You can switch off most of the AI your child chose: Apple's writing and image tools, Gemini on an Android account you supervise, Snapchat's My AI, Meta's AI characters. You cannot switch off the AI that was built into the things they already use. Google says plainly that its AI answers in Search "cannot be turned off," and nothing you set at home reaches a school device.
That second half is the part worth ten minutes tonight, because it is where most of the AI in a child's day now lives.
Start with one number. Pew Research Center surveyed 1,458 US teens aged 13 to 17 and their parents and published the results on February 24, 2026. Fifty-one percent of the parents said their teen uses chatbots. Sixty-four percent of the teens said they do. Those are two different populations answering about the same households, with a margin of error of plus or minus 3.3 points, and the gap between them is the size of the thing you cannot see from the kitchen.
The four switches that do exist, and where they are
Four real controls, roughly in the order most families need them. Each one takes a couple of minutes, and each one has a limit I will name as I go.
On an iPhone or iPad
Apple puts its AI controls inside Screen Time rather than in a menu called AI, which is why parents hunt for them and give up. On your child's device, or on yours if their device is in your Family Sharing group, the restrictions live under Screen Time's content and privacy section, on the screen Apple calls Intelligence and Siri.
Apple's own description of that screen, written for the Mac version, is the clearest statement of what you are turning off: it allows or blocks "access to Apple Intelligence features, including Image Creation, Writing Tools and Intelligence Extensions," and separately allows or blocks Siri content including "Web Search Content in Siri."
Apple's parental controls guide, published on September 14, 2026, puts the same thing in family terms: "You can limit what your child can access through Siri and Siri controls," choosing to allow or block options like "Writing Assistance or Image Creation."
The caveat is on Apple's own page, and it explains a lot of frustrated evenings: "Not all parental controls are available on all Apple devices. Some parental controls are located within Screen Time, and some are enabled through the settings directly." You may have to look in two places.
On an Android phone
Gemini is not in the app list with everything else. It has its own switch, in its own menu, and that is the step most parents stop one short of.
Google's instructions for families, read on October 5, 2026, are four taps: "Go to familylink.google.com or open the Family Link app. Select your child. Tap Controls and then Gemini and then Gemini Apps. Turn Gemini Apps on or off."
Two things on that page matter more than the taps. Google states that "Gemini Apps are not available for supervised accounts in the European Economic Area, Switzerland, and the United Kingdom," so a British parent looking for this switch will not find it, because the thing it switches is not offered to supervised accounts there in the first place. And for younger children, Google notes that "'Hey Google' and Voice Match are not available to use with Gemini Apps for users under 13 (or the applicable age in your country)."
If you want the fuller picture of what Google's assistant does and does not do for a child, I have written it up separately in whether Gemini is safe for kids.
Inside Snapchat
My AI is the one your child did not install and cannot easily get rid of, and it sits at the top of their chat list looking like a friend.
A parent linked through Family Center can switch it off. Snapchat's support page for parents gives three steps: open Family Center, scroll to Settings at the bottom of the page, then "Turn on the toggle to Disable My AI." Snapchat describes the effect precisely: "When this setting is on, My AI will not store or process your teen's queries, nor reply to your teen's messages."
Your child can also clear the conversation themselves, and it is worth knowing what that does not do. Snapchat's page on removing My AI says: "Clearing removes the conversation thread from your Chat feed. It does not delete data you've previously shared with My AI." Tidying the screen is not the same as taking anything back, and that distinction is worth saying out loud to a teenager. There is more on this one in whether Snapchat's My AI is safe for kids.
Inside Instagram
You can turn off your teen's chats with Meta's AI characters. You cannot turn off Meta's AI assistant, and Meta says so in the same breath.
When Meta announced its teen AI controls on October 17, 2025, it said parents would be able to "turn off their teen's access to one-on-one chats with AI characters entirely," and to "block specific AI characters if they don't want to turn off access to AI characters altogether." The controls were to arrive during 2026, beginning with Instagram, and "first in English, to the US, UK, Canada and Australia."
In the same announcement Meta said what would not be switchable: "Meta's AI assistant will remain available to offer helpful information and educational opportunities, with default, age-appropriate protections in place." The characters are optional. The assistant is not. That was the announcement rather than a guarantee about tonight, so check your own supervision menu for what has actually landed, because Meta has been rolling these out country by country.
What you get instead of a switch is a view. On April 23, 2026 Meta added an insights tab for parents supervising Teen Accounts, where they "will be able to see the topics their teen has been asking Meta AI about in that specific app over the past week," available "in the US, UK, Australia, Canada, and Brazil." Topics, not conversations. Meta is clear that it shows the subject rather than the exchange, including for the hardest subjects: "We'll still show parents the topic their teen was asking about, even if Meta AI didn't respond to the question."
What has no off switch at all
The AI your child meets most often is the one at the top of every search result, and Google's own help page says you cannot turn it off.
On its help page explaining AI Overviews in Search, read on October 5, 2026, Google writes: "Features cannot be turned off. However, you can select the Web filter after you perform a search." The same page describes what that filter does: "This filter displays only text-based links without features like AI Overviews."
So there is a way to get the old list of blue links. It is a tab your child taps after they have already searched and already seen the answer, on that search only, and then it is gone again on the next one. It is a per-search workaround, not a setting, and no parental control makes it stick.
