AI writing-marking checklist How to choose an AI writing-marking tool for your school
writingboost.com.au/choosing-an-ai-marking-tool/
- Vendor
- Product
- Reviewed by
- Date
How to use this checklist
Use one copy for each vendor. Ask every vendor the same questions, and ask for the answers in writing with a link to where each one is published. A clear “not yet” tells you more than a vague yes.
- Bring the people who will check the answers: a teaching lead for the classroom questions, your privacy or IT contact for data and security, and whoever approves the purchase.
- Mark each answer Yes, Partly, No or Not stated, and note where the vendor publishes it.
- Treat “Not stated” as a question to follow up, not as a no.
- Check what you’re told against the vendor’s own documents: privacy policy, service-provider list, terms and accessibility statement.
Student data and privacy
1 Where is student data stored, and where does the AI processing happen?
Storage and processing can happen in different places. A tool can keep student work in Australia and still send it overseas to be marked.
Watch for “Hosted in Australia” that turns out to cover storage only.
Answer YesPartlyNoNot stated
Where it’s published / notes
2 Is student writing used to train AI models?
There are two questions here. Does the vendor train or tune its own models on student work? And can its AI providers use that work for their own models? An opt-out means student work is used until someone switches it off.
Watch for Training switched on by default, or an answer that covers the vendor or its providers but not both.
Answer YesPartlyNoNot stated
Where it’s published / notes
3 Has the product been through a Safer Technologies 4 Schools (ST4S) assessment?
ST4S is a national, standardised assessment of the security and privacy of digital products used in schools, run by Education Services Australia with the school sectors. Some systems ask schools to check a product’s ST4S result before using it. A current assessment can save your reviewer asking the same questions again.
Answer YesPartlyNoNot stated
Where it’s published / notes
Marking you can check
4 What evidence is there that the marks are accurate?
“Aligned to the NAPLAN criteria” describes the rubric, not the accuracy. Agreement with expert markers is a separate claim, and better learning is a different one again.
Watch for One headline percentage with no test set behind it, or evidence from a model that is no longer in use.
Answer YesPartlyNoNot stated
Where it’s published / notes
5 Can teachers check and change the marks before students see them?
AI marking can be wrong. Teachers need to be able to check a mark, change it and decide when students see it.
Answer YesPartlyNoNot stated
Where it’s published / notes
Fit for your classrooms
6 Does it fit the curriculum you teach?
A tool marks against a rubric. Know which one, which genres it covers and which year levels it was built and tested for.
Answer YesPartlyNoNot stated
Where it’s published / notes
7 What do students do with the feedback?
A score tells a student how they went, not what to do next. The learning happens when a student uses the feedback in a new draft.
Answer YesPartlyNoNot stated
Where it’s published / notes
8 Can it mark handwritten work?
Many classes still write on paper. A handwriting reader can misread a word, and a misread word changes what gets marked.
Answer YesPartlyNoNot stated
Where it’s published / notes
9 What do school leaders see?
A leader needs to compare like with like: the same task, the same criteria, the same conditions.
Answer YesPartlyNoNot stated
Where it’s published / notes
10 Is there an accessibility statement?
Students with disability need to use the same tool as their class. A statement tells you what has been checked and how to raise a barrier.
Answer YesPartlyNoNot stated
Where it’s published / notes
Price, trial and governance
11 Is the price public, and what are the trial terms?
A public price lets you budget before the sales conversation. A trial needs terms too: who takes part, for how long, and what happens to student work afterwards.
Answer YesPartlyNoNot stated
Where it’s published / notes
12 How does it line up with the Australian Framework for Generative AI in Schools?
Education ministers approved the Framework in October 2023 to guide the responsible use of generative AI in schools. It has six principles: teaching and learning; human and social wellbeing; transparency; fairness; accountability; and privacy, security and safety.
Watch for A tool described as “not generative AI” whose provider list includes generative AI model providers. Ask which models mark the writing.
The Australian Framework for Generative AI in Schools
Answer YesPartlyNoNot stated
Where it’s published / notes
Compare vendors
Write Yes, Partly, No or Not stated in each box.
| Question | Vendor | Vendor | Vendor |
|---|---|---|---|
| 1 Australian storage and AI processing | |||
| 2 No training on student writing | |||
| 3 ST4S assessment | |||
| 4 Published accuracy evidence | |||
| 5 Teacher control | |||
| 6 NAPLAN criteria, year levels, genres | |||
| 7 A revision loop, not just a score | |||
| 8 Handwritten work | |||
| 9 Leader views | |||
| 10 Accessibility statement | |||
| 11 Public price and trial terms | |||
| 12 Australian Framework for Generative AI in Schools |
Writing Boost · writingboost.com.au/choosing-an-ai-marking-tool/
How Writing Boost answers
Our answers to the same twelve questions, with links to where our site says more. Where an answer is partial, or we don’t publish one yet, we say so.
Ask us for any of these in writing for your school’s review.
From writingboost.com.au. Check each answer against the linked page.
-
1 Where is student data stored, and where does the AI processing happen?
Answered on our site
Student personal information is stored in Australia. AI processing of student writing, marking and feedback takes place in Australia. Student AI runs on Amazon Bedrock and Microsoft Azure, both in Australia. Encrypted backups are stored in Sydney.
Some supporting services sit outside that boundary, and our provider table names each one: teacher-generated stimulus images use a separate service that can process prompts in the United States and other countries, and email, website delivery and school sign-in can involve overseas processing. None of them marks student writing.
