How Our AI Scoring Works
We believe you deserve to know exactly how your essay is scored. This page explains our methodology, what the AI evaluates, known limitations, and how we keep improving.
Overview
When you submit an essay on any of our 993 practice exercises, an AI model scores it across the same four criteria that IELTS examiners use. You get a band estimate for each criterion, an overall band, and one specific suggestion to raise your score.
The free scorer uses DeepSeek, a large language model fine-tuned for structured evaluation. Paid subscribers get a second, deeper analysis from Gemini, including paragraph-level feedback, sentence rewrites, and a model answer.
The 4 Scoring Criteria
Every {ielts} {writingTask2} essay is scored on four criteria, each weighted equally. Our AI evaluates each one independently, then averages them (rounded to the nearest 0.5) for the overall band.
Task Achievement (TA)
Does the essay address all parts of the question? Is there a clear position maintained throughout? Are ideas developed and supported with examples?
Coherence & Cohesion (CC)
Is the essay logically organised? Do ideas flow naturally between sentences and paragraphs? Are linking words used appropriately — not mechanically?
Lexical Resource (LR)
Does the writer use a wide enough range of vocabulary? Are words used precisely and naturally, with some less common items? Are spelling errors rare?
Grammatical Range & Accuracy (GR)
Does the essay use a mix of simple and complex sentence structures? Are errors infrequent enough that they don't impede communication?
How Calibration Works
The AI is calibrated against the official IELTS band descriptors published by the British Council, IDP, and Cambridge. We use explicit scoring anchors:
Band 7.0+: Clear position, well-organised, wide vocabulary range, good grammar control. The AI must quote evidence from the essay for any score above 7.0.
Band 6.0–6.5: Addresses all parts, coherent progression, adequate vocabulary with some errors, mix of simple and complex sentences.
Band 5.0–5.5: Partial task response, limited vocabulary, limited grammatical structures.
Below 5.0: Minimal response with significant limitations across all criteria.
Scores are given in 0.5 band increments only — the same increment scale used in the real exam.
Scoring Rules
These rules ensure the AI stays consistent:
- 1
Essays under 250 words are penalised in Task Achievement only (minus 1.0 band), matching the real exam penalty.
- 2
An essay with a clear position, two supported body paragraphs, and some complex sentences scores at least 6.0 — the AI cannot give lower if these elements are present.
- 3
The overall band is the arithmetic mean of four criteria, rounded to the nearest 0.5.
Free vs Paid Scoring
Both tiers use the same criteria. The difference is depth:
Free
- ✓Band scores for all 4 criteria
- ✓Overall band estimate
- ✓One specific improvement suggestion
- ✓Available on all 993 exercises — no signup required
Paid
- ✓Everything in free, plus:
- ✓Paragraph-level strengths and weaknesses with quoted evidence
- ✓Concrete before-and-after sentence corrections
- ✓Band 9 model answer for the same prompt
- ✓Top 3 study priorities ranked by impact
Known Limitations
We are transparent about what AI scoring cannot do:
Not an official IELTS score
Our scores are estimates. Only an official IELTS test centre can give you a real band score. Use our estimates for practice and progress tracking.
Handwriting not assessed
The AI reads typed text. On the paper-based exam, illegible handwriting can lower your score — the AI cannot simulate this.
Accuracy range
In our internal testing, AI scores typically fall within 0.5 bands of where a human examiner would score the same essay. Occasionally, the gap can be 1.0 band.
No {task1} support yet
The current scorer is calibrated for {task2} essays only. {task1} (graphs, letters) uses different criteria and is not yet supported.
Off-topic essays
If the essay is entirely unrelated to the prompt, the AI may still produce a score. A human examiner would likely score this much lower.
Why We Publish This
Most AI scoring tools are black boxes. You submit an essay, get a number, and have no idea how it was calculated. We think that is wrong.
If you understand how scoring works, you can study smarter — targeting the specific criterion that is holding your band down. That is the whole point of our platform.
How We Improve
We regularly review scoring accuracy by comparing AI outputs against human-scored essays. When we find patterns where the AI overscores or underscores, we update the scoring prompt and re-test.
If you believe your essay was scored unfairly, email us. Every report helps us calibrate better.
Questions About Scoring?
If you have questions about how a specific essay was scored, or want to report inaccurate results, contact us.
Contact Us