If You Teach IELTS, You Already Know the Problem

Your students need practice. They need to write essays, get feedback, revise, and write again. The revision cycle is where improvement happens — not in the first draft, but in the gap between the first and second. Between the second and third.

The problem is that feedback is scarce. You cannot mark thirty Task 2 essays every week. Your students cannot wait five days between drafts for your comments.

And the generic "grammar checker" tools they find online do not understand the difference between Band 5 cohesion and Band 7 cohesion. They do not know what an overview is or why it matters for Task 1. They certainly cannot tell the difference between a position that is "discernible" (Band 4) and one that is "clear and well-developed" (Band 8).

EssayHero attempts to fill that gap. Not replace you — fill the gap between your feedback sessions. This post explains exactly how it works, so you can decide whether it is worth recommending to your students.

The Four Criteria

IELTS Writing is assessed across four criteria, each weighted equally at 25% of the total.

Task 2 Criteria

1. Task Response — Does the essay address the prompt? Is there a clear position? Are ideas relevant, extended, and supported?
2. Coherence & Cohesion — Is the essay logically organised? Are cohesive devices used effectively? Is paragraphing appropriate?
3. Lexical Resource — Is the vocabulary range sufficient? Are there less common or idiomatic items? Are word choices accurate?
4. Grammatical Range & Accuracy — Is there variety in sentence structures? Are complex structures accurate? Are errors frequent or rare?

Task 1 Criteria

For Task 1, the first criterion changes:

1. Task Achievement — Does the response cover key features of the visual data? Is there a clear overview? Is data reported accurately?

The other three criteria (Coherence & Cohesion, Lexical Resource, Grammatical Range & Accuracy) are shared across both tasks, though the descriptors differ in detail.

How Scores Are Calculated

EssayHero scores each criterion independently on the 0-9 scale, then calculates the overall band as the average of all four, rounded to the nearest 0.5. This is the same method used in the official IELTS exam.

How the AI Applies the Criteria

The AI receives a detailed system prompt containing the full band descriptors for each criterion at every band level from 0 to 9. These are aligned with the publicly available IELTS band descriptors. The prompt instructs the AI to match the candidate's writing against these descriptors and assign a score that fits.

Not Keyword-Counting

This is not a keyword-counting exercise. The AI reads the full essay, assesses each criterion independently, and provides paragraph-by-paragraph feedback that names the specific criterion being discussed.

When it says "your Coherence & Cohesion benefits from effective use of referencing here," it is drawing on the Band 7 descriptor for that criterion. When it says "to improve your Task Response, develop this supporting idea with a specific example," it is identifying a gap against the Band 6 or 7 descriptor.

Adaptive Feedback by Level

The feedback adapts to the candidate's level:

• Band 3 response — Full worked examples and before-and-after rewrites. The candidate needs scaffolding, not a critique.
• Band 7 response — Genuine praise for what works well, with perhaps one optional suggestion for reaching Band 8. The AI is instructed not to nitpick strong writing.

Task 1 vs Task 2: What Changes

EssayHero uses separate prompt configurations for Task 1 and Task 2. This matters because the two tasks test different skills.

Task 2: Argumentative or Discursive Essays

Task 2 is an argumentative or discursive essay. The AI looks for:

• A clear position
• Well-developed body paragraphs with supporting evidence
• Logical progression
• A conclusion that synthesises the argument
• Minimum word count: 250

The first criterion is Task Response: does the essay address all parts of the prompt?

Task 1: Data Description

Task 1 is a data description task. The AI looks for:

• An overview of the main trends or features (critical for Band 7+ in Task Achievement)
• Accurate data reporting
• Meaningful comparisons
• Appropriate use of Task 1-specific vocabulary (trend language, comparison language, approximation)
• Minimum word count: 150

Why the Distinction Matters

The distinction matters. A student who writes a beautifully argued opinion piece in response to a Task 1 prompt will score poorly on Task Achievement, even if the writing quality is high. The AI is configured to catch this.

Similarly, a Task 1 response that lists every data point without identifying the key trends will be flagged — the Band 5 descriptor specifically mentions "mechanical" recounting without highlighting.

