Analysis

Why Writing Is the One Selective Component Where AI Actually Helps

Written by Hao, a Sydney data analyst and parent of a 2025 NSW Selective student. Last updated 8 July 2026.

Of the four NSW Selective test components (Reading, Mathematical Reasoning, Thinking Skills, and Writing), writing is the one where AI helps most, and not because AI is a good writer. The other three are closed questions with a single right answer and endless practice that marks itself. Writing has no answer key, so the scarce resource is feedback, and feedback is exactly what AI can supply instantly and for almost nothing. This piece lays out that argument, extends it to any prompted writing (NAPLAN, scholarship papers, school essays), and is honest about the two things AI genuinely cannot do here.

When my son was preparing for the NSW Selective test, I tried AI tools across all four components. For three of them the AI was a modest help. For the fourth, writing, it changed how he practised. The reason is not that AI is clever. It is about what kind of learning problem each component is, and it is worth being precise about, because the same reasoning tells you exactly when to trust AI and when not to.

The Four Components, and the One That Is Different

The NSW Selective High School Placement Test has four components: Reading, Mathematical Reasoning, Thinking Skills, and Writing. The first three are multiple choice. Every question has one defensibly correct answer, and every practice book, past paper and app comes with the answer key. For those components the scarce resource is practice volume, and it is not scarce at all. A child can do a hundred questions and mark them without any adult in the room. AI can explain a solution a second way, which is genuinely useful when a child is stuck, but it is a smaller gain, because a full shelf of practice with answer keys already does the marking.

Writing is different in kind. It is a single 30-minute response to a prompt, and there is no correct answer to check against. It is judged against qualitative criteria: whether the ideas are developed, whether the structure holds, whether the vocabulary is precise, whether the sentences vary, whether the punctuation and spelling are controlled. You cannot mark it from the back of the book, because there is no back of the book.

Component	Answer type	What is scarce	AI's marginal value
Reading	One correct answer	Practice volume (abundant, self-marking)	Small
Mathematical Reasoning	One correct answer	Practice volume (abundant, self-marking)	Small
Thinking Skills	One correct answer	Practice volume (abundant, self-marking)	Small
Writing	No correct answer	Specific feedback on the child's own piece	Large

In Writing, the Bottleneck Is Feedback, Not Practice

A child does not get better at writing by writing into a void. They improve through a loop: write something, get specific feedback on it, revise, repeat. The important word is specific. Not "use more varied sentences," but "these four sentences in a row all start the same way, here is where to break the pattern." Not "develop your ideas," but "this paragraph makes a claim and never gives an example." That feedback has to be about this essay, this child, these sentences.

Now ask who supplies it. A private tutor can, at sixty to a hundred dollars an hour, marking perhaps one piece a week, with the feedback arriving days after the writing. A classroom teacher has thirty children and can rarely give more than a few lines per essay. A parent often cannot judge a piece against the marking criteria with any confidence, and feedback from a parent carries an emotional charge that a tired ten-year-old does not always take well. So most children write far less than they should, and get real feedback on a fraction of what they do write. The bottleneck is not effort, and it is not a shortage of prompts. It is feedback.

Why AI Fits This Particular Gap So Well

Four things line up here, and they line up specifically for writing.

• Language is what these models are built for. Deciding whether a paragraph's structure holds or its vocabulary is precise is a judgment about language, not a calculation. That is the one thing a large language model is genuinely good at.
• It collapses the loop. The write, get feedback, revise cycle goes from days to seconds, at almost no cost. A child can fix a paragraph and see the difference before they lose interest, which is the whole game at this age.
• Volume and consistency. No human will patiently mark three essays a week for one child, every week, without fatigue or a bad mood. Software will, and it applies a far more repeatable standard than a tired adult marking piece by piece. No automated score is exact, but the feedback stays steady, and it treats the fiftieth piece with the same patience as the first.
• It is private and low stakes. Children are sensitive about their writing. An AI coach is an unembarrassing place to fail and try again, which matters more than adults tend to remember.

The Part People Get Backwards

The common worry is reasonable on its face: AI makes mistakes, so do not lean on it for exam preparation. But look at where it makes them. Ask an AI a multi-step mathematical reasoning question and it can hand back a clean, confident, wrong answer, and because maths has exactly one right answer, wrong is simply wrong. Ask it instead to judge whether a paragraph's ending lands or its vocabulary is precise, and it is working in its strongest domain, where there is no single answer to get wrong, only a judgment to offer.

So the counterintuitive rule that falls out of this: trust AI least on the maths answer, and most on the writing feedback, as long as you use it as a fast second reader and not as an oracle. Roughly, the part of study people are most nervous about handing to AI is the part it is best suited to, provided it stays a second reader and not the final judge.

This Is Really About Prompted Writing, Not Just Selective

Nothing in the argument so far is specific to Selective. The Selective writing task is timed prompted writing, and so is NAPLAN writing, the ACER scholarship writing paper, and most school essay tasks. The underlying skills, planning quickly, structuring a piece, holding to a genre, editing under time, are trainable through repetition and feedback.

AI can generate unlimited on-level prompts across narrative, persuasive and informative writing, hold a child to timed conditions, offer a plan or a scaffold to start from, then score the result and give targeted feedback, and power the rewrite. The one step it must never take is the writing. If your child is facing any test that asks them to write to a prompt under time, the same loop applies, and the same tool helps for the same reason. One caveat: each assessment has its own expectations, so a Selective practice score is not a NAPLAN or scholarship score. Carry the loop across tests, not the number.

