All guides

AI chatbots

The chatbots your child talks to on purpose, the assistant already built into the device, and the toys that listen. The controls here are real but thinner than on any other app in this list, so more of the work is the conversation.

ChatAbout 23 minutesSource confidence: highChecked 20 November 2026

Menu names shift between software versions and between phone makers. If a name here does not match what you see, look for something close by rather than assuming the setting is gone.

There is no younger version of AI chatbots

There is no younger version of a general-purpose chatbot. The nearest thing is Gemini on a supervised Google account, which a parent has to switch on and which is covered in its own step above.

What to watch for

Before you start

  1. 1

    Work out which ones they can actually reach

    Every other guide on this site starts with an app you decided to allow. This one usually starts with something that arrived on its own. Before you change a setting it is worth knowing how many doors there are, because a control on one of them does nothing about the rest.

    1. Look for a chatbot app they installed and signed into.
    2. Check the assistant that came with the device, which is Apple Intelligence and Siri, Gemini, Copilot or Alexa depending on what they have. Your device's guide on SafeStart has the exact path for switching it off.
    3. Check inside the apps they already use. Several social and messaging apps now have a chatbot in the chat list or behind the search bar.
    4. Check the browser. Every one of these also works as a website, so an app you block is still reachable by typing the address.

    You have a list to work from, which is usually longer than parents expect.

    This matters more here than anywhere else on the site, because the routes are not equivalent. Blocking an app does not block its website, and a setting inside one chatbot says nothing about the chatbot built into the device.

  2. 2

    Check they are old enough for the one they are using

    The age rules differ more between these products than between any other group of apps here, and the one that looks most like a toy is not the one with the lowest age. Knowing the rule also tells you which protections switch on by themselves, because most of them hang off the age on the account.

    1. ChatGPT: OpenAI's terms set a minimum of 13, and say that under-18s need a parent's permission. OpenAI's own help page puts it plainly: ChatGPT is not meant for children under 13.
    2. Gemini: a child under 13 can use it only on a supervised Google account and only if you turn it on. See the next step.
    3. Character.AI: the company removed open-ended chat for under-18s in November 2025, so a teenager cannot have that kind of back-and-forth conversation there at all.
    4. If they are signed in with an adult birthday, fix that first. The teen protections on these products are applied from the age on the account or from the platform's own estimate of it.

    OpenAI says an account it estimates to be under 18, or that states an age between 13 and 17, is placed into its teen experience automatically. That is worth knowing in both directions: it can switch protections on without you asking, and it can switch them off if the birthday on the account is wrong.

  3. 3

    Link your account to theirs where the product allows it

    This is the only real parent-side control most chatbots have, and it is worth setting up for what it does rather than what you might assume it does. It shapes the experience and it can switch features off. It does not let you read the conversations, and being straight with yourself about that now saves a false sense of coverage later.

    ChatGPT: Settings → Parental controls, or chatgpt.com/parentalcontrols

    1. Open Parental controls from your own account and invite your teen by email or text.
    2. They have to accept the invitation before anything applies.
    3. Once linked, go through the settings: reduce sensitive content, turn off model training, turn off memory, and remove voice mode, image generation or group chats if you want to.
    4. Set quiet hours if a time boundary would help, and study hours if you want it to ask questions before giving answers.

    The features you switched off stop working in their account, and you get a notification if the accounts are ever unlinked.

    Three limits, all of them stated by OpenAI. You do not see their conversations or transcripts. You are notified only in rare cases where their systems and reviewers detect a serious safety concern, and the notification carries only what is needed. And your teen can disconnect at any time, though you are told when they do. The protections also only apply inside their own account, so the same chatbot signed out, or signed in as you, is unprotected.

    Who holds this: Either of you can change this, and neither of you is told when the other does.

    If someone is being cruel to them

    A linked account will not show you someone being cruel to your child, because none of this reaches their conversations. What it can do is keep a chatbot from being the only place they take it, by making the safety notification route exist at all.

  4. 4

    Decide about Gemini on a supervised Google account

    This is the one case where a child under 13 can use a general-purpose chatbot through a route the platform supports, and it is off until a parent turns it on. So the decision is genuinely yours, which is unusual here, and Google is unusually direct about the limits of its own filtering.

    Family Link → [your child] → Controls → Gemini → Gemini Apps. On older versions of Family Link this sits at Controls → Content restrictions → Gemini → Gemini Apps.

