# Can AI Teach Kids to Read? The Part That Has to Happen Out Loud **Author:** Dan Fitzpatrick **Canonical URL:** https://parents.theaieducator.io/posts/can-ai-teach-kids-to-read **Published:** 2026-09-16T07:26:30.000Z **Updated:** 2026-09-16T07:26:38.782Z **Category:** AI by Age ## TL;DR Partly. AI can read to your child, widen their vocabulary and give them more to talk about, and a tool that listens while your child reads aloud can improve their accuracy. But the mechanical part of reading, turning letters into sounds, only happens when the sound comes out of your child, out loud, to a person. The US Department of Education's practice guide on foundational reading gives its strongest evidence rating to exactly the two things a machine cannot do for a child. Tonight, have your child read their school book out loud to you for five minutes, and when they get stuck, point at the first letter and wait. If reading is not coming at all, ask the school for an assessment before you buy an app. ## Key takeaways - Reading is decoding plus understanding, and a five or six year old is almost entirely working on decoding, which is the part AI cannot do for them. - The two recommendations with the strongest evidence rating in the US Department of Education's foundational reading practice guide are the two that only work when the sound comes out of your child. - Tools that listen while a child reads aloud have research behind them. In a 2026 trial across two US districts, children simply given access to an AI literacy platform used it for two to five minutes a week and nearly half never opened it. - Your child's school reading book contains only the letter sounds they have been taught so far. An AI story cannot do that, because it does not know the sequence. - Run the Sounding Out Test before you buy: in five minutes with the tool, how long is your child's own voice sounding out words they have not seen before? ## Article Partly, and the part it cannot do is the part that matters most at five and six. AI can read to your child and widen their vocabulary, and a tool that listens while your child reads aloud can improve their accuracy. But turning letters into sounds has to happen in your child's own mouth, out loud, to someone. If reading is not coming, the school's assessment comes before any app. That is the short answer. The longer one is worth five minutes, because "reading" is really two jobs, and AI is good at one of them and close to useless at the other. ## Which part of reading do you actually mean? Reading is decoding plus understanding, and a six year old is almost entirely working on decoding. Decoding is the mechanical part: seeing the letters, knowing what sound each one makes, and blending those sounds into a word. Understanding is everything else: knowing what the words mean, following a story, working out what is implied. A fluent adult reader does both at once and has forgotten that decoding was ever separate. A five year old has not, because decoding is all they are doing. That distinction is not a writer's tidy metaphor. It is how the official guidance is built. The [Institute of Education Sciences practice guide on foundational reading skills](https://ies.ed.gov/ncee/wwc/practiceguide/21), published by the US Department of Education in July 2016 and revised in December 2019, sets out four recommendations for kindergarten through third grade, each with a rating for how solid the research behind it is. Two of the four carry the panel's highest rating, strong evidence: - Develop awareness of the segments of sounds in speech and how they link to letters - Teach students to decode words, analyze word parts, and write and recognize words Reading connected text every day gets a moderate rating. Teaching academic language and vocabulary gets the panel's lowest rating, minimal evidence, which is a judgment about how conclusive the studies are rather than a verdict on whether vocabulary matters. It matters enormously. The research just has not pinned down the causal link as tightly. Now hold that list up against a chatbot. Here is what I make of it, and it is the whole argument of this article. The two recommendations with the strongest evidence behind them are the two that only work when the sound comes out of your child. The one AI is brilliant at, talking and explaining and widening vocabulary, is the one the panel rated weakest. A machine that talks is playing on the easy end of the board. ## What the research on AI reading tutors actually found It works when the child reads aloud and the machine listens. It does almost nothing when the machine reads and the child listens. There is a good randomized trial on this. [Bai, Hubers, Cucchiarini, van Hout and Strik, published in Technology, Knowledge and Learning on June 27, 2025](https://pmc.ncbi.nlm.nih.gov/articles/PMC13526040/), gave 525 Dutch first-graders aged six and seven, across 44 schools, six weeks with a reading tutor that used automatic speech recognition: the child reads out loud, the software listens and responds to the words they get wrong. Every version of the setup produced gains in reading accuracy. When children were reading whole sentences, clear and explicit feedback about the error beat subtler correction. There was a trade-off the authors are honest about: