Imagine Minh, a second-year business student in Hanoi. She has an offer from a university in Melbourne, conditional on an IELTS score of 6.5 overall with no band below 6.0. Her reading and listening are comfortably above 7. Her speaking is improving every week. Her writing, though, is a mystery, even to her.
During the semester she writes reports and essays all the time, and she has learned to work the way most students now do. She outlines with an AI assistant, checks citations with a generator, runs a final draft through a grammar tool and sometimes a detector. That workflow gets her good marks. Then she sits a mock IELTS writing test in a quiet room with a 60-minute timer and nothing else, and her essay comes back as a 5.5.
The gap is not a mystery once you see it. Minh did not get worse at writing. Her practice had stopped looking like the test.
The one exam where your tools stay outside
IELTS Writing is two tasks in 60 minutes. Task 1 asks you to describe a chart, table, map or process in at least 150 words (or, on the General Training test, to write a letter). Task 2 asks for an argument essay of at least 250 words, and it carries twice the weight of Task 1. Whether you take the paper or the computer-delivered test, there is no spellchecker, no thesaurus, no autocomplete and no assistant. Every sentence you hand in is a sentence you built yourself, in real time.
Examiners mark that writing against four published criteria: how fully you answer the task, how clearly your ideas are organised and linked, the range and precision of your vocabulary, and the range and accuracy of your grammar. A student who leans on tools during the semester can be strong on ideas and organisation while her unassisted grammar and word choice quietly fall behind. Those are exactly the habits a tool corrects without telling you, and exactly the ones an examiner notices.
None of this means AI is bad for IELTS preparation. It means the role has to change. During the semester, AI is often your co-writer. For IELTS, it should only be your examiner: you write alone, under the same conditions as the real test, and the tool tells you afterwards what an examiner would have seen.
Practice that looks like the test
The most useful change is also the simplest. Write every practice essay against a clock, with no help, on a real exam-style question. Forty minutes for Task 2, twenty for Task 1. Stop when the time runs out, even if the conclusion is missing, because that is information too: it tells you that your planning took too long or your paragraphs ran on.
This is the routine we designed IELTS Writing Practice around. It is a writing simulator for Academic Task 1, General Training Task 1 and Task 2. The timer starts with your first word, a word counter shows when you have reached the minimum, and you can pick a question from the bank, try a recent exam question, or type in a question your teacher gave you. For Academic Task 1 you can upload the chart itself, so the feedback can check your figures and comparisons against the picture rather than guessing at them.
The point is not the interface. It is that the conditions are honest. An essay written in an hour with a dictionary and a grammar checker open beside it tells you very little about the essay you will write on test day.
Feedback that names the habit, not just the number
When the timer stops, the essay is scored across the four official criteria, and then it is read sentence by sentence. Each correction shows your original words, a suggested version and a one-line reason, labelled with the criterion it affects. “The number of tourists increase” becomes “increased” because the chart shows a past period and the subject is singular. A paragraph that opens with “Moreover” but adds no new idea is flagged under coherence, not grammar, because the problem is the logic, not the word.
This is where students like Minh usually find their pattern. One essay with twelve corrections is overwhelming. Five essays with the same three corrections repeated, such as article errors, missing tense shifts on graphs and over-linked paragraphs, form a short and very fixable list. That list is worth more than any band score, because it tells you what to practise in the next essay.
A sensible weekly cycle looks like this. Write one timed essay. Read the feedback the same day while you still remember why you wrote each sentence. Rewrite only the paragraph affected by your weakest criterion, rather than polishing the whole essay. Then write a new essay on a different question a few days later and see whether the same corrections come back. If they do not, the habit has started to change.
Can you trust an AI score at all?
This is a fair question, and it matters more than it seems. If a tool marks you too generously, you book your test too early. If it marks you too harshly, you spend weeks fixing an essay structure that was already fine. Either way, you waste the study time you have left before your offer deadline.
The only fair way to check a scoring tool is to give it essays whose real band is already known. Cambridge and the official IELTS partners publish sample answers with the band an examiner awarded, so we collected 97 of them, covering Task 1 and Task 2 on both the Academic and General Training tests, and scored each one with five popular tools, including ChatGPT. On average, IELTS Writing Practice was 0.53 of a band away from the examiner and landed within half a band 74.2% of the time. ChatGPT, pasted the same essays, was within half a band 46.4% of the time: a helpful reader, but not a reliable examiner.
We ran that test ourselves, so read it with that in mind. That is also why the full method, the source books and the limitations are published in our 97-essay IELTS writing scoring benchmark, where anyone can check the approach. Whatever tool you use, the practical lesson is the same: treat a single score as a range of about half a band either side, and trust the trend over five or six essays more than any one result.
Learn from essays that have already been judged
Feedback tells you what went wrong. Strong answers show you what right looks like, and for IELTS, “strong” should mean an essay whose band has been checked, not one that simply sounds impressive.
Reading Band 8 and 9 Task 2 sample essays side by side with your own is one of the fastest ways to see the distance between a 6 and an 8. You will rarely find rare vocabulary. You will find a clear position stated early, one main idea per paragraph, examples that actually support the point, and sentences that vary in structure without becoming complicated. Borrow those patterns, not whole paragraphs. A memorised essay is easy for an examiner to spot and almost never fits the question you are actually given.
Back to Minh
Minh did not need a new study plan. She needed her practice to match the test. Over the next six weeks she wrote two timed essays a week with no tools open, read the feedback the same evening, and rewrote one paragraph each time. Her corrections kept coming back to the same three things: articles, tense on Task 1 graphs, and introductions that repeated the question instead of answering it. By week four, two of those had mostly disappeared from her feedback.
The rest of her semester workflow did not change. She still used AI to plan reports and check citations, because at university that is allowed and often expected. She simply stopped using it in the one place where it would not be there on the day.
If you are heading abroad next year, that is the whole idea. Use your tools to study faster, but practise IELTS writing the way you will sit it: alone, on the clock, with honest feedback afterwards. The score you get on test day will then look a lot like the score you have already been seeing.