AI Mode, the fuller conversational version inside Search, is a different case. Google's Search Labs page for AI Mode lists the requirements to join the experiment, and the first is to "Be 18 or older." The same page notes that "Search Labs and its experiments are not available to Google Workspace accounts, including Google Workspace for Education accounts," which is why your child may meet it on their own phone and not on the school's.
That one unswitchable thing outweighs the four switches above, because homework starts in a search box. Ofcom's research into UK children's media lives, published on May 21, 2026, found AI is used by over half of 8 to 17 year olds, 56 percent, rising to two thirds of 16 to 17 year olds, and that one in ten, 11 percent, have used AI "as someone to talk to or 'as a friend.'" My reading of those numbers, rather than Ofcom's, is that the companionship figure gets the headlines and the first figure is the one that changes your evening: for most children this is not a confidante, it is the thing that answers the question before they have finished typing it.
And children are not well placed to check what it tells them. Internet Matters surveyed 2,000 parents and 1,000 UK children aged 9 to 17 and reported on September 10, 2026 that nearly two thirds of those children, 64 percent, came across false or misleading content online in a six month period, and that almost a quarter, 23 percent, are not confident about recognizing misinformation.
The Switch Test
Before you go hunting for a setting, ask one question about the AI that is worrying you: did it arrive, or was it already there?
The Switch Test. If it arrived as an app, a character or a chat your child added, there is a switch and you can find it. If it is part of the search box, the keyboard, the camera or the assistant, there is no switch, and the job is a conversation rather than a setting.
Five seconds of that saves an hour in the menus, and it tells you which half of the evening you are in. Half of this job is settings. The other half is not, and no amount of looking will turn it into settings.
The AI your child chose has an off switch. The AI that was already there does not, and that is now most of what they meet.
If the thing you actually want to block is a named app rather than a built-in feature, that is a different and simpler job, written out step by step for an iPhone and for an Android phone or Chromebook.
What I See in Schools
The settings conversation and the safeguarding conversation are not the same conversation, and in schools they sit in different rooms.
I spent years as a teacher and then as an assistant headteacher, and I sit on a school's board of trustees now. What I noticed in safeguarding meetings, long before any of this was about AI, is that the technical question and the welfare question never arrive together. One is a question about a device. The other is a question about a child. The pattern I see most often with parents is the first being used to answer the second: an evening spent on settings, and real relief at the end of it, about a worry that was never going to be settled by a menu.
The settings are still worth doing. They are just the small half.
The other thing I would say from inside the building is that your controls stop at your front door. A school device runs on the school's account under the school's rules, and a staff member, not you, decides which AI services are switched on for your child's year group. I have set out what that actually means in what a school can see of a child's AI use. If your child's school has never written to parents about which AI is on in the classroom and what staff have been trained to do with it, that gap is worth naming at the next parents' evening, and it is the work I do with schools through staff training on AI.
How this changes by age
Under 8. Switch nearly all of it off, and say why in one sentence. At this age the AI they meet is a voice answering a question they were about to wonder about for themselves, and the wondering is the valuable part. You could say: "Let's have a guess first. Then we can check."
8 to 12. This is the age to do the four switches properly, because this is when a phone arrives with everything already turned on and nobody mentions it. Do it with them in the room rather than while they are at school. You could say: "I'm turning two of these off tonight. Tell me if one of them was something you actually used, and I'll think again."
13 to 15. Switching things off starts to work against you here, because a second account on a different email address undoes all of it in four minutes. Aim for the Snapchat and Instagram controls, which are tied to supervision rather than to the device, and spend the rest of your effort on the search box. You could say: "When Google answers at the top, who wrote that? Where would you check it?"
16 to 18. Controls are mostly the wrong tool now, and the useful habit is the one they will take into exams, applications and a first job: knowing what the machine is confident about and wrong about. You could say: "What's the last thing it told you that turned out to be wrong?"
If the hard part is finding the words rather than the menus, I have written a separate guide on how to explain AI to a child.
What none of these settings can guarantee
Being straight about the limits is the only way any of this is useful.
None of them cover a friend's phone, a shared tablet or a games console. None of them cover a second account made on a different email address. None of them reach your child's school device or school account, where an administrator decides what is available. The regional notes matter, as Google's exclusion of supervised accounts in the UK, the European Economic Area and Switzerland shows, so a control that exists in one country may simply not be there in yours. And every parental control on every platform can be worked around by a determined teenager with time and a search engine, which is a reason not to rely on them alone rather than a reason to skip them.
Two things take precedence over anything here. Your child's school sets the rules for its own devices and its own work, and where those conflict with your home rule, the school's rule applies at school. And if what is worrying you is not the setting but your child, if something they have said or written has frightened you, that belongs with your pediatrician or GP, or your school's safeguarding lead, tonight rather than at the weekend.
What I would do tonight
Do the four switches that apply to your house, which will take about fifteen minutes in total. Then sit with the one you cannot switch off and ask your child to search for something in front of you, and read the answer at the top together. That second part is the one that keeps working after they turn 18, and it is the whole of what I mean by letting AI help with the doing and keeping the thinking with your child.
If it is useful to have the next of these as it is written, I send one email a week to parents on exactly this, and you can sign up here.
Dan Fitzpatrick is a former teacher and assistant headteacher who now works with schools on AI. More about Dan.
