ChatGPT's parental controls are worth setting and are not enough on their own. As of October 16, 2026 a linked parent can set quiet hours and study hours, keep chats out of training, and switch off voice, memory and image generation. Quiet hours start off and training starts on. The controls never show a conversation, and a logged-out browser sidesteps them. The minimum age is 13. Tonight: set quiet hours.
Last updated October 16, 2026. I re-read OpenAI's parental controls and age prediction help pages today. New in this version: which settings start on and which start off, the limit on the disordered-eating alert, the current menu for an adult wrongly placed in the teen experience (it is now called Age), what happens to the link when a teen turns 18, and what OpenAI says its age-check partner does with an ID. I also corrected a date I had taken from page metadata rather than from the page.
I use these tools every day, I have advised the UK Department for Education and KHDA in Dubai on AI in education, and I would still not describe any of this as "control." Think of it as a seatbelt. It is the right thing to fit, it takes about ten minutes, and it does not make the road safe. Here is how to fit it, and what it will not stop.
Are ChatGPT's parental controls enough on their own?
No, and it is worth understanding why before you spend the ten minutes. Everything on that settings page governs one account in one app: your child's account, while they are signed in to it. That is a real thing to govern. It is also about a third of the problem.
The other two thirds are these. The same chatbot answers without an account, in any browser, on any device. And most of the AI a child meets in an ordinary day is not ChatGPT at all. It is the assistant inside Snapchat, the summary sitting above their Google results, the chatbot built into a game.
So the honest answer is: set the controls, because a child who is signed in most of the time is covered most of the time. Then set the two layers the app cannot reach, which are the device and the conversation. All three are below.
By age, in one line each. Under 13, no account, by OpenAI's own rule. Thirteen to 15, link the accounts and set quiet hours and study hours on day one. Sixteen to 18, link if they will let you, and expect to be unlinked eventually. The detail for each band is further down.
If you do one thing tonight, set quiet hours from bedtime to breakfast. It takes two minutes, it is the setting families actually keep, and late-night use is the use that worries me most.
What do ChatGPT's parental controls actually do?
OpenAI introduced parental controls on September 29, 2025 and expanded them on July 13, 2026, and its help center lists the settings a parent with a linked account can manage. Checked against that page on October 16, 2026, they are: reduce sensitive content, stop the teen's chats being used to improve the model, switch off saved memories, switch off voice mode, switch off image generation, switch off the cloud browser and Codex network access, turn Study Mode on, set study hours in which new chats start in Study Mode, and set quiet hours in which the teen cannot use ChatGPT at all.
Then, on August 18, 2026, OpenAI launched ChatGPT for Teens. Any account the company believes belongs to someone under 18, whether from the date of birth given or from its age prediction system, is moved into it automatically, whether or not a parent is linked. It brings stricter content filters, break reminders, homework reminders that steer a request back toward working it through, and rules against romantic language, emotional dependence and role-play. If you want the product itself rather than the controls, what ChatGPT for Teens does and does not do has its own page.
Which settings are already on, and which do you have to switch on?
This is the part most guides skip, and it changes what you do tonight. OpenAI's help center, read on October 16, 2026, gives a default for each control once the accounts are linked:
| Control | Starting setting, per OpenAI's page |
|---|---|
| Reduce sensitive content | Enabled |
| Improve the model for everyone | Enabled |
| Reference saved memories | Enabled where available; may be disabled by default in some regions |
| Voice Mode | Enabled |
| Image generation | Enabled |
| Cloud browser | Enabled |
| Codex network access | Enabled |
| Study mode | Disabled |
| Study hours | Disabled, no hours scheduled |
| Quiet hours | Disabled, no hours scheduled |
My reading of the table, and it is mine rather than OpenAI's: linking the accounts changes less than most parents assume. The two protections people most often want, a night-time cutoff and a nudge toward Study Mode, do nothing until you set them. And a teen's chats can be used to improve the model unless you switch that off. Linking is step one, not the finish. The same page says safety notifications are on by default once accounts are linked, so that part you do not have to set.
One thing the page cannot tell you is what your own child's account shows, so open the Parental controls page under your child's name and read each line instead of trusting the defaults above. What happens to the chats themselves is a separate question, covered in what ChatGPT keeps about your child and the two settings that change it.
What is age prediction, and what if an adult is put in the teen experience?
Age prediction deserves its own paragraph, because parents are now meeting it from both sides. OpenAI's help page on age prediction says the system looks at signals linked to the account, including the general topics discussed and the times of day ChatGPT is used, plus how long the account has existed, and that it can apply even if a date of birth was entered at signup. The page admits that sometimes ChatGPT may get it wrong.
