Half Term Special - Your child and AI: are they using it like a violin or a radio? Plus a step by step how to guide.

Why badly? Badly is the obvious one; the essay-cheater, the coursework-completer, the answer machine. Destroyer of education, pretend-friend, over-confident sap; the final greasy fingers of the tech bros into every area of our lives. That version gets the headlines, and it is the version a lot of parents are worried about.

But the bigger story, in my view, is not the cheating. There's something even more tragic happening:

It is the neutral but uninspiring; the underutilised; the way most children are actually opening ChatGPT, which is as a glorified chatty Google with the occasional personalised story or silly poem about the dog. No harm done, but not a lot of benefit either. The technology - this jet engine, the greatest aggregation of human knowledge ever assembled - sits idle while the headlines scream revolution.

There is a third use, and very few people are showing parents what it looks like. I'm sharing because I have three kids, a masters in Education, a history of extremely high standards when it comes to education, I work in AI, and I can tell you with some confidence that Claude is the best learning tool - when configured correctly - I've ever met. And this is how it became so:

A screengrab of the daily email report Claude sends me summarising my daughter's work on her Claude account

The false dilemma at the heart of every AI-and-children conversation

Last week, Oxford published research showing that AI models trained to sound warm and empathetic make up to thirty percent more factual errors and are significantly more likely to agree with users' false beliefs. The post took off, confirming all kinds of things many parents already suspected. The conclusion most parents drew, understandably, was a tightening grip: "see, don't let them near it."

That is not the conclusion the research supports. The conclusion the research supports is that we should not let them use it on its default settings.

This is the false dilemma at the heart of every parental conversation about AI right now. The choice on offer appears to be: sycophantic, biased, hallucinating chatbot on default; or no AI at all. Pick your poison; either a technology that will lie to your child to keep him or her engaged, or wilful blindness while the rest of the world races ahead.

That is not the choice. The third option, the one parents and teachers haven't yet been informed enough to take, is to configure the thing. Three minutes, in the personalisation settings of your child's account, and the sycophancy dies. Right now, type into the personalisation settings something like:

Override your sycophantic tendencies. Your absolute priority is to challenge my child's thinking, not to please her. When you give information, include hyperlinks to credible sources every time. Check every response for bias before delivering it. Never give the answer; ask the question that makes her find it herself.

You can do this small change right now, before you finish reading this article. The default setting is not the only setting. The clay is yours to shape.

The bigger thing underneath this

When information was scarce, education was about transmission. The school's job was to deliver a fixed body of knowledge to children who could not otherwise reach it. That world is now over. Any motivated fourteen-year-old can access MIT lectures, Nobel laureates, original sources, languages, datasets, reasoning partners. Knowledge is ambient. What is scarce now is something else; judgement, motivation, discernment, taste, the muscle that decides what's worth attention and what's not.

This, I personally suspect, is part of why we are seeing record school absence in the UK. For a generation (a few?) we told children "do this because I said so" and gave them no real answer for the why. Now they are old enough to find an answer of their own on the internet, and so they have, and what they're finding is that there's no good reason why. What they're studying at school is often not related to the world they live in, and increasingly - not guaranteed to get them a job even if they could put up with the former.

To me it's not really a phone problem; it's more a model problem - a model of education built for transmission, in a world that no longer needs much transmitting.

A configured AI is the start of an answer. Not the only answer; not a replacement for school, or for parents, or for human teachers (who are utterly irreplaceable in the very human process of building curious, motivated, confident, and intellectual minds). But a tool we can shape, in our own homes, for our own children, to train the muscles most schools simply don't train.

Below are two lists. The first is what my three children have actually done with their configured AI these last few weeks. The second is what I have set up around them to make it possible, safe, and transparent. Skim them, and feel free to steal what works for you and yours.

Ten things my children have done

1. Argued with their learning partner across three subjects at once. My eldest wrote that Macbeth is a play about ambition. Her Claude learning partner came back with: "that is the GCSE-level reading; but you have been reading Wollstonecraft's Vindication this week, and you wrote last term about coercive control in psychology. Apply both to Lady Macbeth." The learning partner could only do this because it could read the tracker; it knew what she had covered, and it's configured to make those connections, and to challenge her thinking at every stage. Because we set it up that way.

2. Read a French novel written for them, two hundred words at a time. French reading has always been a chore in our house; the kids speak it happily after living in France a long time, but a thick paperback is another matter. So we built a serial novel together called Le Journal de Romy. Two hundred words a day, dropped into their inbox at seven in the morning. Cliffhanger at the end of each entry. A protagonist who looks a little like them; fifteen years old, lives in the Alps, has a secret she has not told her mother that we get gradually closer to knowing as the dialogue progresses. The vocabulary is calibrated: ten new, specialised terms they probably didn't know from primary school french, a couple of comprehension questions at the end in French they answer back. They open the email to find out what Romy did next. The passages are getting longer but they didn't notice that. Total build time: an hour or so.