Where I Would Push Back on My Own Argument

I have made the case for AI on writing, so let me argue against myself, because two of the limits are real.

First, AI is weakest at the exact thing that separates good writing from great. Mechanics, structure and clarity, it judges well. But the jump from a solid essay to a memorable one usually comes from originality, a distinct voice, an idea a marker has not already seen forty times that morning. Those are matters of taste, and AI does not have taste. It can reliably get a child to clean and competent. It cannot reliably tell them when their writing has become genuinely interesting. For that, a child still needs a human reader whose judgment they trust.

Second, any tight feedback loop, used lazily, rewards playing it safe. If a child writes only to satisfy a scorer, human or AI, they drift toward a template: tidy paragraphs, varied sentences, ambitious words, nothing actually risked. Real voice often comes from breaking the pattern, and a tool that patiently rewards the safe version every single time can quietly train that instinct out. The feedback only helps if the child reads it, understands the why, and eventually stops needing it. Leaned on as a crutch, it can slow the very judgment it is meant to build.

How to Use It Without the Traps

• Never let AI write or rewrite the essay. The child must do the writing, and the real test is typed on a computer, to time, and unaided. Sentence-level suggestions and a plan to start from are fine. Whole paragraphs written for them are not. This is our own hard rule at EurekaWrite, and it is the right one for any tool.
• Treat the score as a practice estimate. NSW never returns a raw writing mark, so no tool can predict one. Read a practice score as a single-marker-style estimate with a six-area diagnostic, not an official itemised report. Take a band as give or take one, and pay more attention to the specific feedback than to the total.
• Use it for checkable specifics. AI is most trustworthy on "this word is misspelled, this sentence is a run-on, this paragraph has no topic sentence." Trust it less on "how original is this idea," and keep a human reader for that.
• Keep the practice typed and timed. Since 2025 the Selective test is computer-based, so the response is typed on screen within 30 minutes. Practise the way the test runs: plan quickly on paper or screen, then type to time. Typing fluency and on-screen editing are now part of the skill, and a typed draft flows straight into feedback with nothing to photograph or copy out.

In one glance, the line between using AI and leaning on it:

Good use of AI	Risky use of AI
Work out what the prompt is really asking for	Ask it to write the response
Get feedback on structure, clarity and flow	Paste its improved paragraph in as your own
Spot repeated sentence patterns and weak spots	Chase a higher number for its own sake
Check spelling and punctuation	Treat the score as an official mark
Pick one or two things to fix, then revise it yourself	Rewrite the whole piece in an AI voice
Generate fresh practice prompts	Memorise AI model answers

Privacy and Safety When a Child Uses AI

Because this involves a child's schoolwork, a few rules are worth setting before you start:

• Use a nickname, not the child's real name, and do not enter their school, address, birthday, photo or contact details.
• Keep genuinely personal family material out of practice essays, or change the identifying details.
• Read the tool's privacy terms first, and prefer one built for this that does not reuse the writing. Ours is set out in our Privacy Policy.
• Treat AI feedback as a draft for a person to judge, not an answer to copy. A parent or the student should have the final say on any change.
• Check any school rules for assignments first, since this page is about home practice, not classroom submissions.

What Good AI-Assisted Writing Practice Looks Like

One honest note before the routine. When my son sat the test, the writing was still handwritten. From 2025 the test moved to a computer-based format, so the response is now typed on screen. That change tightens the loop below rather than breaking it, because a typed draft goes straight into feedback with no step in between.

Here is the routine I would run now:

1. Type a response, to time. Pick a prompt and give it the real conditions. For Selective, that is about five minutes to plan (on paper or screen), twenty two to type, three to check.
2. Get it scored, and read the feedback. Look at the area-by-area comments, not just the total. The total is the least useful part.
3. Choose two or three specific fixes, not ten. The high-value ones are usually the same: make one general line specific and picturable, break a run of same-length sentences, and give the ending a real close instead of trailing off.
4. Rewrite one paragraph applying them. You do not have to rewrite the whole piece. Fixing one paragraph well teaches more than redoing all of it badly.
5. Next session, a new prompt, ideally a different genre and form. The task is unknown until the day and might be a story, a persuasive letter, an advice sheet or a report, so rotate across narrative, persuasive and informative writing and the forms within them.

Over a school term that is twenty to thirty pieces with feedback on every one, which almost no child gets any other way. If you want the prompts and the timed conditions ready to go, our writing practice workspace runs this loop, and the genre guides show the specific moves for narrative, persuasive and informative writing.

The Honest Bottom Line

It is worth naming an objection here, because NSW Education makes it directly: the test needs no coaching, and over-coaching can harm a child's wellbeing. I agree with the spirit of that, and the distinction matters. Drilling multiple-choice tricks is coaching. Writing more, reading honest feedback, and revising is just learning to write, which is a skill a child keeps long after the test. Keep it light and low-stakes, and stop well before it stops being useful.

AI does not replace reading widely, which is still where vocabulary and a real sense of good writing come from. It does not replace a good teacher, or the child's own effort, or a parent's read on whether a piece is actually worth reading. What it does is remove the single biggest practical barrier, feedback, for the one component of the four that is starved of it.

Used as a fast, patient second reader, and not as an oracle or a ghostwriter, it was the most useful tool I found in the whole of my son's preparation, for exactly one of the four papers. That is a narrow claim, and I am fairly sure it is a true one. If you want to understand how far to trust a practice score before you rely on it, our accuracy page shows how we test the scorer against human-marked essays, and what a good score really means explains why there is no official number at all.

Common Questions