-
2 Is student writing used to train AI models?
Answered on our site
No. We do not use student writing or other student information, identified or de-identified, to train or fine-tune AI models. That applies to every school; there is nothing to opt out of.
Our student AI arrangements also prohibit our AI providers from using student information for their own model training, fine-tuning or product improvement, and the AWS AI services opt-out applies to our account.
-
3 Has the product been through a Safer Technologies 4 Schools (ST4S) assessment?
Not on our site yet
Our site doesn’t mention an ST4S assessment. Ask us for our current status.
-
4 What evidence is there that the marks are accurate?
Partly answered
We checked Writing Boost’s marks against 52 officially marked NAPLAN scripts it hadn’t seen before, narrative and persuasive, and found 98% adjacent agreement: criterion marks within one point of the official mark.
A Writing Boost score is not an official NAPLAN result, and class comparisons use the Writing Boost year-level cohort, a platform reference rather than a national benchmark. For the full evaluation, including exact agreement and where the marking disagrees most, ask us.
-
5 Can teachers check and change the marks before students see them?
Answered on our site
Schools choose whether AI features are enabled, and whether students see scores; written feedback and rewriting stay available when scores are hidden. A school admin can let teachers adjust writing marks for the students they teach. Every change needs a reason, and the original marks are kept.
Teachers switch writing tools on or off for each task, and check and correct the transcript of handwritten work before it is marked. Feedback reaches the student as soon as marking is ready, so the next draft can start straight away; teachers read it alongside the writing and can adjust a mark afterwards.
-
6 Does it fit the curriculum you teach?
Answered on our site
Teachers can set six text types: narrative, persuasive, informative, procedure, explanation and poetry. Narrative and persuasive are marked on their official NAPLAN criteria. The other four use adapted NAPLAN criteria, out of 48: the same criteria and scales, with names and descriptors rewritten for the genre. A procedure, for example, is marked on procedural control and on segmentation and layout. Adapted results never produce a NAPLAN proficiency level.
Students can be in any year from Foundation to Year 12. The criteria are the same at every year level, and the feedback is pitched to the student’s year. Scores are not official NAPLAN results or predictions of proficiency levels. Writing Boost is independent and is not affiliated with or endorsed by ACARA.
Prep/Foundation, Year 1 and Years 10–12 are included free with a paid plan, where offered. Paid year levels are Years 2–9.
-
7 What do students do with the feedback?
Answered on our site
Each piece comes back with what works and one thing to work on, in a passage the student can find. Students keep the feedback beside their writing, change a passage in their own words and compare the two drafts. The original piece and its scores stay as they were, and every piece stays on the student’s bookshelf. Students remain the authors: they choose the words and write the next draft.
-
8 Can it mark handwritten work?
Answered on our site
Yes. Students write on paper and give their pages to their teacher. The teacher captures the pages, checks and corrects the editable transcript against the original, then submits the work for marking. The original page stays in view during review. Handwriting goes through the same Australian AI processing as typed work.
-
9 What do school leaders see?
Answered on our site
Every class is marked on the same criteria. A leader can set one task for a whole year level and read its classes side by side, criterion by criterion, with the Writing Boost average for that year level and genre as a reference. That average is a platform reference, not a national benchmark. A task a teacher sets for one class stays in that class’s picture. Teacher and leadership access are included in the school plan, and networks can use System Leadership for shared tasks across schools.
-
10 Is there an accessibility statement?
Answered on our site
Students can change how text looks across the student app: OpenDyslexic or Atkinson Hyperlegible letters, larger text, roomier spacing, and a cream, grey, blue or green page. The writing desk has its own text size, from 12 to 24 point. When writing tools are on, Hear it back reads a selection or the whole piece in the device’s own voice, at a slower, natural or faster pace, highlighting each word where the browser supports it.
Spelling words, and prompts with recorded audio, can be played aloud. Students can write by hand instead of typing. The app follows the device’s reduced-motion setting, and teacher screens keep a minimum text size.
Our accessibility page lists these supports and explains how to report a barrier. We don’t claim WCAG conformance or certification. If your school needs a formal statement or a test report, raise it with us during evaluation.
-
11 Is the price public, and what are the trial terms?
Answered on our site
Our pricing is public. Writing Boost + Spelling Boost in one school plan, from $1,000 per year, excluding GST. Teacher and leadership access and guided onboarding are included. Find your school on the pricing page, choose your year levels and check enrolments to see your annual price. No demo or contact details are required to explore.
A requested proposal is a fixed quote and doesn’t place an order. We don’t publish a standard trial: a trial or pilot is agreed with our team first, including participants, dates, support and data arrangements. If a school doesn’t subscribe, student trial data is deleted or de-identified within 90 days after the trial ends. Renewal, cancellation and enrolment changes follow the accepted agreement.
-
12 How does it line up with the Australian Framework for Generative AI in Schools?
Partly answered
We haven’t published a mapping to the Australian Framework for Generative AI in Schools yet. Here is how the marking is built.
Writing Boost uses a hybrid marker: deterministic code and generative AI, each doing the part it does best. Code that gives the same result every time handles what should never vary. On the NAPLAN criteria, spelling is scored by an algorithm, not by the AI model. Likely personal details, such as full names, email addresses and phone numbers, are swapped for placeholders before a piece is marked. The student’s writing reaches the model as marked-out material to assess, not as instructions. Code checks the AI’s feedback before it is saved: an example has to quote the student’s own words, and a spelling correction has to match an error the algorithm found.
A generative AI model, running in Australia, reads the writing against the criteria and writes the feedback. Teachers keep the final say.