accuracy improved, and reading got slower. The researchers also found the software agreed with teachers less well about errors on single words than on sentences. Read the design of that study again, because it is the opposite of what most parents picture when they hear "AI reading app". The child does the sounding out. The machine does the listening. Then there is the awkward finding. In June 2026, [Carly Robinson, David Gormley, Ana Trindade Ribeiro and Susanna Loeb published two randomized trials](https://edworkingpapers.com/ai26-1451) of an AI literacy platform across two US school districts: 174 children in grades 1 to 5 in one, 181 in grades 1 to 3 in the other. Children who were simply given access used it for two to five minutes a week, and nearly half of them never opened it at all. Adding a human tutor whose job was purely motivation and accountability, not teaching, raised engagement by 71 to 80 percent. It still was not enough. The platform's own provider says children typically need at least 30 minutes a week before academic benefits show up. The authors report plainly that "the intervention did not improve reading achievement", and their closing line is the one I would put on the fridge: > "Rather than replacing human relationships, AI-based personalized instruction may still depend on them." So the honest summary of the evidence is narrower than either the app marketing or the panic. A tool that listens to a child read aloud and corrects them can help. A tool nobody opens cannot. And the thing that decides which one you bought is not the technology. It is whether an adult is sitting there. ## The Sounding Out Test One check tells you what you are buying. > **The Sounding Out Test.** The Sounding Out Test is the one check I would run on anything sold as AI reading help: in five minutes with it, how long is your child's own voice sounding out words they have not seen before? If the answer is none of it, the tool may be building vocabulary, comprehension or appetite, and all three matter. It is not teaching your child to read. Run it literally. Sit next to your child, open the thing, and watch the clock. Most products fail this test instantly, because talking is easy to build and listening to a six year old mispronounce a word is hard. ## What I See in Schools Early reading is the most precisely engineered thing a school does, and most AI tools know nothing about it. I taught in secondary schools and I have trained more than 150,000 teachers, so I come to this from the other end of a child's schooling. What strikes me about the early years is how deliberate they are compared with everything that follows. By the time a child is writing essays, teachers are working with judgment and professional instinct. At five, they are working through a sequence. Here is the part that changes how you shop. The book your child brings home from school is chosen to contain only the letter sounds they have already been taught. That is the whole point of it, and it is what phonics teaching means in practice: the letter sounds are introduced in a fixed order, a few at a time, and the reading book never runs ahead of the order. England's [Reading Framework, published by the Department for Education in July 2023](https://assets.publishing.service.gov.uk/media/664f600c05e5fe28788fc437/The_reading_framework_.pdf), says pupils should "read aloud accurately books that are consistent with their developing phonic knowledge and that do not require them to use other strategies to work out words", and that such books are "carefully structured in cumulative steps for pupils learning to read, so that they can decode every word as their knowledge of the alphabetic code increases". The framework is blunt about why this is engineered rather than left to chance: "Most texts are not decodable for children who are beginning to learn to read." An AI will write your child a story about a dinosaur who runs a bakery. It has no idea which sounds your child was taught on Tuesday. **Your child's school gives them a book that is deliberately too simple, because it contains only the sounds they have been taught so far. An AI will write them a story about anything they ask for. One of those is reading practice. The other is a bedtime story with extra steps.** You can push back on this a little, and it is worth knowing how far. You can ask a chatbot to write a story using only certain letter sounds, and it will make a decent attempt. It will also slip in words outside the set, so you would have to check every word, and to check every word you would need to know which sounds your child has covered. Which is a question only the school can answer, and one almost no parent has thought to ask. That gap is the thing I would fix. Not the app. The gap between a school running a carefully sequenced reading program and a family at home with a tablet, where nobody has ever told the family what the sequence is. ## What is appropriate at each age The answer changes fast between three and nine, faster than at any later stage. **Under five: talk, do not teach.