Key Takeaways
- EssayHero assesses HKDSE English Paper 2 essays across three criteria — Content, Language and Style, Organisation — each scored 1-7, totalling 3-21
- Our scoring is aligned with HKEAA marking standards and calibrated against officially graded exemplar essays
- The AI applies genre-specific criteria for all ten HKDSE text types (speeches, letters, articles, reports, and more)
- This is a practice feedback tool, not an exam score predictor
With 61 Days Until the Exam, Your Students Need More Feedback
The 2026 HKDSE is approaching. Your students are writing practice essays. The question is whether they are getting enough feedback on those essays to actually improve before April.
If you are a teacher or tutor reading this, you already know the answer. You are marking as fast as you can, but there are only so many essays you can give detailed criterion-level feedback on in a week.
Your students need to write more. They need feedback faster. And they need it linked to the specific criteria that examiners use.
Why EssayHero Exists
I built EssayHero originally for my own students in Hong Kong. It started as a tool to give them practice feedback between my own marking rounds — something better than nothing while they waited for the next batch of corrected essays.
It has since grown, but the core purpose has not changed: faster feedback cycles for students who are practising.
What This Post Covers
This post explains exactly how the HKDSE assessment works, what the AI looks for, and where it falls short. If you are going to recommend it to your students, you deserve to know all of this.
The Three Criteria
Every HKDSE Paper 2 essay is assessed against three criteria, each scored on a scale of 1 to 7.
Content: Relevance and Task Fulfilment
Content evaluates relevance, detail, creativity, reader engagement, and fulfilment of task requirements.
At the top of the scale, the AI looks for writing that is relevant and extensive, shows awareness of purpose, engages the reader's interest, and demonstrates creativity and imagination where appropriate. At the lower end, it looks for whether the student has made at least a few relevant content points.
Language and Style: Accuracy and Sophistication
Language and Style covers sentence structures, punctuation, grammar, vocabulary, and register appropriateness.
A score of 5 or above requires a wide range of sentence structures used accurately, with appropriate vocabulary that includes some ambitious and sophisticated language. The register, tone, and style must be appropriate to the text type. A score of 2 reflects simple sentences and basic vocabulary with sufficient accuracy to be comprehensible.
Organisation: Structure and Coherence
Organisation assesses structure coherence, paragraphing effectiveness, and cohesion between sentences and paragraphs.
Strong organisation means wholly coherent structure appropriate to the genre, effective paragraphing, and sophisticated cohesion. Weaker organisation shows basic paragraph awareness with simple links between sentences.
Total Score Range
The total is the sum of the three criterion scores, giving a range of 3 to 21.
How the 1-7 Scale Works
The official HKEAA Level Descriptors use a 1-5 framework. Our 1-7 scale extends this to provide more granularity at the top end.
Score Mapping to Official HKEAA Levels
| Our Score | HKEAA Level | Description |
|---|---|---|
| 6-7 | Beyond Level 5 | Outstanding or exceptional quality (deliberately rare) |
| 5 | Level 5 | Strong Level 5 work |
| 4 | Level 4 | Competent work with room for improvement |
| 3 | Level 3 | Developing competence |
| 2 | Level 2 | Basic achievement |
| 1 | Level 1 | Limited achievement |
This mapping means that a student scoring 5 on our system is producing work that aligns with the highest official HKEAA level descriptor. Scores of 6 and 7 are stretch targets — they indicate writing that goes beyond what the descriptors require, approaching native-speaker fluency and publishable quality.
Why We Extended the Scale
We made this design decision deliberately. The 1-5 official scale compresses strong performance into a single level.
Extending to 7 gives students at the top end more room to see improvement and gives teachers a more nuanced picture of where their strongest students stand.
Text Types Matter
HKDSE Paper 2 is distinctive because it requires students to write in specific text types. Each text type has its own conventions, and a student who writes a technically competent essay in the wrong genre will lose marks.
The Ten HKDSE Text Types
EssayHero assesses all official text types:
- Articles
- Blog entries
- Emails
- Formal letters
- Informal letters
- Proposals
- Reports
- Reviews
- Short stories
- Speeches
Genre-Specific Assessment
EssayHero applies genre-specific assessment criteria on top of the three core criteria.
For a speech, the AI checks for:
- Audience engagement techniques
- Oral markers ("Firstly," "Let me now turn to")
- Rhetorical techniques
- Memorable conclusion
For a formal letter, it checks for:
- Format conventions
- Formal register throughout
- Clear purpose statement
For a short story, it evaluates:
- Narrative arc
- Characterisation
- Dialogue quality
- Descriptive language
Genre Mismatch Detection
If a student selects "Article" but writes something that reads like a speech, the AI flags the mismatch and assesses based on what was actually written. This helps students understand that choosing the right text type is not just a formality — it shapes the entire assessment.
Calibration Against Official Standards
We have calibrated our scoring against HKEAA-graded exemplar essays using Google Gemini 3 Flash Preview. Our current Level QWK is 0.833 (almost perfect agreement) across 119 exemplar essays, with a within-one rate of 98.3% and near-zero bias (-0.08). For full results, see How We Validate Our Scores.
Our Calibration Process
Calibration is ongoing and involves multiple checks:
- Score distribution analysis — Ensuring the AI's score patterns match expected ranges
- Internal consistency checks — Verifying that criteria scores align logically
- Teacher feedback integration — Incorporating comparisons from educators who use both systems
- Pattern adjustment — Correcting tendencies toward over-generous or over-harsh marking
What Calibration Means
Calibration does not mean our scores are identical to what an HKEAA examiner would give. It means they are broadly aligned. An AI score of 5 on Content indicates work that is in the right neighbourhood of strong performance, not a guarantee of a specific examination result.
