What is adaptive learning? A parent's guide

Adaptive learning is software that changes what a learner sees next based on their own answers. It keeps a model of what the child knows, schedules when things come back, and picks the next question at a chosen difficulty. Done well, in independent school trials, it adds about 0.05 to 0.20 standard deviations; it is not magic.

Peabody (peabody.kids) is a free AI-driven adaptive learning platform for kids, free through the end of 2026, and it is the worked example at the end. AI writes the stories; code decides the rest.

The three parts of any adaptive app

Everything else (the characters, the points, the AI) sits on top. If a product can't say what its three parts are, it is personalised in name only.

What the research says about each part

Learner models. The famous models are less different in practice than in marketing. Gervet and colleagues (2020) compared them on nine datasets and found simple logistic models win on small and medium datasets, and deep neural networks lead only above about a million interactions, by 0.06 AUC at most (JEDM). For a product with few children per question, a rating per child and per item, like the Elo scheme Math Garden uses, works from the first answer (Pelánek 2016).

Schedules. Spacing practice over days is one of the most reliable findings in learning science (Cepeda et al. 2008, 1,354 people, PubMed), and the active ingredient of an expanding schedule is waiting before the first test of something new (Karpicke and Roediger 2007). Fitted schedulers predict recall better than fixed ones on their makers' benchmarks; nobody has shown in a trial that children learn more with them.

Difficulty. Math Garden, a Dutch system with thousands of children, picks questions a child has about a 75% chance of getting right, because failing half the time is not acceptable in daily practice (Hofman et al. 2018). A 2019 modelling paper put the fastest-learning point for simple learners at about 85% right (Wilson et al.). Both say: mostly right, sometimes wrong.

What the trials show

On Kraft's (2020) scale for education programs at scale, 0.05 to 0.20 is a medium effect and anything above 0.20 is large. That is the honest size of adaptive practice: worth having, not a replacement for a teacher.

What adaptive learning is not

• Not learning styles. Adapting to what a child knows is not the same as adapting to a supposed "visual" or "auditory" type; more on that in learning styles are a myth.
• Not an open AI chatbot. The best AI-tutor results come from tightly constrained tutors with the solution in front of them; in the largest school trial, the open chat layer added nothing over practice.
• Not engagement. A system that adapts to keep a child on the screen longer is optimising the wrong thing. The research report behind this post sets the rule plainly: the reward for any adapting loop must be a learning outcome, never time or streaks.
• Not a diagnosis. An app's level is a placement inside that app, not a grade-level score.

How Peabody adapts

Peabody's three parts are ordinary code, written up rule by rule in how Peabody picks the next question.

1. Learner model: levels and a ladder. Each fact or word climbs from unseen to "quick and right on two different days". Each subject is a ladder (counting up to dividing; first sounds up to word meanings), and a grade only sets the first rung.
2. Schedule: a fixed ladder. Known items come back after 1, 3 and 7 days, then three weeks. A question answered right is never asked again; a miss comes back once.
3. Difficulty: about 85% known. A round is about 30 questions, at most two new ones, each taught first. Two misses in the last five stop new items for the round.

It counts rather than fits a model, which is cruder than Math Garden and easy to read. Peabody has no published efficacy study, so none of the numbers above are claims about it. Start free and watch which questions come back, or open the free library to see reading levels at work.

Learn more elsewhere

• J-PAL: teaching at the right level: 13 randomized trials of teaching children at their level, not their grade.
• EEF toolkit: mastery learning: a UK research charity rates mastery learning by impact, cost and strength of evidence.
• NSW CESE: cognitive load theory: the clearest free explainer of why the right help depends on what a learner already knows.

For about seventy more, checked and sorted by who they're for, see where to learn about AI.

Questions parents ask

What is an example of adaptive learning?

A maths app that notices your child gets every sum within 10 right, quickly, and moves on to sums within 20, while bringing back a missed fact three questions later. The key is that the next question depends on the last answers.

Is adaptive learning the same as personalised learning?

Adaptive learning is one kind of personalised learning: the software adjusts the content. Personalised learning can also mean letting a child choose topics or pace, or a teacher tailoring a plan. Asking "what decides the next question?" tells you which you are looking at.

Does adaptive learning use AI?

Sometimes. Many adaptive systems use statistics, not language models, and some use plain rules. In Peabody, AI writes and screens stories; levels, reviews and the next question are decided by code.

Is adaptive learning good for young children?

The constrained skills young children practise (letter sounds, counting, number facts) are where learning apps show their largest effects, and adapting the difficulty keeps practice mostly successful. The evidence for children under eight is thinner than for older students, so watch how your child responds.