** Before a child is decoding, the useful work is spoken. Stories, questions, songs, made-up characters, being read to. This is the one place a chatbot earns its keep, as a source of things to talk about rather than as a teacher, and it should be you holding the device. [Ofcom's Children and Parents report, published on May 21, 2026](https://www.ofcom.org.uk/siteassets/resources/documents/research-and-data/media-literacy-research/children/2026-children-and-parents-report/children-and-parents-media-use-and-attitudes-report-2025-6.pdf?v=418231) from 5,008 interviews with UK parents, found that among parents of three to seven year olds whose child goes online for creative activities, just over half, 52 percent, say their child has used AI in those activities. That is a subgroup rather than all young children, but it tells you this is already ordinary. Be careful about the device that answers everything instantly, which is a separate question I have written about in [what the tests on AI toys found](https://parents.theaieducator.io/posts/are-ai-toys-safe-for-kids). **You could say:** "Let's ask it to make up a character, and then you tell me what happens next." **Five to seven: this is the decoding window, and it belongs to you and the school.** These are kindergarten through second grade in the US, Reception through Year 2 in the UK, and they are the years where a few minutes a day of your child reading out loud to a person does more than any product. If you want technology in it, use the kind that listens. Google's own documentation for [Read Along](https://support.google.com/readalong/answer/12279465), read on September 16, 2026, describes a reading buddy called Diya that "uses Google's advanced text-to-speech and voice recognition technologies to listen and respond to students with real-time feedback and encouragement as they read aloud", and says it works without Wi-Fi. When Google launched the web version in August 2022 it stated that "all the speech recognition happens in the browser so children's voice data remains private and we do not send it to any servers". I am naming it as an example of the category that matches the evidence, not as a recommendation over what your school uses. Check its current availability where you live before you rely on it. **You could say:** "You read it to me first, then I'll read it back to you and you catch my mistakes." **Seven to nine: the job shifts from decoding to fluency and meaning.** Once a child can decode most of what they meet, AI becomes more useful, because now the bottleneck is understanding rather than sounding out. This is where asking a chatbot what a word means, or to explain a confusing paragraph, starts to earn its place. It is also where the first arguments about doing the work begin. If you are looking at a paid tutoring product at this age, the test to apply is in [whether AI tutors actually work and which kind helps](https://parents.theaieducator.io/posts/ai-tutor-for-kids), and it is a different test from this one. **From nine onward the question changes entirely**, from teaching a skill to setting boundaries around a tool your child can already use well. The guide for that band is [AI for nine to twelve year olds](https://parents.theaieducator.io/posts/ai-for-9-to-12-year-olds). ## What to do if reading is not coming The mistake I see most often is buying a reading app for a child who needs a reading assessment. If your child is six or seven and reading is not coming, the temptation is to buy something. I understand the instinct: a product feels like action, and a wait for a school appointment does not. But no app diagnoses anything, and a child who is struggling to hear the individual sounds in words needs a person to work that out. Ask the school for a conversation about your child's reading, ask what their phonics assessment shows, and ask what they would do next. If there is a question of dyslexia or another specific difficulty, that is a specialist assessment, not a subscription. The school's judgment and a professional's assessment come before anything in this article. ## What none of this can guarantee No tool can hear what a person hears, and no setting makes a machine a teacher. Speech recognition on a young child's voice is hard. The Dutch trial found the software agreed with teachers less well about errors on single words than on sentences, which is worth remembering before you treat a score from one of these tools as a measurement. And a system trained on typical speech has more trouble with a child whose speech is not typical. If your child has a speech difference or a strong regional accent, ask about that before you rely on a tool that corrects pronunciation, because being told you are wrong when you are right is discouraging in a way no parent wants attached to reading. Two other limits are worth saying plainly. The mainstream chatbots are not built for this age at all: [OpenAI's own help center](https://help.openai.com/en/articles/8313401-is-chatgpt-safe-for-all-ages), read on September 16, 2026, states that ChatGPT "is not meant for children under 13" and that where it is used in education "the actual interaction with ChatGPT must be conducted by an adult". That is the company's rule, not mine. And your child's school rule comes first on anything to do with schoolwork or a school device. There is a plain-language version of the age rules across the main tools in [can kids use ChatGPT](https://parents.theaieducator.io/posts/can-kids-use-chatgpt). ## What to ask