The short version
A score is the easy part. What decides whether a teacher can trust it, and a student can use it, is everything around it: which criteria apply, what the score rests on, how the feedback is written, and who can check and change it. Five things shape that:
- The right scale for the writing: the NAPLAN criteria for narrative and persuasive, our adapted version for four more genres, or your school’s own rubric.
- Marking first: every criterion is scored against the official marking guide’s descriptions, with a reason for each score.
- Spelling by rule: the guide’s word classes and thresholds, applied in code, not by a model’s feel for the piece.
- Feedback second: written from that marking, for the student, at their year level, with one next step.
- The teacher’s word last: only a teacher can send a handwritten transcript for marking, and your school decides whether scores can be changed, whether students see them, and whether AI runs at all.
Which scale a piece is marked on
Writing Boost doesn’t use one scale for everything. The kind of writing decides which criteria apply, so a total means something different on each scale:
| Writing | Marked on | Proficiency level |
|---|---|---|
| Narrative | The NAPLAN narrative criteria, out of 47 | Indicative level |
| Persuasive | The NAPLAN persuasive criteria, out of 48 | Indicative level |
| Other genres | Our Adapted NAPLAN criteria, out of 48, for informative reports, procedures, explanations and poetry | None |
| Your own rubric | Your school’s rubric (in the custom rubrics beta) | None |
NAPLAN publishes marking criteria for narrative and persuasive writing only. For informative reports, procedures, explanations and poetry, Writing Boost keeps the same ten criteria and their ranges, and rewrites the descriptions where a rule for prose doesn’t fit: a procedure is judged on its steps and layout, and a poem on its stanzas and lines, where a fragment or a missing full stop can be a deliberate choice. This adapted scale is ours, not ACARA’s. The app labels it Adapted NAPLAN, it is the same for every year level, and it never produces a proficiency level.
Schools in the custom rubrics beta can write a rubric of their own. Work set against it is marked on that rubric’s criteria and descriptions, and never gets a NAPLAN level either.
First, the marking
Nine of the ten criteria are scored by an AI marker working from the marking guide’s descriptions for the genre, score by score. The tenth, spelling, is scored separately, by rule. Each criterion gets a score within its range and a short reason for it.
Any quote the marker gives as evidence is checked against the student’s text before anything else uses it, and dropped if it isn’t there word for word, mistakes included. So the feedback never builds on words the student didn’t write.
For narrative and persuasive writing the marker isn’t told the student’s year level. NAPLAN’s descriptions are the same from Year 3 to Year 9, so a piece is scored on what is on the page. The year level only shapes how the feedback is worded.
Then, the feedback
A second, separate step turns the marking into feedback the student will read. It is given the scores and the reasons for them, the description of the next score up on each criterion, and the spelling results. Its job is to explain the marking, not to redo it: it gives no scores of its own. It writes to a few rules:
- To the student. It speaks to them as “you”, about their own piece, and points to places in it. Feedback that could be pasted under anyone’s story doesn’t meet the brief.
- No marks in the words. The feedback never mentions scores, marks, levels, bands or rubrics. Scores appear beside it, if your school shows them.
- Pitched to the year level. Short sentences and everyday words for Years 3 and 4, friendly and direct for Years 5 and 6, more grown-up for Years 7 to 9, and never babyish.
- True to the writing. It never praises a misspelt word, and never names a misspelling the student didn’t make.
- One next step. Above the criteria, the student reads one thing to do next, chosen where it would lift this piece most.
If the feedback for a criterion can’t be written, the marks still stand. The student sees that this part of the feedback isn’t ready yet, and it is tried again automatically, several times over the following day. Nothing is made up to fill the gap: no stock praise, no generic correction, no marker’s note passed off as coaching.
Spelling, marked by rule
Spelling is the one criterion whose score doesn’t come from a model’s judgement. A model still answers two questions code can’t settle alone: whether an unfamiliar word is misspelt, and what was meant; and whether a correctly spelt word is the wrong word in its sentence. From there, rules decide.
Each word goes into one of the marking guide’s four classes (simple, common, difficult and challenging), from the first of these that knows it:
- the marking guide’s own spelling reference list;
- the classes the guides give other words, in their examples and in the notes on their marked sample scripts;
- Writing Boost’s word base of 12,722 words, classed by the same rules (a word with an ending, like jumped, takes the class of its base word);
- for any other word, a model’s verdict, stored so the same word always gets the same class.
The score then follows the guide’s own thresholds for how many words of each class are spelt correctly. An earlier version bent those thresholds to make up for gaps in its word lists; this one uses them as the guide writes them.