    1. Open Family Link, or familylink.google.com, and pick your child.
    2. Find Gemini Apps and switch it on or off.
    3. If you turn it on and nothing appears, give it a few hours and restart Gemini. Google says access can take time to arrive.
    Recommended by age
    Under 7Off
    7–11Off unless you are going to sit with them while they use it
    11–12Off unless you are going to sit with them while they use it
    13–15On if you have had the conversation below, off if you have not
    16–17Agree it together

    Gemini either works on their supervised account or it does not, and you hold that switch.

    Worth knowing before you rely on it. Google says its filters try to limit inappropriate content but are not perfect, and that your child may encounter things you would rather they did not. Gemini Apps are not available for supervised accounts in the European Economic Area, Switzerland or the United Kingdom, so if you cannot find the switch, that may be why. The Keep Activity setting is not available under 13. And this switch covers the Gemini apps, not the AI answers that appear in ordinary Google search results.

    Who holds this: You set this from your own account, and your child cannot change it back.

  5. 5

    Turn off memory and training where the option exists

    Memory is the setting that changes what a chatbot is. With it on, the thing remembers what your child told it last week and brings it up, which is most of what makes a child feel known by it. Training is a separate question about where their words end up, and the two are usually different switches.

    1. ChatGPT, from a linked parent account: turn off Memory so it does not save or use memories, and turn off the setting that lets their conversations improve the model.
    2. Check the same two things on any other chatbot they use. Most have both somewhere in settings, under memory, personalisation or data controls.
    3. On a supervised Google account under 13, Google says the Keep Activity setting is not available, so there is nothing for you to switch there.

    The chatbot stops carrying things forward from one conversation to the next, so it feels less like someone who knows them.

    Two honest limits. Turning memory off changes what happens from here, and you should not assume it clears what the product has already stored, which is a separate request on most of them. And this is a setting inside one product, so it says nothing about the others on your list from step one.

    Who holds this: Either of you can change this, and neither of you is told when the other does.

  6. 6

    Agree what never goes into a chatbot

    There is no setting for this one, which is why it is the step that does the most work. A chatbot asks questions like a person does, and answering them feels like a conversation rather than filling in a form, so children hand over things they would never type into a website.

    1. Agree the list out loud: no full name, no school, no address, no phone number, no photos of themselves or anyone else.
    2. Add the one children find hardest: nothing about another person. Friends' names, a sibling's problem, something they were told in confidence.
    3. Agree that passwords and account details never go in, whatever reason it gives for asking.
    4. Give them the test to use instead of a rule to remember: would you say this out loud to someone at the bus stop?

    This rule has to hold everywhere on the list from step one, because the chatbot inside a social app and the assistant on the device are the same kind of thing with the same reach. It is one rule for all of them, not a setting you repeat.

  7. 7

    Check what a connected toy is actually doing

    A toy that talks back is doing the same job as a chatbot in a smaller case, and because it looks like a toy it gets left in a bedroom with nothing switched off. A younger child will tell it things, because that is what it invites.

    1. Check whether it has a microphone or a camera, and whether there is a way to switch them off or mute it.
    2. Check whether it needs an account or an app. If it does, the account is where any controls will be, and it is worth a look before the toy is unwrapped.
    3. Look for the maker's own page about what it records and how long they keep it. If you cannot find one, that is itself worth weighing before you buy.
    4. Keep it out of the bedroom at night, or switch it off, for the same reason you would a speaker that listens.
    5. If it talks like a character, say out loud to a younger child that it is a toy and the voice is not a person.

    This step is deliberately general, because these toys have no shared settings menu and no single company behind them, so SafeStart cannot name a path the way it does for an app. Treat anything that listens and answers as a chatbot your child did not sign into.

  8. 8

    Ask it something you already know, together

    This is the one thing on the page that changes how your child treats a chatbot, and it takes about two minutes. Being told it can be wrong does very little. Watching it be wrong, confidently, about something they know, does the rest of the work for you.

    1. Pick something they are certain about. A score, a date, something about your own town.
    2. Ask it together and read the answer.
    3. Then push back on a correct answer and see whether it folds and agrees with you. That tells them more about what it is than any explanation would.
    4. While you are in there, open whatever you blocked in the earlier steps and check it is actually blocked.

    They have seen it be wrong and seen it agree too easily, which is the part that sticks.

    If something a chatbot has said already worries you, screenshot the conversation before anything else, because these can be cleared from the app and some products do it for you. Most have a way to report or flag a single answer, usually on the message itself. Then treat it as you would anything else that has upset them, and if it goes further than that, the Get help now page on SafeStart covers who to contact in your country.

Safety check

Worth talking about

Official sources

Has something already happened?If your child has been contacted by an adult, is being threatened over an image, or is being bullied, settings are not the first thing to deal with. Here is what to do, and who to tell.

Menus move. If something on screen does not match these steps, check the official link above.