the school One question, and it has a short answer. Ask your child's teacher: which letter sounds has my child been taught so far, and is there anything you would like them to practice at home? It is a small question and it does something no app can do. It tells you which words your child can be expected to work out for themselves, which is the difference between reading practice and being read to. If you want a second one, ask whether any reading software the school uses is matched to the same phonics sequence as the reading books. A school that has thought about it will know. A school that has not will now be thinking about it, which is usually how these things start moving. ## The one thing to do tonight Five minutes, your child's voice, your ears. Get the school reading book, not a screen, and have your child read it out loud to you. When they get stuck, do not say the word. Point at the first letter and wait. The waiting is the lesson. That is the thing the evidence keeps pointing at and the thing no product has managed to replace: a child sounding out a word they cannot yet read, with somebody they love sitting next to them. Then, if you want technology in the mix, run the Sounding Out Test on whatever you were about to buy. If your child's school has not spoken to parents about how it is using AI, including in early reading, that is the work I do with schools: [checking whether a school is actually ready for the tools it is already using](https://theaieducator.io/school-ai-readiness). It is worth sending to your child's principal or head teacher. And I write a [Sunday newsletter](https://theaieducator.io/#newsletter) read by more than 44,000 teachers and parents, which keeps up with this without the panic. Parents are very welcome. *Dan Fitzpatrick is The AI Educator: a former teacher and assistant headteacher, a Forbes contributor, and the author of bestselling books on AI in education. He has trained more than 150,000 educators across more than 30 countries and uses these tools every day. [More about Dan](https://theaieducator.io/about).* ## Frequently Asked Questions ### At what age can AI actually help my child with reading? From about seven, once your child can decode most of what they meet. Before that the useful help is spoken rather than technological: being read to, talking, and a few minutes a day of reading out loud to a person. Between five and seven, the only kind of tool that matches the evidence is one that listens to your child read and corrects them, rather than one that reads to your child. ### Are paid AI reading apps worth the money? Only if your child will actually use them with an adult nearby. The most useful finding for a parent about to subscribe comes from Robinson, Gormley, Ribeiro and Loeb's two randomized trials published in June 2026: children given access to an AI literacy platform averaged two to five minutes a week, nearly half of those left to use it independently never opened it, and the platform's own provider says at least 30 minutes a week is needed before benefits appear. Adding a person purely for encouragement raised engagement but still did not improve reading achievement. ### Can ChatGPT help my six year old learn to read? It is not built for a six year old at all. OpenAI's help center, read on September 16, 2026, states that ChatGPT "is not meant for children under 13" and that in education "the actual interaction with ChatGPT must be conducted by an adult". So anything it does for a six year old happens through you. It can invent a story to talk about, which is worth something, but it cannot hear your child read and it does not know which letter sounds they have been taught. ### What is a decodable book, and can I ask AI to write one? A decodable book contains only the letter sounds a child has already been taught, so they can work out every word themselves. England's Department for Education Reading Framework calls for books "carefully structured in cumulative steps" for exactly this reason, and it is what your child brings home from Reception or kindergarten. You can ask a chatbot to write a story restricted to certain sounds and it will try, but it will slip in words outside the set, so you would have to check every word against a sequence only the school can tell you. ### Will an AI reading tool tell me whether my child has dyslexia? No, and it should not be used to try. Speech recognition on young children's voices is imperfect, and the Dutch trial published in June 2025 found the software agreed with teachers less well about errors on single words than on sentences. If reading is not coming, ask your child's school what their phonics assessment shows and what they would do next. A specific difficulty such as dyslexia needs a specialist assessment, not a subscription. ### Is it harmful if AI reads stories to my child? No, and there is real value in it, as long as it is not replacing the reading your child does themselves or the reading you do together. Listening to stories builds vocabulary and the appetite for books, both of which matter. The thing to watch is whether the machine has quietly taken over the part where your child does the work, and whether it has replaced the few minutes a day of reading aloud to a person, which is the part with the evidence behind it.