Home → How Our AI Works

How Our AI Works — And What It Can't Do

Most companies tell you what their AI is good at. This page also tells you where it falls short, because you're making a decision about your child.

Last updated: August 2026

I built Brightstar for my own daughter. When I show other parents an AI tool that marks children's writing, the first sensible question is always some version of "how much should I trust this?" This page is my honest answer.

What the AI actually does

Brightstar uses Claude, an AI assistant made by Anthropic, for two features. Everything else on the site — the practice questions, the worksheets, the practice tests — involves no AI at all.

1. AI Writing Feedback

Your child writes a response to a prompt. When they submit it, the text is sent to Claude along with instructions to mark it against the ten official NAPLAN criteria. Claude returns a score for each criterion, one piece of specific praise, and one thing to improve. That comes back to your child's screen in a few seconds.

2. AI Reading Comprehension

Your child picks a topic — dinosaurs, netball, quokkas, anything. Claude writes an original passage at a suitable reading level, plus five comprehension questions covering literal, inferential and vocabulary understanding.

What it does well

✓ Genuine strengths

What it can't do

This is the part most companies leave out.

✗ Real limitations

The honest summary: our AI is a good practice coach and a poor examiner. Use the score to see roughly where things stand, and use the tips to decide what to work on next. Don't use it to predict your child's NAPLAN result — nobody can do that, including us.

Things we're often asked to clarify

ConcernThe reality
"Does the AI write for my child?"No. Your child writes every word. The AI reads the finished piece and comments on it — it never drafts, rewrites or completes work.
"Are these real NAPLAN questions?"No. Every question and prompt on Brightstar is original, written in a NAPLAN-style format. ACARA's actual test material is copyright and we don't reproduce it.
"Is my child's writing stored?"No. The text is sent for marking and returned. It isn't stored on our servers, and under Anthropic's API policy it isn't used to train models.
"Do you know who my child is?"No. There are no learner accounts, and we don't ask for a name, school or location — the request carries the writing text, the prompt and the year level. Anything your child types is sent, so please keep identifying details out of the writing.
"Is this endorsed by ACARA?"No. Brightstar is completely independent and has no affiliation with, endorsement from, or approval by ACARA.
"Will it replace a teacher?"No, and it shouldn't. It's practice between lessons, not a substitute for one.

Other limitations worth knowing

What we promise instead

  1. We'll be accurate about NAPLAN. Current terminology, correct criteria, real score ranges — and corrections when we get something wrong.
  2. We won't collect your child's data. No accounts, no tracking of children, nothing sold, ever.
  3. We won't fabricate anything. No invented testimonials, no inflated numbers, no claims of endorsement we don't have.
  4. We'll stay free. No paywall appearing once you're invested.
  5. We'll tell you what we can't do. This page exists because you deserve to know before you rely on us.

For teachers and schools

If you're considering pointing families to Brightstar, there's a page covering what it includes, how Brightstar creates and reviews educational content, and exactly what data is and isn't collected — plus a printable one-pager and newsletter text you're welcome to use: Brightstar for schools.

Found something wrong?

Please tell us. Every error a parent or teacher has reported has been fixed, usually within a day or two. That feedback has made Brightstar meaningfully better, and it's the main reason to trust it.

Try it and judge for yourself

Free writing feedback on all 10 NAPLAN criteria. No signup and no cost. See how we handle data. Make up your own mind.

Try AI writing feedback →

Frequently asked questions

Is AI writing feedback as accurate as a real NAPLAN marker?

No. Our feedback is an estimate designed to help a child improve. Trained markers apply detailed descriptors and national moderation that we can't replicate. Treat our score as a guide to what to work on, not a prediction of the official result.

Will the AI give the same score twice for the same writing?

Not always. AI models vary slightly between runs, so a criterion may score a point differently on a second submission. The advice about what to improve is usually consistent — and that's the part that matters.

Which AI model does Brightstar use?

Claude, made by Anthropic. We use their faster model so feedback arrives in seconds rather than minutes, which matters for children's attention spans.

Is it safe for my child to use unsupervised?

There's no chat, no open-ended conversation, and no way for a child to talk freely with the AI. They submit writing or choose a topic, and receive structured feedback. There's nothing to sign up for and no way to contact anyone.

What if the AI says something wrong?

Trust your own judgement, and please let us know or email hello@brightstarprep.com. We fix reported errors quickly.

Written by Naveen — a parent in Sydney who built Brightstar for his daughter. I wrote this page because I'd want it before letting an AI tool near my own child's schoolwork. Questions, corrections or concerns — please send them to hello@brightstarprep.com or through the feedback form.