How to Revise an IELTS Writing Task 2 Essay One Criterion at a Time

Most people preparing for IELTS Writing Task 2 practice the same way: write an essay, get a number, feel good or bad about it, write another essay. The number rarely moves much, because nothing in that loop tells you what to change or whether your change worked.

This tutorial is about a different habit: treat each essay as a draft, revise it against one criterion at a time, and check which part of the score moved. It works with a teacher, a study partner, or an AI tool. I'll show the process first, then how I use it with a tool I built.

Step 1: Know the four things you are scored on

Task 2 is assessed on four criteria, each weighted equally:

Task Response – Did you answer the actual question, take a clear position, and support it?
Coherence and Cohesion – Can a reader follow your argument from paragraph to paragraph?
Lexical Resource – Is your vocabulary precise and varied, and used correctly?
Grammatical Range and Accuracy – Do you use a mix of sentence structures without frequent errors?

The key point: a single overall band hides which of these is holding you back. An essay can have strong grammar and still lose marks because it only half answers the question. If you don't know which criterion is weak, you'll waste revision time polishing the wrong thing.

Step 2: Write under realistic conditions

Pick a real Task 2 style question, for example:

Some cities now charge drivers a fee to enter the centre on weekdays. To what extent do you agree that this is a fair way to reduce traffic?

Then write it the way the exam asks:

Set a timer for 40 minutes.
Aim for at least 250 words. Going under 250 is penalised under Task Response, so treat it as a hard floor.
Write it yourself. Don't paste the question into a chatbot and edit what comes out. You're training your own writing, and a generated essay teaches you nothing about where your drafts go wrong.

Save this first version untouched. You'll need it for comparison.

Step 3: Get feedback per criterion, tied to your own sentences

Useful feedback has two parts for each criterion:

Evidence: a specific line from your essay that shows the problem.
One action: a single thing to change.

"Your coherence is weak" is not useful. "Your second paragraph opens with 'Moreover' but introduces a counter-argument; change the linking word or move the sentence" is useful.

If a teacher or partner is reading your essay, ask them for exactly this format: four criteria, one quoted sentence each, one change each. Resist asking them to rewrite your paragraphs. A rewritten paragraph shows you their writing, not a path to improving yours.

Step 4: Revise one criterion at a time

Now open your saved draft and make changes, but only for one criterion per pass. A practical order:

Task Response first. If your position is unclear or a paragraph drifts off the question, fixing vocabulary there is wasted effort. Check that your introduction states your view, and that each body paragraph supports it with a reason and an example.
Coherence second. Read only the first sentence of each paragraph. Do they tell the argument on their own? Then look at linking words: are they doing real logical work, or are they decoration ("Furthermore", "In addition") stacked in front of unrelated points?
Lexical Resource third. Look for the word you repeated most. Replace some instances with precise alternatives, but only if you're sure of the meaning. A wrong "advanced" word costs more than a plain correct one.
Grammar last. Fix errors in the sentences you've kept. Then check range: if every sentence is subject-verb-object, combine two with a relative clause or a conditional.

Doing it in this order stops you from polishing sentences you're about to delete.

Step 5: Re-check and compare

This is the step most people skip. After revising, get the essay assessed again against the same four criteria and put the results side by side with the first version:

Criterion	Draft 1	Draft 2
Task Response	?	?
Coherence	?	?
Vocabulary	?	?
Grammar	?	?

Three things can happen for each row: it went up, it stayed, or it went down. All three are informative:

Up: the change worked. Note what you did so you do it in the first draft next time.
Stayed: your change didn't address the real problem, or it was too small. Look at the quoted sentence again.
Down: common with vocabulary. Swapping in unfamiliar words often introduces errors. That's a signal to go back to precise, ordinary language.

Also be honest about noise. A half-band difference between two assessments, human or AI, can be random variation. Look for consistent movement across several essays, not one result.

Doing this loop with an AI estimate

Running this loop with a human reviewer is ideal but slow, since you need someone to read the same essay twice. That's why I built WriteFit, a small tool that runs exactly this loop for Task 2:

You write the essay yourself against a question that stays on screen (up to 40 minutes, 250-word minimum, with early submission allowed). It does not draft the essay for you.
It returns an estimate on the four criteria. Each one comes with a line quoted from your essay and one revision task, not a replacement essay.
You revise, score again, and the page marks which score went up, stayed, or went down.

Two honest caveats. First, it's an estimate, not an official examiner score. The methodology page explains how the estimate was checked against a small set of essays, and notes that a half-band change can still be noise. Second, an AI estimate is a practice signal; it's most useful for spotting which criterion to work on, not for predicting your exam result.

The first score on an account is free, and so is the first rescore of that essay, which is enough to try one full draft-revise-compare cycle.

A weekly routine that actually moves the score

If you have a few weeks before the exam, this is a simple structure:

Two new essays a week, written under timed conditions.
One full revision pass per essay, following the criterion order above.
A short log: for each essay, write down which criterion was weakest and what change moved it.

After four or five essays, your log usually shows a pattern, such as "my Task Response drops whenever the question has two parts" or "I overuse 'Moreover'." That pattern is far more valuable than any single band score, because it tells you what to watch for in the first draft on exam day, when there's no chance to revise.