Free, Go, Plus, Pro, and Business users on the web and in the ChatGPT apps for iOS and Android," and specifically not to "ChatGPT Enterprise, Edu, or ChatGPT for Healthcare." A child on the school's account may not have the tidy control panel at all. Availability, OpenAI says, depends on "your plan, region, account, and workspace settings."

> **The Read and Train Test.** Ask two questions of any AI app your child uses: who can read what my child types, and who can learn from it. The account your child is signed into decides both answers, and it usually decides them in opposite directions.

## What I See in Schools

Before this work I was Director of Digital Strategy in further education, and these days most of what I do is AI strategy with schools and trusts. What goes wrong here is almost never carelessness with children's data. It is a school that switched a tool on without anyone deciding who holds the admin console, and what they are allowed to look at with it.

The technical answer is usually fine. Somebody has read the data processing agreement. What is missing is the human rule: whether a member of staff may read a child's conversation because they are curious rather than concerned, who signs that off, and what happens to the record afterward. In the schools where this is handled well, one named person holds that access, there is a written reason required before anyone uses it, and the safeguarding lead knows the rule. In the schools where it is not, the access exists, nobody has agreed the rule, and nobody has looked.

That is a fair question for a parent to ask, and it is not an aggressive one. Most schools will be glad somebody asked.

## What to change tonight, in ten minutes

This is a ten-minute job and it does not require your child's cooperation for the first part.

**One. Look at memory before you touch anything else.** Settings, then Personalization, then Memory. Read the summary. If something in there surprises you, that is your conversation for this week, and it is a better opening than any rule you could impose.

**Two. Decide the training setting deliberately.** In Settings, under Data controls, find "Improve the model for everyone" and choose it rather than inherit it. On a linked teen account you can set this yourself, along with [the other nine parental controls](https://parents.theaieducator.io/posts/chatgpt-parental-controls). On your child's own account, this is a decision to make with them, because it is theirs to reverse in one tap.

Then, if the option is available on your plan and in your region, open the Privacy Center, which OpenAI describes as "a place in ChatGPT to learn about your privacy options and find the settings that manage them." It gathers memory, ad personalization, location sharing, temporary chats, connected apps, model improvement and data export into one screen, which beats hunting through menus.

What none of this does is clear out what is already stored. For that you delete specific chats, delete memories, or export and then delete the account, which are separate actions with separate consequences.

## How this differs by age

**Under 8.** They should not have an account. OpenAI's privacy policy says the services are "not directed to, or intended for, children under 13" and that it does "not knowingly collect Personal Data from children under 13." If a younger child is using AI, it is on your account, with you there, and the data is yours rather than theirs. [The stated minimum ages across the main tools](https://parents.theaieducator.io/posts/can-kids-use-chatgpt) are worth knowing before you decide.

**8 to 12.** Still your account, and the useful habit starts here: nothing goes in that you would not put on a postcard. Full name, school, address, photographs of themselves. Say it as a rule about postcards rather than a rule about data and it will stick.

**13 to 15.** This is where the teen experience applies. OpenAI says [ChatGPT for Teens](https://openai.com/index/chatgpt-for-teens/), announced on August 18, 2026, is applied where "our system estimates someone is under 18 or they state their age is between 13 and 17." Link the account if your child will agree to it, set the training preference together, and check the memory summary with them rather than behind them. OpenAI has been explicit about the trade it is making here, writing in [September 2025](https://openai.com/index/teen-safety-freedom-and-privacy/) that "we prioritize safety ahead of privacy and freedom for teens."

**16 to 18.** Persuasion is the only tool left, and the privacy argument works better here than the safety one. The objection that lands with a seventeen-year-old is being profiled, not being protected. Show them the ad personalization and memory settings and let them decide.

## What to say to your child

> **You could say (8 to 12):** "Anything you type into that goes to a company, not into a diary. So no full name, no school, no address. Same rule as a postcard."

> **You could say (13 to 15):** "There is a setting that decides whether your chats get used to train the thing. Shall we look at it together and pick, rather than just leaving it?"

> **You could say (16 to 18):** "Have you seen what it thinks it knows about you? It keeps a summary. Worth a look, if only to delete the bits that are wrong."

Not one of those is an accusation, and that matters more than the wording. A child who thinks they are being investigated will move to a logged-out browser, and then you have lost both the setting and the conversation.

## What to ask the school

Four questions, and any of them can be asked by email without a meeting.

1. Is my child using a school-managed AI account, and if so on which product?
2. Who in school can access conversations on that account, and what reason do they need to give?
3. Does the school's configuration keep those conversations out of model training, and can you confirm that in writing?
4. How long are they retained, and what happens to them when my child leaves?

