AI Marking Platform Comparison: A Buyer's Guide for Hong Kong Schools

Last updated: 2026-07-23

Under the AI for Empowering Learning and Teaching Funding Programme, many schools have put AI marking near the top of their shopping list. The reason is easy to see. Marking takes up more teacher time than almost anything else, and the improvement is easy to measure. The catch is that products labelled 'AI marking' vary enormously. This article gives you a way to compare them, not a ranking of individual products. A disclosure: Criteva develops a local marking platform, Criteva Marking, and the criteria below apply just as much to our own product.

Why schools need to compare AI marking platforms carefully now

EDB Circular Memorandum 221/2025 asks participating schools to bring AI-assisted teaching into at least three subjects and to develop at least six teaching demonstration examples. AI-assisted marking is one of the most common starting points. It deals directly with a teacher's marking load, the improvement is easy to quantify, and it fits neatly into the 2027 interim report.

Under the same label, though, actual capability varies widely. Some products simply hand an essay to a general chatbot and print out the comments. Others have the official marking scheme built in and attach the student's own words, mark by mark, as evidence. Choose the wrong one and teachers end up correcting bad scores one by one during review, which is slower than marking by hand. That is why a little time spent comparing before you buy is worth it.

Five criteria for comparing AI marking tools

1. Is the marking scheme localised?

Hong Kong public exams have their own marking culture. The marking references and analytic rubrics for each HKDSE subject are not the same as those in international curricula. A tool trained on generic marking standards can award marks quite differently from a local examiner. When you compare, ask directly: which marking scheme is built into the system, and has it been calibrated against past Hong Kong papers?

2. Is every mark backed by evidence?

A reliable AI marking system should score point by point, and every mark should point back to where it came from in the student's answer. Teachers cannot review a mark that has no evidence behind it, and they cannot explain it to students or parents either. During a trial, pull a few scripts and check whether the system can clearly show which sentence in the answer earned each mark.

3. Does the teacher keep the final say?

AI marking belongs in the role of a first pass, and the final decision has to rest with the teacher. The system should let teachers review and change every score easily, rather than printing a total that cannot be adjusted. This also touches on the professional responsibility owed to the EDB. The assessment data in the report carries the teacher's name in the end.

4. How is student data handled?

Student scripts are personal data. You need clear answers on where the data is stored, how long it is kept, whether it is used to train models, and whether the arrangement meets the Personal Data (Privacy) Ordinance. Our article on the school AI tool privacy checklist has a fuller list.

5. Total cost and pricing model

Annual fees, per-student pricing, per-subject pricing: work all three out as a three-year total cost against your school's actual size. The funding can be used until 31 August 2028, so the sensible unit of comparison is the three-year total, not a single year's list price. Also confirm whether the trial period is free, whether adding subjects or classes costs extra, and whether you can export your data once the term ends.

Types of AI marking solution in the Hong Kong market

The solutions on the market fall into roughly three groups. The first is a general AI tool, such as a chatbot, paired with instructions the teacher writes. It costs the least, but scoring is hard to keep consistent, and student data compliance is hard to handle. The second is an international marking platform. The features are mature, but the marking schemes are mostly built around overseas curricula and may not match the analytic requirements of the HKDSE. The third is a local platform, designed around the Hong Kong curriculum and official marking schemes. These usually have a clear edge in matching the schemes and in Chinese support, and when you buy one, the thing to examine is how well it delivers on the five criteria above.

Whichever group you choose, the EDB reminded schools in its clarification of February 2026 to assess carefully whether packaged solutions suit the school's own needs. A sensible approach is to trial on a small scale, compare the evidence, and only then decide on a whole-school purchase.

Questions to ask suppliers when you trial an AI marking platform

Where does the built-in marking scheme come from, and which papers was it calibrated against?
Can you show the mark-by-mark evidence for a single script?
How many steps does it take a teacher to change an AI score, and is the change history kept?
Which region's servers store student scripts, how long are they kept, and are they used to train models?
Based on our student numbers, what is the three-year total cost, and which subjects does it cover?
Can one subject panel trial it free for a term before we make a whole-school decision?

Common questions

Can AI marking be bought with the AI for Empowering Learning and Teaching funding?

Yes. Circular Memorandum 221/2025 lets schools use the funding to buy the software and services needed to run AI-assisted teaching, and AI-assisted marking is a common use. Schools should make sure the purchase follows established procurement procedures and keep receipts and evidence of use for the report.

Is AI scoring accurate? Can teachers rely on it?

It depends on how the system is designed. With official marking schemes built in and evidence attached to each mark, teachers can review quickly and correct individual scores. Black-box scoring with no evidence cannot be reviewed. The right expectation is that AI does the first pass and the teacher makes the final decision, rather than replacing the teacher's judgement altogether.

Do primary schools need an AI marking platform too?

Marking needs in primary schools centre on Chinese and English writing and on checking the steps of maths work, while most mature products on the market are built around public exams. A primary school is better off trialling in a writing subject on a small scale, confirming that it genuinely saves teachers time, and then considering whether to expand.

Sources

Want to know your school's next step?

Tell us where your school is with the funding programme. We help with evidence, from choosing tools to filing the reports.

Book a free school consultation Request a proposal