What the Band Descriptors Actually Say

Each band level is defined by a set of positive features (what the writing demonstrates) and negative features (what limits the score).

Examples of Band Descriptors

At Band 9, you see phrases like:

• "the message can be followed effortlessly"
• "minor errors are extremely rare"

At Band 5, you see:

• "the range of structures is limited and rather repetitive"
• "grammatical errors may be frequent"

How the AI Matches Writing to Descriptors

The AI matches writing against these descriptors:

• A response where "complex sentences are attempted, but they tend to be faulty, and the greatest accuracy is achieved on simple sentences" fits the Band 5 descriptor for Grammatical Range & Accuracy
• A response where "a variety of complex structures is used with some flexibility and accuracy" fits Band 7

The Key Principle: Full Fit Required

The key principle is that a script must fully fit the positive features of a descriptor to receive that band. Negative features limit the rating.

The AI is calibrated to follow this principle: it does not award Band 7 for Coherence & Cohesion simply because the essay is "reasonably well organised." It checks whether information and ideas are "logically organised" with "a clear progression throughout" and whether "paragraphing is generally used effectively."

These are specific requirements, not vague impressions.

The Revision Cycle

The real value is not in the score itself but in what happens after.

A candidate who receives a Band 5.5 and reads that their Task Response lacks supporting examples can revise the essay, add specific examples, and resubmit. If the revised version scores Band 6 or 6.5, they can see precisely which criterion improved and why.

Where Improvement Actually Happens

This iteration is where IELTS preparation actually works. You know this from your teaching:

• The student who writes one essay per week improves more slowly
• The student who writes a draft, gets feedback, revises, and writes again improves faster

EssayHero makes that cycle faster by providing structured feedback at any time, not just during your office hours.

Strictness Modes

The strictness modes help calibrate expectations:

What We Cannot Do

This is the section that matters most.

We Cannot Replicate IELTS Examiner Training

Official IELTS examiners go through rigorous certification by the British Council, IDP, or Cambridge. They are regularly re-certified and their marking is monitored for consistency.

The AI applies the published band descriptors but has not been through this standardisation process. It is applying criteria, not replicating the full examiner experience.

Scores Are Indicative, Not Official

We Cannot See Task 1 Visuals

For Task 1, the AI assesses the quality of the writing, the structure, and the language. But it cannot see the original chart, graph, or diagram.

It cannot verify whether the candidate has read the data correctly. If a student describes an upward trend that was actually downward, the AI may not catch it. Data accuracy checking relies on what the candidate writes, not on the original source.

We Cannot Assess Beyond Writing

IELTS has four components: Listening, Reading, Writing, and Speaking. EssayHero only covers Writing. An overall IELTS band score depends on all four.

AI Scoring Has Inherent Limitations

Large language models are pattern matchers, not examiners. They handle common essay types well but may be less reliable with unusual topics, unconventional structures, or highly idiomatic writing.

For borderline cases where a human examiner would deliberate, the AI applies the closest matching descriptor without that deliberation.

Full Transparency

The complete band descriptors, the detailed prompt instructions, the feedback rules, and the scoring calibration guidelines are all published on the IELTS methodology page. Everything the AI sees when it assesses an essay is visible there.

The source code is open under AGPL-3.0.

Privacy Commitment

Essays are processed and discarded. They are not stored unless the candidate opts in by creating an account. They are not used for AI training.

Privacy matters, especially when students are sharing their writing.

Try It Yourself

EssayHero is free. No account required. If you want to see what the feedback looks like on an IELTS essay, go to essayhero.app/?exam=ielts-task2, paste a practice essay, and read the output. The feedback names each criterion, explains why the score was given, and suggests specific improvements.

If the feedback helps your students iterate faster between drafts, share it with them. If you think the scoring does not align with your experience, or the feedback misses something important, I would genuinely like to hear about it. Email hello@essayhero.app.

I built this for students who need more feedback than they are currently getting. If it helps your students practise more effectively, I am glad. If not, I understand.

EssayHero is free, has no commercial aims, and is built by a Hong Kong teacher for students worldwide. Questions? Email hello@essayhero.app.