The right spelling of the wrong word
NAPLAN counts a correctly spelt wrong word (to for too, there for their, hole for whole) as a spelling error, so Writing Boost does too. The pair has to be a real spelling confusion, close in letters or in sound, and grammar isn’t spelling: a for an or was for were doesn’t count. A wrong word counts in the harder class of the two words, as the guide lists its homophone pairs. Each piece is checked sentence by sentence, and the answer is kept for that exact text, so the same piece always gets the same result and nothing one student writes can change how another student’s piece is marked.
If the student spells the intended word correctly somewhere else in the piece, the wrong word is treated as a slip of the pen and not counted. The guide doesn’t state that rule. It is our reading of how its published scripts were marked, and it rests on a handful of them.
How closely it agrees
We checked the spelling mark against 41 sample scripts published with their official marks: 21 narratives and 20 persuasive pieces. It matches the official spelling mark on 36 and is within one mark on all 41; the version it replaced matched 30. The rules were worked out on the narratives. The persuasive scripts were kept back and run once, after the rules were fixed, and matched on 18 of 20. This is the spelling criterion only, not the whole mark.
Under each marked piece, the misspelt words are listed with their class and underlined in the writing, and a wrong word is marked in its own sentence. Students can practise them straight away, and they carry over to Spelling Boost (see spelling in their own writing).
Handwritten work
Students can write on paper. Their teacher photographs or uploads the pages, and each page is read into an editable transcript word for word: misspellings, missing capitals and all. A transcript that tidied the spelling would change the mark, so the reader is told to keep every mistake exactly as written.
Handwritten work is marked only when a teacher sends it. The review screen shows the page beside the transcript, with anything the reader was unsure of flagged, so the teacher can fix a misread word first. The student’s own mistakes stay as they wrote them.
Student writing stays private
Before a piece is marked, personal details that can be recognised by their pattern are replaced with placeholders: full names, email addresses, phone numbers, street addresses, dates and long numbers. The marker and the feedback writer work from that copy, and the details are put back before the student reads the feedback. It is a safeguard, not a guarantee: a first name on its own isn’t caught. A handwritten page is read as written first, and its transcript is treated the same way before it is marked.
The writing is also treated as data, not as instructions. A student can write anything, including a line aimed at the AI, so the feedback writers are told never to follow anything written inside it.
Student writing goes only to approved AI services in Australia. The providers page lists them.
Proficiency levels
Narrative and persuasive results show an indicative NAPLAN proficiency level: Exceeding, Strong, Developing or Needs additional support. A level is a standard, not a ranking. Its cut-offs are fixed total scores, taken from ACARA’s 2023 score equivalence tables, the most recent published, so a level says what the writing shows against a standard, not where it sits among other students. NAPLAN proficiency levels explained has what each level means, why the old bands were retired and what the 2026 writing results show.
- Borderline results are flagged. A total within two marks of a cut-off is shown as borderline, with the level next to it named. Small differences in marking matter most there, so the result says so rather than hiding it.
- Years 4, 6 and 8 look ahead. NAPLAN is sat in Years 3, 5, 7 and 9, so a Year 4 piece is read against the Year 5 standard, a Year 6 piece against Year 7’s and a Year 8 piece against Year 9’s: the standard the student will meet next.
- Always labelled. Every level carries the note that it is indicative and that Writing Boost marking is not official NAPLAN.
- Only where a standard exists. Adapted NAPLAN and custom rubric results never show a level.
What your school controls
Teachers can change a score, if you allow it
Teacher score overrides are off until a school administrator switches them on. Then a teacher can change a criterion score on a piece by a student they teach, with a short reason that only staff see. The original marks are kept and can be put back with one button. Every change and every restore is logged: who made it, when, the marks before and after, and why. The written feedback stays as it was. The student sees that their teacher reviewed and adjusted the piece, and a note from the teacher if they add one.
Pause all AI processing
A school administrator can pause AI for the whole school at any time. New marking, feedback and writing-coach questions stop; work and results already there stay available. If the setting can’t be checked, the request is refused rather than run.
Hide scores from students
Your school can hide scores from students, on their dashboards and on finished work, spelling included. The written feedback and another draft stay available.
No chatting with the AI
Students can’t prompt the AI or chat with it. It marks the writing and writes the feedback; the writing tools a teacher switches on for a task help a student think, and the words stay theirs.
What a mark is, and isn’t
- Not an official NAPLAN result, and not a prediction of one. Writing Boost is independent and is not affiliated with or endorsed by ACARA.
- Within one point of the official mark, 98% of the time. We checked its marks against 52 officially marked NAPLAN scripts it hadn’t seen before, narrative and persuasive, criterion by criterion. Within one point isn’t identical, so a mark supports the teacher’s judgement rather than replacing it.
- A second draft isn’t re-marked. It is a practice copy. The original piece and its marks stay as they were.