If the answer to the first is no, and your child is using AI for schoolwork on a personal account, then the settings above are entirely in your hands, which is simpler. In a UK secondary school the person who will know the answer is usually the data protection officer rather than the head of year.

## What none of this can guarantee

No setting stops a determined child from opening a logged-out browser or a second account, and none of these controls reaches a friend's phone. Turning training off does not remove what is already stored. A teen account can be unlinked. Age limits on these products are self-declared and weakly enforced, which is why OpenAI is building age prediction at all, writing that "if there is doubt, we'll play it safe and default to the under-18 experience."

The law is moving, and it is moving in your favor, slowly. [The US COPPA rule](https://www.ftc.gov/business-guidance/resources/complying-coppa-frequently-asked-questions), which covers under-13s and was amended on April 22, 2025, requires operators to "retain personal information collected online from a child for only as long as is necessary to fulfill the purpose for which it was collected." In the UK, the Information Commissioner's Office said in its [Children's Code progress update](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/childrens-information/childrens-code-guidance-and-resources/children-s-code-strategy-progress-update-august-2026/next-steps) of August 18, 2026 that it continues "to scan the horizon for risks to children's data protection rights in areas such as the wider mobile games sector, AI chatbots and health apps," and that it will "also consider children as part of our code on AI and automated decision-making." Neither of those protects your child this evening. The memory summary and the training setting do, a little.

And if what is actually worrying you is not the data but what your child is telling the machine at midnight, that is a different problem with a better answer, and it starts with a person rather than a setting: your doctor, your child's school safeguarding lead, or whoever your child likes talking to most.

If your child's school has not yet told parents which AI accounts children are using and who can read them, that is a reasonable thing to ask for, and it is the work I do with schools: [finding out how ready a school actually is](https://theaieducator.io/school-ai-readiness?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=what-data-does-chatgpt-collect-about-my-child) before the tools arrive rather than after. I also write [a Sunday newsletter](https://theaieducator.io/?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=what-data-does-chatgpt-collect-about-my-child#newsletter) that keeps teachers and parents up to date on changes like this one without the panic. Parents are very welcome.

*Dan Fitzpatrick is a former secondary teacher, assistant headteacher and Director of Digital Strategy in further education who now helps schools and families make sense of AI. [More about Dan](https://theaieducator.io/about?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=what-data-does-chatgpt-collect-about-my-child).*

## Frequently Asked Questions

### Can I ask OpenAI to delete everything it holds about my child?

You can delete the account, which is the most complete route available to you, and OpenAI's privacy policy (last updated February 6, 2026) says it removes personal data on request "within 30 days unless we need to retain it for longer." Deleting individual chats and individual memories are separate actions with separate effects, so doing one does not do the other. Deleting the account also removes your child's chat history and any purchase, so export the data first if there is schoolwork in it worth keeping.

### Does ChatGPT show my child advertising based on their chats?

Ad personalization is one of the controls OpenAI lists in the new Privacy Center, alongside memory and location sharing, and its help page describes those ad controls as applying to Free and Go users. The honest position is that this is a moving area and the setting matters more than the current policy: find ad personalization in the Privacy Center, or in Settings if the Privacy Center has not reached your plan or region yet, and turn it off deliberately rather than leaving it as you found it.

### Is Gemini or Copilot any different from ChatGPT on data?

Apply the same two questions rather than assuming, because each product answers them differently and the answers change every few months: who can read what my child types, and who can learn from it. Every major assistant publishes a data controls or privacy page of its own, and that page, not a news article about it, is the thing to read. The pattern that does hold across all of them is that a managed school or work account behaves differently from a personal one.

### My child has been using ChatGPT for months before we changed anything. Is it too late?

No, but the two jobs are different. Changing the training setting only affects conversations from that point on, so the second job is clearing what is already there: delete the specific chats you would rather were gone, then open the memory summary and delete the entries that hold anything personal. Do that part with your child if they are old enough to have an account of their own, because they can undo all of it in one tap and are much less likely to if they helped.

### What should I ask a UK secondary school about my child's AI account?

Ask the data protection officer rather than the head of year, because this is a data question and they will have the paperwork. Ask which AI products children sign into with a school account, who in school can access those conversations and what reason they must record, whether the configuration keeps conversations out of model training, how long they are retained, and what happens when your child leaves. The Information Commissioner's Office said in its Children's Code progress update of August 18, 2026 that it is watching AI chatbots as a risk area for children's data, so a school being asked this in 2026 should not be surprised by it.