The SEAG has 56 scored questions. Of those, 44 are multiple choice. Marking those is easy. A computer matches the bubbled letter against the correct letter and the mark drops in instantly. The other 12 questions are different. They are free response, meaning your child writes the answer in words or numbers instead of picking from a list.
Free response questions are the ones parents dreaded marking at home. A spelling word. A short phrase drawn from the comprehension passage. A numerical answer that might be written as 3.5 or 3.50 or 7/2. Someone has to read each answer and make a judgement call. On SEAG Success, our AI marker does that. You still have the final say.
How it works
When your child finishes a practice paper, the 44 multiple choice questions mark themselves instantly. The 12 free response questions go to our AI marking engine, which runs on Claude, a large language model built by Anthropic.
The AI reads the written answer, compares it against the correct answer and a set of accepted variations, and decides whether the response is correct, incorrect, or borderline. For numerical answers it recognises equivalent forms (3.5 and 3.50 are the same number). For English answers involving spelling, grammar or a short comprehension phrase, it looks at whether the response shows the skill the question was testing. Minor wording variation is allowed where that is appropriate.
It happens in seconds. By the time your child clicks through to their results page, every question has been marked, including the written ones. There is no waiting and no post-paper marking session for you.
Which questions get AI-marked, and which do not?
The 12 free response questions split roughly evenly between English and Maths. The Maths ones (typically Questions 51 to 56) are numerical, and they are handled by a numeric comparison engine, not the language model. That is a much simpler check. The system tests whether your child's number matches the correct number, accounting for decimals, fractions and units. Numeric marking is effectively 100 per cent accurate because the comparison is objective.
The AI language model is used for the English free response questions, where the answer involves actual words and a small amount of judgement is needed. That is where machine marking has to be cleverer than a string match, and where the accuracy conversation matters.
How accurate is it?
Honestly? Very high, for this kind of question. No marking system is perfect, including human ones. But SEAG free response questions are short by design. A word. A short phrase. A sentence at most. That narrow answer space is exactly the ground on which modern language models are reliable.
Across thousands of answers in our own testing, the AI agrees with our human judgement the vast majority of the time. The place we see disagreement is borderline spelling, where a child's answer is close to correct but not quite there. In those cases the AI tends to mark strictly, awarding the mark only when the answer clearly meets the standard. We prefer that. Better to practise under a slightly stricter marker and be pleasantly surprised on test day than the other way round.
For a sense of the English skills the AI is checking, look at our Spelling Superstars lesson and Punctuation Perfection lesson. Both walk through the error types the free response questions tend to test.
Parent review and override
Every AI-marked answer is visible to you on the parent dashboard. You see your child's response, the accepted answer, and the judgement the AI made. If you disagree, you can override the mark with one click and the score updates.
This is the part of the design that matters most. The AI is not making decisions about your child's progress that you cannot see. It is doing the routine sorting so you do not have to, and handing you the borderline calls. Before we built this, parents who wanted their child to do full practice papers had to sit down after every one and mark 12 written answers. A lot of parents told us that was the point at which regular practice stopped. The AI removes the bottleneck. You step in when your judgement is useful.
What about marking mistakes?
They happen. If the AI marks an answer wrong that you think should be right, or awards a mark you think should not have been given, override it on the dashboard. The overridden mark is the one that counts for your child's score and progress tracking.
If you spot a pattern where the AI is getting a specific question wrong repeatedly, let us know. We look at that feedback and tune the accepted-answer lists so the system gets better over time. We would rather hear about a marking issue than have you work around it quietly.
Privacy and data
Your child's answers are sent to the AI marking engine, processed, and the result comes back to our platform. We do not use your child's answers to train AI models. The answers go in, the mark comes out, and the full record sits in your account on the dashboard. Our privacy policy covers the detail in full. Handling student data carefully is not optional for us. It is the baseline.
Why it matters for preparation
Full-paper practice under realistic conditions is one of the single most useful things a P7 child can do before the SEAG. Short of sitting in an exam hall, it is the closest experience available. Without AI marking, the free response questions were the reason many families only ever did the multiple choice sections at home. With the marking handled, your child finishes a full 56-question paper and sees their complete score within minutes.
More practice, faster feedback, better preparation. For where full practice papers sit in the year, our P6-to-P7 roadmap places them month by month. And our walkthrough of the SEAG paper explains why each of those 12 free response questions is worth practising.
