AI screen time: the mode matters more than the minutes

The timer answers a question you weren't asking

Think about what you do when the limit is up and the device goes away. You feel organised for about an hour, and then the original worry comes back, because it was never really about duration. It was: did anything good happen in there? Did they learn something, or did something get done to them? A timer cannot answer that, which is why enforcing one so rarely feels like relief.

This is the whole reason AI breaks the screen-time frame rather than just stretching it. A video is a video; you can roughly infer what the forty minutes contained. An AI session contains anything at all — a child who worked through a hard fractions problem with hints and got there, or a child who pasted the worksheet in and copied out the answers. Identical minutes, opposite outcomes, and no way to tell from the clock. So the useful move is not a tighter limit. It is a better instrument.

The problem with counting minutes

Screen-time limits became the standard advice because they're measurable and easy to enforce. But a limit is a blunt instrument: it caps the harm of the worst kind of screen use and equally caps the benefit of the best kind. Most parents already know this intuitively — you feel differently about your child video-calling their grandmother than about the eleventh short-form video in a row, even though the clock reads the same.

Major paediatric guidance has moved in this direction too: away from a single universal number for school-age children and toward a focus on what the time displaces — sleep, physical activity, meals, homework, and time with people. That's a more useful frame, and it's the one AI demands, because AI sits at both extremes at once.

Passive AI vs. active AI — what to look for

• Passive: the answer machine — Worksheet in, answers out, child moves on. No decision, no revision, nothing attempted. This is the mode that worries teachers, and it is the default behaviour of every general-purpose chatbot, because nothing in one is designed to withhold an answer.
• Passive: the endless companion — Open-ended chat with no goal. It's engaging, it's warm, and it has no stopping point by design — which is precisely the property that makes short-form video hard to put down.
• Active: working it out with help — Stuck, a hint, another hint, then a worked example — and the child still does the next one. The thinking stays with them, which is the only version of homework help that leaves anything behind. It is also the version that takes longer, so it looks worse on a timer.
• Active: directing a build — The child decides what should exist, describes it, tests what comes back, and fixes it. There's a plan, a failure, and a revision — the same shape as any good learning activity, on or off screen.

The tell: does the session produce anything?

If you want one question that sorts almost every AI session into the right bucket, it's this: at the end, is there something that exists which didn't before? A game, a story, a drawing, a fixed bug, a changed mind. Sessions that produce something have a natural end, give you something to talk about, and leave the child with evidence they can be proud of.

Sessions that produce nothing are the ones to watch — not because chatting is wicked, but because unbounded engagement is what the design of most consumer software is optimised for, and children are not well equipped to notice when they're being optimised at.

Is ChatGPT safe for children? The risks, by age

Five rules that work better than a timer

1. 1. Finish something, then stop Replace "thirty minutes" with "until the level works." It teaches completion rather than clock-watching, and it makes the stopping point the child's achievement rather than your interruption.
2. 2. Keep AI use in shared rooms Same rule as any screen at this age, and it's doing more work than any content filter. You overhear the interesting parts, which is where the actual parenting happens.
3. 3. Ask for the tour "Show me what you made and how it works." Thirty seconds, and it converts a solitary session into a shared one — and quietly reveals whether the child understood any of it.
4. 4. Protect sleep and bodies first Screens out of bedrooms overnight, and screen time that doesn't eat exercise, meals, or homework. This is the part of the traditional advice that has held up best.
5. 5. Budget passive time, not total time If you're going to cap something, cap the consumption — video, feeds, open-ended chat — and let creation have a longer leash. That's the distinction a raw minute count can't make.

Trade the limit for an instrument

Every rule above is really the same move: stop measuring the session and start seeing it. And once you put it that way, the obvious question is why a parent should have to reconstruct the session at all from tours and dinner-table questions. Most of what you want to know was right there while it happened.

