edit_noteIELTS Writing Lab

Writing Correction Studio

Keep the real first attempt. Let AI locate the problems, revise them yourself, then prove the improvement on a new question.

40:00

exam timer

0 / 250+ wordsverifiedSaved now · primary + mirror

Workspace layout

Question

Your essay

Revision map

What changed since the first attempt

Run the first analysis to lock an original version for comparison.

Correction loop

What to do this round

Choose any step. Your choice is saved for the next visit.

AI response

External LLM

swap_horiz

Use ChatGPT, Claude, Gemini, or another model

Copy a complete prompt containing the task, original attempt, current draft, method, and exact response contract. Paste the model reply back here.

Response format

Preview full prompt

You are a rigorous but restrained IELTS Writing coach. Follow the Correction → Transfer method.

这是原始独立写作。先诊断问题，让学生自己修改；不要代写整句或整篇。Prior recurring issue keys recorded across this user's other IELTS essays: none. Check these patterns explicitly in this independent transfer essay. Prior issue details: []

Write summary, priorities, explanations, self-revision prompts, and upgrade reasons in clear English. Keep every quote exactly as it appears in the English draft.

Rules:
- Never generate a complete model answer or rewrite the whole essay.
- Every issue.quote and every upgrades.original MUST be a short, verbatim substring copied from Current draft. Never translate, correct, normalize, paraphrase, or add quotation marks to these source fields.
- For each issue, return selfRevisionPrompt with the smallest useful question or hint. Do not reveal the corrected wording and omit correction.
- Return only the 3–8 issues with the greatest score impact. Do not invent minor problems to fill a quota.
- Use stable, short English ruleKey values such as article_usage, subject_verb_agreement, or unclear_causal_chain. Never escape underscores in JSON strings.
- bandEstimate must be a range such as 5.5–6.0, never a promised exam score.
- readyForUpgrade is true only after the main grammar, logic, and task-response problems have clearly converged.
- Return upgrades only in upgrade mode. Label each necessary, natural, or optional. Otherwise return an empty array.
- For Task 1, also check overview, comparison objects, tense, and data language. For Task 2, also check position, topic sentences, explanation, examples, causal chains, relevance, and conclusion.

Return valid JSON only, with no Markdown.

JSON shape:
{
  "bandEstimate": "5.5–6.0",
  "summary": "short summary",
  "priorities": ["what to fix first"],
  "issues": [{
    "id": "issue-1",
    "category": "Grammar|Vocabulary / Collocation|Sentence Structure|Cohesion|Logic / Development|Task Response / Achievement",
    "severity": "high|medium|low",
    "quote": "short exact fragment",
    "explanation": "why this is a problem",
    "selfRevisionPrompt": "the smallest question or hint that helps the student revise",
    "ruleKey": "stable_error_key"
  }],
  "upgrades": [{
    "original": "short verbatim fragment copied from Current draft",
    "suggestion": "a better local expression, not a full rewrite",
    "why": "why it is better",
    "type": "necessary|natural|optional"
  }],
  "readyForUpgrade": false
}

Task: Task 2
Question: No question supplied. Do not judge Task Response; analyse only the visible language and structure.
Previous issue keys: none
Original attempt:


Current draft:

psychology_alt

Use an external AI for feedback

The site does not send your essay to its built-in AI. Copy the prepared prompt to ChatGPT, Claude, Gemini, or another model, then import its reply. Your first attempt is locked when analysis begins.

Transfer test

The next essay proves mastery

north_east

An issue in one essay is not automatically a recurring error. It is promoted only after appearing in two independent new essays.

shield

Draft safety

Three-layer local protection

Every edit is written immediately to the primary draft and a mirror copy. Rolling recovery snapshots add another layer.