An adult wrongly placed in the teen experience can check the Age setting under Settings, then Account, and verify there if the option appears. The check is handled by a third party called Persona and, depending on the country, may ask for a live selfie or a government ID. OpenAI says Persona deletes uploaded IDs and selfies within 7 days and that OpenAI receives only the date of birth or the age estimate, not the ID or the selfie. That is OpenAI describing its own arrangement, so treat it as a statement by the company and read the page yourself before handing over a document.
My reading of the whole system: it is a sensible net, and it also means a determined seventeen year old holding an adult's ID has a documented route out. Treat age prediction as a net with holes in it rather than a lock.
The minimum age has not moved. OpenAI's terms of use, effective January 1, 2026, say "You must be at least 13 years old or the minimum age required in your country to consent to use the Services" and "If you are under 18 you must have your parent or legal guardian's permission to use the Services." A child under 13 should not have a ChatGPT account at all, and the age rules across the main chatbots vary more than most parents expect.
How do I set up ChatGPT parental controls?
Both of you need a ChatGPT account. Either of you can start the link. In your account go to Settings, then Parental controls, then Add family member, choose phone or email, and send the invitation. Your teenager receives a text or email with a link, and has to accept it. A teen can also send the invitation from their side (Settings, then Parental controls, then Invite a parent), which is a good sign if they do it unprompted. Once linked, the settings above appear under your child's name, and every change takes effect on their account.
Two settings are worth making on the first day: quiet hours, and I would start with bedtime to breakfast, and study hours across the homework window, so that when they open a new chat to "help" with an essay it asks them questions instead of writing it. Switch off "Improve the model for everyone" at the same sitting if you do not want your teen's chats used that way. The Internet Matters guide walks through the same steps; I read it on October 16, 2026, and the page itself shows no date.
Can I see my child's ChatGPT conversations?
No. OpenAI's help center says plainly that linking accounts "does not give a parent or guardian access to a teen's conversations, chat history, or real-time activity." It adds that parental controls do not let parents or guardians read or monitor a teen's conversations. What you get instead is a safety notification, by email, text and app alert, in a short list of situations: when trained reviewers identify a serious self-harm concern, when they identify behavior that may be associated with disordered eating, and when the teen's account is deactivated for conversations about acts of violence. The notification carries a brief description of the type of concern, not the conversation. OpenAI states that "safety notifications are not real-time monitoring," and its original announcement says the company "might sometimes raise an alarm when there isn't real danger."
Two limits worth knowing, both from OpenAI's pages. The help page says the disordered-eating notifications apply only to conversations in Chat, and are not available in Work or Codex. And the ChatGPT for Teens page says more eating-disorder notifications are being added, with limits on what is shared. So the list will grow, and it will still not be a transcript.
I think that is the right design, and I know it is not what many parents want to hear. A sixteen year old who knows every word is read will use a different app. A sixteen year old who knows you will hear if something serious happens, and nothing else, may keep using the one with the seatbelt on.
Your child may well have looked this up from the other side, and the answer is not the same everywhere they use AI: on a school device, what the school can see is not what you can see.
What can the controls not do?
They cannot see a logged-out browser. Internet Matters puts it in one line: "children can use the chatbot without logging in, which will allow them to avoid any controls you have set." ChatGPT answers without an account, on any browser, on any device.
They cannot survive an unlink. Either side can unlink at any time, you are notified when a teen does it, and after that you can no longer manage settings or receive safety notifications for that account.
They end on their own. When a teen turns 18, OpenAI's page says the parental-controls connection ends, teen protections are turned off, and the account holder gets full control of their settings. Plan for that birthday instead of being surprised by it.
They cannot reach a second account registered with a different date of birth, although age prediction now catches some of those.
And they do nothing about the AI your child already uses elsewhere: Snapchat's My AI, the AI summaries at the top of a Google search, the chatbot inside a game. That is the real limit of everything on this page. It governs one app, and your child can open a different chatbot in about ten seconds.
Google's arrangement is different and worth knowing. A parent can switch Gemini on for a child under 13 through a supervised Family Link account and gets an email the first time the child uses it, but Google's own support page states that "Gemini Apps are not available for supervised accounts in the European Economic Area, Switzerland, and the United Kingdom." So if you are in Britain and your ten year old says they use Gemini, it is not on a supervised account, and that is worth asking about tonight.
The device layer, and what it can actually reach
This is the layer most parents skip, and it is the one that covers the logged-out browser.