So the standard worth holding any AI tool to is not how strict its timer is. It is what it can show you afterwards, without you reading a transcript over your child's shoulder: what they worked on, where they got stuck, whether they needed a nudge or a full explanation, and whether the thing they got right on Monday is still right on Thursday. A tool that can answer those four questions makes the limit mostly unnecessary. A tool that can't will have you back to counting minutes by the end of the month, because counting is the only instrument it left you.

This is also the honest reason the screen-time conversation has felt unsolvable for a decade. Parents were handed a stopwatch to measure something a stopwatch cannot see, and then told the problem was their own consistency.

What a parent can actually see: how Luchi teaches, and what it reports back

Common questions

How much AI time is too much for a 9-year-old?

There's no evidence-backed universal number, and anyone quoting one precisely is guessing. The more answerable version: is it displacing sleep, movement, homework, or people? If not, and if the child is producing things rather than consuming, you're likely in reasonable territory.

Does creative AI use really count as 'good' screen time?

When the child is making decisions, testing results, and fixing what's wrong — yes, that's active learning that happens to be on a screen. When the AI is doing the thinking and the child is watching, it's consumption with extra steps.

My child wants to just chat with the AI. Is that harmful?

Not inherently, but it's the mode with the least to show for it and no natural end. Steer it toward a purpose — build something, explain something, solve something — and it becomes a much better use of the same minutes.

Should AI time count inside their normal screen limit?

Split it. Passive AI belongs in the same bucket as video; project-based creation is closer to homework or a craft. Lumping them together is what makes the limit feel arbitrary to your child — because it is.

How do I know what my child actually did in there?

Three things in ascending order of reliability: ask to see what exists at the end, ask them to explain one part of it in their own words, and use a tool that keeps a record you can read later. Reading the transcript yourself is the option most parents reach for first and the one that works worst — it is long, it is boring, and a wall of chat tells you what was typed rather than what was learned.

What a reasonable week actually looks like

Parents rarely want principles; they want to know whether their Tuesday is normal. So here's a shape that works for a lot of families with school-age kids, offered as a starting point rather than a prescription: two or three creation sessions a week, thirty to sixty minutes each, in a shared room, each ending with something finished and shown to someone. Passive screen use handled separately under whatever limit you already use. Nothing at all in the hour before bed.

What makes that work isn't the numbers — it's that the sessions have an ending built in. The most common failure in households that feel out of control isn't excessive minutes; it's that nothing in the activity ever signals that it's done, so every session ends in a negotiation. Choose activities that finish, and most of the conflict disappears without a timer being involved.

The other thing worth protecting is the gap afterwards. A child who closes the laptop and immediately opens a tablet hasn't really stopped. Ten minutes of something physical between screens does more for how the rest of the evening goes than any adjustment to the limit itself.

Talking about it without it becoming a fight

Screen-time conversations go badly when they're framed as a judgement about the child's self-control. They go much better when framed as a judgement about the product's design — because that's both true and something you're on the same side of.

"Do you think this app wants you to stop?" is a genuinely interesting question to a ten-year-old, and it invites them into the analysis rather than putting them on the defensive. Children have a strong sense of fairness and they do not like being manipulated. Pointing that instinct at the design of the software is more effective than any rule you can impose, and it's a skill they keep long after your limits stop applying.

Stop counting the minutes. See what happened in them.

Luchi helps with homework through hints, not answers, keeps a record of what your kid knows, and writes you a weekly note from what actually happened.

Get on the list

How Luchi teaches

Part of the series: Is AI safe for kids?

Related guides in this series

• Is ChatGPT safe for kids? — A straight answer by age — the risks that actually come up, and the five habits that make AI safe when you do let kids use it.
• What is my kid doing on AI? — You can’t read every transcript. Four checks that take under a minute each, what each one tells you, and the one question a ten-year-old actually answers.
• AI apps a parent can actually monitor — Alert, transcript, or record — three different things sold under one word. What seven products say they show a parent, quoted from their own sites.