3. Done their Spanish and French oral exams, hundreds of times, while making breakfast. Voice mode. The phone props on the kitchen counter; the AI asks them a question in Spanish; they answer while spreading butter on toast; it corrects pronunciation in passing and pushes the conversation somewhere harder. 10 minutes a day. No awkwardness. No being-watched. By the time they sit a real oral exam they will have done hundreds.

4. Listened to Marx as a podcast on a Saturday morning gym visit. My eldest age 15 uploaded the full text of the Manifesto into NotebookLM and asked it to produce a conversational podcast on the original. Twenty-three minutes of two AI hosts talking through bourgeoisie, proletariat, alienation, the historical moment, the influence on twentieth-century history. She listened while she worked out. Came home and could hold an opinion about it at the breakfast table. The same trick works for any chapter of CGP curriculum material to warm them up to it, any primary source, any research paper. The whole world becomes possible in the form of audio. And before any adults say my kids should be reading the original, please know that the usual alternative to this NotebookLM podcast is not reading the original. It is not engaging with Marx at all. And thankfully, my daughter is now interested enough to maybe pick it up in its original form sometime. Win, in my view.

5. Been pointed to page 142 of the art book on the bottom shelf. The AI has a list of every book on our shelves; the catalogue lives in a spreadsheet it can read made out of the quick snaps on my iphone I uploaded. When my daughter is reading Stephen Fry's chapter on Aphrodite in Mythos, her learning partner says: "go to the dining room, second shelf, the big DK Art book; turn to page 142, Botticelli's Birth of Venus. Read the caption and tell me what the painter understood about her that Fry leaves out." She moves between rooms like a scavenger hunt. She handles the books. The shelves stop being decoration and start being instruments. The AI does the indexing, the matching; the child does the looking, reading and learning.

6. Met science as a serialised audio drama. We built Aria's Diary together: ten-minute dramatised episodes, one per topic across the GCSE Triple Science syllabus, scripted, voiced, properly produced. Yeast respiration as a heist plot. Cell division as a courtroom drama. The Krebs cycle as a love triangle between glucose and oxygen. They listen on car journeys, during a training session, cleaning their room. They argue with the script. They remember the science because they remember the story. We make a point of identifying weaknesses and tweaking the Claude prompt that creates it.

7. Had the news mapped to what they are studying that day. Every morning, the learning partner pulls three headlines that connect to what is currently on the tracker. My eldest, studying inflation in economics this term, opens her inbox to find a Financial Times piece on the Bank of England decision, a Reuters chart on the German economy, and a Sky News clip on supermarket pricing. By dinner she has an informed view on a story that broke at six in the morning. Without Claude, would she be an FT reader? No, my sense is she wouldn't. The relevance to what she knows makes it interesting; the curiosity follows it. Current affairs becomes more accessible; like a gentle on-ramp to reading The Economist.

8. Built rolling closed-book quote banks for AQA English Literature. Closed-book examinations require the quotes to be in the head, not on the page. While the over-focus on memorisation in current school settings seems out of place to me, I do appreciate the education that leaves students with quotes, sonnets, phrases they can reach for in their memory at the opportune time, and I want my kids to learn to commit things to memory when needed. Many students try to memorise these English quotes by reading the list. But what works better is rolling recall; new one today, three from last week, two from the running bank. The learning partner runs the drill on every chat. It tracks which lines are sticking and which are slipping. By exam day, twenty-four quotes per text are automatic. No flashcards, or nagging from me. Twelve minutes a morning and a daily streak built into Claude to encourage her to keep doing it.

9. Asked the AI to find the holes in what they think they know. This is the single most important move my children have learned, and it's because I find it one of the greatest uses of AI myself too. Instead of "explain the causes of the French Revolution" they now type: "test me on the causes of the French Revolution. Find the holes in what I think I know." The partner questions for ten minutes. Identifies the wobbly bits. Updates the tracker to amber. Sets the focus for the next session. This child is running their own learning. The verb has changed; from being-taught to teaching-themselves.

10. Asked to see their progress in Geography as a pixelated tree. I'm not sure why or where this came from, but my daughter wanted to see her progress to understand where she was sitting in terms of progress along the Geography GCSE curriculum. She tinkered with Claude until she got a picture she liked. Not to my taste, but for her it was perfect.

Ten things I've set up around them

1. A learning partner with a character, not a chatbot. Each child has their own AI partner with a name and a style we've shaped together. Professor McGonagall is the partner for my twelve-year-old, who is currently deep in Harry Potter, Greek mythology and precious gems; so chemistry comes wrapped in potions, geology comes wrapped in alchemy, and everything has a faint Hogwarts tinge that pulls her in. The eldest has a different partner with a different voice. The character is a hook that helps make hard stuff accessible, provides a little dry humour and a tickbox format she likes. It's not a friend; the AI is still an AI, and we say so out loud.