On an iPhone, Screen Time lives in Settings, then Screen Time. Downtime is the setting that matters here: Apple's own description is that it blocks apps and notifications for the periods you schedule, leaving only phone calls and the apps you choose to allow. Schedule it across the night and the browser goes with everything else. App Limits will also cap a category such as social networking.
Screen Time will also block one named app outright, which is what most parents want first when a particular chatbot has become the problem. It takes about two minutes, and Apple's own guidance is that the app then "appears dimmed on the Home Screen, but not deleted" (Apple Support, published September 14, 2026), so your child will see it and will know. The part worth reading before you do it is what the block leaves untouched: blocking an AI app on an iPhone does nothing about Siri, the AI answers sitting above a Google search, or the assistant at the top of a chat list in an app your child already had.
On Android, the equivalent sits in Family Link, where a parent sets a daily limit, a bedtime schedule and per-app rules from the parent's own phone.
Be clear with yourself about what this layer does and does not do. It controls when a device is usable, not what is said inside an app. No device setting reads a chatbot conversation, and a child with a school laptop, a games console or a friend's phone is outside it entirely. It closes the late-night gap, which is the gap worth closing first.
What Good Looks Like
The families who get this right think in three layers, and I give them that structure as a rule:
Controls work in three layers: the account, which is what the app allows; the device, which is what the phone allows; and the conversation, which is what your child tells you. Only the third layer covers the logged-out browser.
The account layer is ChatGPT's controls, done in ten minutes. The device layer is Downtime or Family Link, done in five. The conversation layer is ten minutes a week.
It is the third layer that parents underestimate, and here is what I see from inside schools that makes me say so. When a school's filtering blocks a chatbot, the block sits on the school network and the school device, and pupils work out within about a fortnight which of their own devices the block does not reach. When a safeguarding lead looks into something, what they can usually retrieve is a record of sites visited and files in the school account, not the contents of a conversation held in a personal app on a personal phone. And the question I am asked most often by parents at the end of an AI session in a school hall is almost never about a setting. It is a version of "how do I know what they are actually doing with it." The parents who have an answer to that question are not the ones with the strictest settings. They are the ones who have made it ordinary to ask.
So make it ordinary. Ask what they used it for this week and what it got wrong. The second half of that question is the one that works, because it invites your child to be the expert and to show you something, which is a far easier conversation than being audited.
The evidence says the gap is real. Pew Research Center surveyed 1,458 US parents of teens aged 13 to 17 and published the findings in February 2026: 51 percent of those parents said their teen uses AI chatbots, and about three in ten said they were not sure. In the matching survey of US teens, 64 percent said they use chatbots. Those are two different groups of people answering their own question, so the 13-point difference is a signal rather than a precise measure. That gap is not closed by a setting. It is closed by asking.
What to set, by age
Under 13. No ChatGPT account, by OpenAI's own rule. If they want to see what a chatbot is, they use yours, next to you. If you are outside the UK and Europe and want Gemini, use a supervised Family Link account and leave it switched off until you have watched them use it a few times.
13 to 15. Link the accounts on day one, with quiet hours from bedtime and study hours across homework. Leave image generation off. You could say: "This isn't me reading your chats. I can't, and I've checked. It's me making sure it asks you questions instead of doing your homework, and that it switches off at night."
16 to 18. Link the accounts if they will let you, and accept it if they unlink at seventeen, because that is what the setting is for. Keep the conversation layer going. You could say: "You'll be off to university or work soon and nobody will set this for you. Show me your settings, and tell me what you'd switch off for a younger sibling."
One caveat that sits above all of this: if you are worried about your child's mental health rather than their homework, no setting is the right tool. A safety notification is a backstop, not a diagnosis, and it is not designed to catch everything. Talk to your family doctor, your pediatrician or your child's school, and in the US you can reach the 988 Suicide and Crisis Lifeline by call or text, or in the UK Samaritans on 116 123 and Childline on 0800 1111. Where your child's school has its own rule about AI tools, or a doctor or safeguarding lead advises otherwise, theirs comes before anything here.
What to ask the school
Ask whether the school has spoken to parents about AI controls, and whether it has a recommended setup. Most have not, and it is a small thing that makes a big difference at parents' evening. If your child's school is working out where it stands on AI, this is the work I do with schools: an honest look at how ready the school is, including what parents have been told. It is worth sending to your head of year.
The settings will change again, probably before Christmas. My Sunday newsletter tracks what moved each week for more than 44,000 teachers and parents, without the panic.
Dan Fitzpatrick is a former teacher and assistant headteacher, a Forbes contributor, and the founder of The AI Educator. Read Dan's story.