The Feedback Is Criterion-Linked
Every piece of feedback the AI produces names the specific criterion it relates to. Instead of generic comments like "good vocabulary," the student sees feedback tied to Language and Style, Content, or Organisation by name.
This trains students to think in the same terms that examiners use.
Adaptive Feedback Depth
The depth of feedback adapts to the student's score level:
| Score Range | Level | Feedback Approach |
|---|---|---|
| 1-2 | Critical | Full worked examples with before-and-after rewrites, limited to 2-3 critical issues per paragraph to avoid overwhelming struggling students |
| 3-4 | Developing | One worked example per issue type, with specific practice strategies |
| 5 | Competent | Brief acknowledgement of competence, plus one targeted suggestion for reaching the next level |
| 6-7 | Strong | Genuine, specific praise with at most one optional stretch goal (AI is explicitly instructed not to nitpick) |
This adaptive approach means a student who scores 2 on Organisation gets a fundamentally different feedback experience than one who scores 6. The struggling student gets scaffolding. The strong student gets validation and direction.
What We Cannot Do
This is the section that matters most, and I will be direct.
Cannot Replicate Holistic Examiner Judgement
Experienced HKDSE examiners develop a sense of quality that goes beyond individual criterion descriptors. They read thousands of essays and develop calibrated intuitions that an AI does not have.
Our AI applies criteria systematically and consistently, but it lacks the professional judgement that comes from years of marking experience.
Cannot Predict Exam Scores
If your student scores 16/21 on EssayHero, that does not mean they will score 16/21 in April.
The AI does not know:
- The specific question paper
- The marking team's calibration discussions
- The year's grade boundaries
Our scores are useful for tracking improvement over time, not as exam predictions.
Cannot Fully Assess Creative Writing
For short stories and other creative text types, the AI can evaluate structure, language, and technique. But the subjective appreciation of originality, emotional resonance, and creative risk-taking is inherently limited.
A human reader who is moved by a story will credit it in ways the AI cannot.
Cannot Assess Handwriting or Presentation
This is a digital tool. Handwriting quality, neatness, and physical presentation are outside scope.
Cannot Replace Your Feedback
You know your students. You know what they have been working on, where they were last month, and what specific weaknesses they need to address.
The AI provides standardised, criterion-based feedback. Your feedback provides context, relationship, and professional judgement.
Complementary, Not Replacement
Both AI and teacher feedback are valuable. Neither replaces the other.
What It Is Good For
After that list of limitations, the honest question is: what is the point?
More Practice Cycles
The point is more practice cycles. A student who writes a practice essay on Tuesday evening can paste it into EssayHero, get paragraph-by-paragraph feedback linked to Content, Language and Style, and Organisation, identify the weakest areas, revise, and bring a better draft to class on Wednesday.
That revision cycle is where learning happens, and most students do not get enough of it because feedback is scarce.
Text-Type-Specific Guidance
The text-type-specific feedback helps students understand genre conventions. Many students lose marks not because their English is weak but because they do not understand what makes a good speech different from a good article.
The genre-specific criteria make these distinctions explicit.
Pinpoint Weaknesses
The paragraph-by-paragraph format means students can identify exactly where their essay weakens. Not just "your organisation could be better" but "paragraph three lacks a clear topic sentence, and the transition from paragraph two is abrupt."
Full Transparency
The complete assessment criteria, level descriptors, text-type-specific criteria, and the full prompt text that the AI receives are published on the HKDSE methodology page. Every section of the prompt is available for inspection.
EssayHero is open source under AGPL-3.0.
Independence from HKEAA
The assessment criteria used in our prompts are our interpretation and operationalisation of the HKDSE marking standards. They are not a reproduction of HKEAA's proprietary marking schemes.
Privacy and Data Handling
Essays are processed and discarded. They are not stored (unless the student opts in by creating an account), not used for model training, and not accessible to anyone after the feedback is generated.
Try It With Your Students
EssayHero is free. No account required.
How to Test It
If you want to see what the feedback looks like on an HKDSE Paper 2 essay:
- Go to essayhero.app/?exam=hkdse-paper2-partb
- Paste a sample essay
- Read the output
The "Try a Sample" button will load a demo essay if you want a quick look.
Share Your Feedback
If you think the feedback could help your students practise more effectively between assignments, share it with them. If you think the criteria do not align with HKEAA standards, or the feedback is not useful, I would genuinely like to hear why.
Email hello@essayhero.app.
The Value Proposition
With 61 days to go, every practice essay that gets meaningful feedback is an opportunity for improvement.
EssayHero does not replace your marking. But it might mean your students arrive at your desk with drafts that have already addressed the structural and language weaknesses they could have caught themselves.
EssayHero is free, has no commercial aims, and is built by a Hong Kong teacher for Hong Kong students. Questions or feedback? Email hello@essayhero.app.
Related Articles
How EssayHero Marks Your Essays
A Hong Kong teacher explains the thinking behind EssayHero's marking system and when to trust it
Read moreHow We Validate Our Scores
Our methodology for testing AI scoring accuracy against 120 official HKEAA exemplar essays, with full results and limitations.
Read moreHow We Validate Our Scores
Our methodology for testing AI scoring accuracy against 119 official HKEAA exemplar essays, with full results and limitations.
Read more