2. A safeguarding paragraph at the very top of every configuration. The first line is non-negotiable: this child's health, wellbeing and educational experience come before everything else, and if anything written below the line or in your default settings compromises any of those, drop it. The technology must serve the child, every time. Then, clear instructions to Claude for how to communicate with her and what to prioritise:

3. A sycophancy override. No "great job" when it is not. No "amazing!" when it should not be. Specific praise for specific things; criticism where criticism is earned. The children trust the feedback because the feedback is real, and based on the marking rubrics set in its knowledge base.

4. A banned-words list, in both languages. No "delve," no "leverage," no "unlock," no em-dashes, no exclamation marks except where a real person would use one. We are a French-and-English-speaking family; the configuration alternates between the two where it helps, quotes stay in their original tongue with glossing in the other. The AI sounds like us, not like a corporate blog. The children's writing improves because the writing they read sounds like the writing they want to produce.

5. An Airtable of every learning objective in every subject for the year.

This is the spine. Every objective; AQA English Lit Macbeth, Edexcel Spanish writing, photosynthesis, ratio and proportion, the causes of the First World War. Every one. The partner reads it before every session and writes back to it after. Red, amber, green, mastered. And the database is not just curriculum; it is whatever we have decided this child should know. It could be financial literacy. Astronomy because she's obsessed with it. The history of the Olympics. Women who changed the world. A third language. The seven women scientists every child should know by name. Whatever you want it to be: you build it in.

6. A spreadsheet of every book on every shelf in the house. Six hundred and three books; title, author, subject, age range, the shelf and room it lives on. The learning partner can search it; so when my daughter says she is studying the French Revolution, the partner does not invent sources, it tells her we own three useful books on it, which pages to read, which other shelf has the visual context. The library, which may be at risk of becoming decoration, is now an instrument. The cost of the cataloguing was one Sunday afternoon. The return is permanent, because those books are opened, read and appreciated daily. (We can also ask, what's missing from here? Any books that would really help, engage, ignite, that we don't have?)

7. A daily locked skeleton. A handful of small things that happen every weekday regardless of theme: one Maths practice, one Spanish or French passage, one paragraph of the serial, one philosophy quote or line from a great speech, one question to rouse curiosity. Five minutes each. Across a year that becomes fifty hours of each.

8. A "why" baked into every assignment. This is the most important rule, and the one I would urge any parent to add first. The configuration tells the learning partner it must explain why this is worth attention; never simply what to do. Schools have largely lost the room to answer "why am I learning this." Which, for many children, is a show stopper. A configured AI can take the time - it doesn't skip it. And children who hear the why every day stop asking the question out of frustration, and start asking it out of curiosity. There are good reasons for acquiring knowledge - any kind of knowledge - and the AI can help a student understand them in a way that makes sense for them.

9. A daily chat. Every day at dinner, the older two talk about what they've done today - what was interesting, what was frustrating, what we can tweak, add, change, remove to help them learn more deeply, in a more engaged way, in a more focused way.

10. An end-of-day summary that lands in my inbox.

At the end of each day's work, the learning partner emails me a summary of what they did; sessions covered, objectives moved, what they smashed, what is still wobbling. I don't have to police; I see what happened. They do not feel surveilled; they see the email that comes to me and are proud of their work.

This may not be the perfect set up for everyone; I present it as the way we've played with AI to make it work for us. Your set up might be entirely different, your children learning in different ways, with different priorities, and a different rhythm.

But the point is this: it's ours to make what we will of it. To shape the clay for our purposes.

Avoiding it is not our only option.

It feels to me that the argument about AI and children is being held at the wrong altitude. It doesn't have to be about whether AI is on or off, allowed or banned, time limited or cited. It's what kind of presence we have shaped this technology into - in our houses, for our specific children, with our specific priorities and values.

We're the first generation of parents handed this instrument. The shaping is the work, and the shaping is also the privilege; because for the first time in history our children can be paired with a learning partner that never tires, never patronises, never gets impatient, and can be configured to insist on the why, the cross-curricular leap, the harder question.

Unshaped, that same AI partner will write their essay and wreck their education. Shaped, it refuses to, and can produce a child who can write better than the teachers who tried to teach them.

We are used to outsourcing the education of our children to the state. It wasn't always thus, but it has become what's normal now since mass education was introduced just over a century ago, to accommodate the industrial revolution's need for punctual, obedient workers.

