You're probably comparing a few options, and most of them will tell you they're adaptive and AI-powered. That isn't very useful, because everything is now.
This page explains the specific mechanisms: how EigenOwl decides which question to show, what the AI tutor does when your child is stuck, how exam answers get marked, and what other students can see. It ends with the things you should be sceptical about, because there are some.
EigenOwl is built and maintained by a practising Cambridge IGCSE and IB Diploma Physics teacher — someone who teaches these syllabuses, marks student work against these mark schemes every term, and sees where marks are actually lost. It isn't a content team working from a specification.
Most revision platforms are question banks with a filter. You pick a topic, you get questions on that topic, and the platform learns nothing.
EigenOwl holds a mastery score for your child against every assessment objective in their syllabus — not "forces" as a topic, but the specific skills the examiner assesses within it. Each answered question moves the relevant score.
Questions are then chosen at the edge of what your child can currently do: hard enough to require thought, not so hard that they stall. Three correct in a row and difficulty rises. A stall and it steps back rather than repeating the same wall.
The part that's unusual: those mastery scores fall over time if a topic isn't revisited. Understanding genuinely fades, and most platforms let a score from October still read as 90% in March. Here it decays, which means the number reflects what your child would remember today. Nobody enjoys watching a score drop. That is the intended effect.
The reasonable worry about AI in education is that a child stops thinking and starts copying. Nova is built specifically against that.
Nova is Socratic. When your child is stuck, it doesn't explain the solution — it asks the question that makes the next step visible. Its instructions are explicit on this point: never state the answer directly, never say "that's wrong", and treat a wrong answer as the most useful thing in the conversation, because it reveals how the student is actually thinking.
It's designed for the situations that matter most — unfamiliar contexts, ambiguous observations, problems where the method isn't obvious. Precisely where a question bank is useless and where a student most needs a patient adult who won't just tell them.
One detail that shows the care in it: Nova is instructed not to teach simplified models that will later have to be unlearned. If it simplifies, it says so — "this is a simplified picture, there's a deeper version you'll meet later" — rather than leaving a convenient falsehood behind.
Nova is not infinitely stubborn. After genuine repeated effort it will offer a small hint — narrowing the gap rather than closing it. A student determined to extract an answer can eventually wear it down, as they could a human tutor. The design makes that harder and slower than simply thinking it through, which is the realistic goal.
Nova runs on a large language model, and language models are occasionally confidently wrong. It's constrained to the syllabus and to Socratic questioning, which reduces this considerably, but it doesn't eliminate it. If something Nova says contradicts your child's teacher, the teacher is right.
Multiple-choice is easy to mark automatically, which is why most platforms stop there. But physics exams are won and lost on structured questions — explanations, derivations, multi-step calculations — and a percentage tells your child nothing about where they went wrong.
In Exam mode, longer answers are graded against the marking conventions for your child's specific board, point by point. Not a score out of ten, but which individual marks were earned and which weren't.
Results break down by assessment objective and subtopic, so patterns become visible — for instance, consistently losing explanation marks while never losing calculation marks. That's a specific, fixable problem, and it's invisible in a percentage.
Students choose a nickname when they sign up. In the discussion forum, the Challenge Arena leaderboard, and anywhere else students encounter each other, only the nickname appears.
Your child's real name is visible to them and to their teacher. It is not shown to other students, and it is not sent to the browsers of other students. Work, mastery scores and exam results are likewise private to your child and their teacher.
The competitive features — leaderboards, tiers, badges, streaks — are optional in the sense that ignoring them costs nothing. Some students are motivated by them; some find them stressful. Nothing in the adaptive engine depends on taking part.
Twenty to thirty minutes, three or four times a week, will do more than a single long session at the weekend. That isn't a scheduling preference — it's what the spacing research shows, and the platform is built around it. Mastery decay is calibrated for regular return, and sessions are designed to be finishable in a sitting.
If your child is doing three-hour marathons on EigenOwl, it isn't working as intended. The honest advice is fewer minutes, more often.
No subscription and no renewals. You buy an individual topic, a curated bundle at a discount, or the full course at the largest discount — and that one payment gives you three years of access.
Three years is deliberate rather than arbitrary: it covers the full two years of an IGCSE or IB Diploma course, plus a further year in case your child resits to improve a grade. It also means I can retire content when an exam board revises its syllabus, instead of leaving your child with material mapped to a specification that no longer exists. Buying two topics a month before the exam is a legitimate way to use EigenOwl, not a loophole.
One subtopic is free, permanently, with no card required. That exists so you can judge the thing rather than the marketing.
Students in conflict-affected regions and countries facing severe economic hardship receive full access free, applied automatically at sign-up. Education shouldn't be gated by geography.
Everything on this page is checkable, and none of it asks you to take my word for it. The research is named so you can look it up. The free subtopic is genuinely free. And the questions, the marking and Nova all show you what they are within a few minutes of use:
I answer these myself — I'm a teacher, not a support desk.