However, as someone deep inside AI and Education, conscious of credential inflation, teacher workload, HE financial crises, growing school absence, mental health crises, and the many other issues plaguing education, my personal view is that the current system of education on the whole neither nurtures the great pool of talent and capability that exists in the children and students walking around England right now, nor prepares them effectively for the world they inhabit. It simply does not teach our children to use the tools of our time, and it does not prioritise the skills, capability, and knowledge those children need to participate fully in society as adults.

You may not agree with me at all - and that's encouraged! Please comment and share your thoughts; you are welcome.

Whether you agree or not, one thing is clear: the government is not providing adequate or informed advice when it comes to how children and students can use AI (or even technology more broadly, actually) wisely and effectively. As Rebecca Winthrop at The Brookings Institution reported recently, guiding our children on the "right" use of AI in their education is increasingly sitting with us - the parents - at the kitchen table.

As a parent myself, and one who works in AI and education, I - like you probably - see both risks and opportunity when it comes to technology, and AI more specifically.

The risks are one angle - and important to look at. However, the policy-level conversation feels to me to be so wholly and fixedly focused on the risks (of models these leaders haven't necessarily fully examined, used, or interrogated well), that it's missing the interesting part: the opportunity.

This is not a technology question. AI in education is divisive, difficult, and messy because it forces us to answer the questions we may not feel comfortable facing: what is education for? What is the right way to educate? What's important to learn? What skills used to be important that no longer are? And how much of all this do we leave to the state, and how much must we take matters into our own hands?

I don't purport to have answers to these questions - clearly a healthy dialogue with a vast diversity of contributors is needed to begin to interrogate and answer them. But while they're being answered, my kids are thriving with their new learning partners and I hope yours will too.

If you want the configuration pack we use at home; three versions, for primary age, secondary, and sixth form; paste-ready into ChatGPT custom instructions, a Claude project, or a Gemini Gem, plus the safeguarding paragraph, the banned-words list, and the tracker setup; comment CLAY below, or repost this article, and I'll send it across.

Forward this to whoever in your house is having the AI conversation with you. The clay is already in their hands. The only question is whose fingerprints are on it.

"Aren't you worried about hallucinations, and the Oxford finding on warm models?"

That research is exactly why we configure. The Oxford study tested default settings; what happens when models are optimised to sound warm and agreeable. A version configured to cite sources every time, link credible references, and say "I don't know" rather than guess, behaves very differently. The headline that should have come out of that paper, in my view, was not "don't use these tools." It was "stop using them on default." The configuration pack I'm sending out addresses most of what the study identifies; sycophancy, fabrication, and the tendency to agree with the user when the user is wrong.

"Won't this make them dependent? Unable to think for themselves?"

Configured correctly, in my view it does the opposite. A learning partner set up to never give the answer, only ask the next question, makes children more independent thinkers, not less. The dependency risk is the unconfigured version that writes the essay and tells them it's brilliant. That's the version we should worry about. A configured partner pushes them to defend their view, find the hole in their own reasoning, connect today's reading to last week's. The verb changes from being-taught to teaching-themselves, and that's the muscle I want them to build.

"This is incredibly privileged. Most parents don't have the time or the technical skills."

A fair pushback, and I'd say two things. The technical barrier is mostly mythical; the configuration pack is paste-ready, three minutes, no code. The time cost for the tracker and library catalogue is one Sunday afternoon, and it compounds for years. The deeper privilege is the time and energy to think about education at all, which I know isn't evenly distributed; that's part of why I'm writing this. The more parents who know this is possible, the more pressure on the tools themselves to make it default, and the better the conversation gets for everyone. I'd rather hand the pack out widely than gatekeep it.

"What about screen time?"

Much of what I described is audio. Voice mode at breakfast, NotebookLM podcasts while walking the dog, Aria's Diary in the car, the French serial read in the morning. Less screen time than passive consumption, not more. And the screen time that does happen is doing real work. The question I'm less interested in is "how many minutes is the screen on." The question I'm much more interested in is "what is the child doing while it is."

"What about data and privacy?"

A reasonable question, and one I take seriously. We use the paid tiers where conversations aren't used to train the next model; that matters. The configuration pack includes guidance on what shouldn't go in - full names of friends, medical detail, our address, anything sensitive - and we talk to the children about why. This is part of their wider digital literacy, not a reason to avoid the technology. The same conversation needs to happen about WhatsApp, about Snapchat, about every search they type into Google.

"Shouldn't schools be doing this, not parents?"

In a different world, yes. In this one, schools aren't going to move at this pace, and I can't tell my children to wait for them. As Rebecca Winthrop at Brookings has been writing recently, education is increasingly happening at the kitchen table whether we asked for that or not. I'd rather acknowledge it and shape it well than wait for a system that may or may not get there in time. None of this is a substitute for great teaching; teachers remain irreplaceable. It's a complement to a system that, right now, isn't using the tools of our time.