There is no correct number of practice essays for IELTS Writing, and anyone who hands you one has guessed. How much you need to write depends on the gap between your current band and your target, on which of the four marking criteria is costing you the marks, and on whether anything marks your essays at all. Ten unmarked essays and ten marked essays are not the same activity, so no single figure covers both. When a site tells you ten, or twenty, or one a day for a month, ask what that was measured on. If there is no sample size and no scoring behind it, it is a slogan.
We are not going to give you a number either. Here is what to do instead, and the exact conditions under which we will publish a real one.
#Why can nobody give you a universal number of essays?
No universal number exists because the amount of writing practice you need is set by variables that differ from person to person. Four of them matter.
- The size of your gap. Lifting one criterion from 6.0 to 6.5 is a different job from lifting all four from 5.5 to 7.0. The first is a narrow, targeted job. The second is a language project, not an exam project.
- Which criterion is costing you. The British Council states that examiners use the IELTS band descriptors to assess Writing and Speaking. A Task Response problem is usually fixed at the planning stage. A Grammatical Range and Accuracy problem takes far longer. Volume works differently on the two.
- Whether anything marks the essay. This is the big one, and the next section is about it.
- How much time you have, and what else is in your week. Four essays you finish and review beat twelve you plan on a schedule you will not keep.
Task 1 and Task 2 are separate skills. If you count essays, count them per task type.
#What does an unmarked practice essay actually teach you?
An unmarked practice essay teaches you very little about your band, because the errors that cap your score are almost always the ones you cannot see in your own writing. If you could see them, you would not be making them. Writing ten essays with the same article error, the same three linking phrases and the same unanswered half of the question does not remove those habits. It rehearses them.
Unmarked writing is not worthless. It builds typing or handwriting speed, planning speed under pressure, and the stamina to produce two tasks in an hour without collapsing at the end. What it does not do is move a criterion you have not identified.
So the useful version of the question is not "how many essays". It is "how many marked essays, and how many of those did I rewrite afterwards".
#Who or what should mark your practice essays?
Something other than you has to mark them, and in practice that means AI, a teacher, or both. The clearest guidance on the AI half comes from IDP, one of the organisations that owns IELTS. IDP says to "focus on using AI for correction and feedback, but not creation", and to always write your drafts yourself. It warns that AI is "well-known for justifying its mistakes and giving overwhelmingly positive feedback", and advises against relying on AI alone.
IDP goes further than most AI prep companies will quote. It tells candidates to refrain from using LLM tools to grade their essays at all, because the grade will not match how human examiners mark against the official band descriptors. We sell AI scoring and we still think you should read that sentence. IELTS Writing and Speaking are marked by trained human examiners, and IELTS.org states that selected results are marked twice for consistency.
The distinction that makes practice AI defensible is one IELTS.org draws itself, in its May 2026 insight article on automarking: "Using automarkers in low-stakes practice contexts has very different implications from a high-stakes university entry test." A tool that tells you which criterion to work on this week is doing a different job from one that decides your visa.
The most careful published study we know of is Koraishi (2024) in Language Teaching Research Quarterly. It compared ChatGPT 4 against the official published band scores for 55 real Writing Task 2 samples and reported an intraclass correlation of 0.814. The author's own conclusion is that ChatGPT "should not be implemented as an official rater, at least not yet", because individual scripts come out well off the mark even when the averages line up. That is why the method below re-measures over blocks, not after every essay.
#What we can say from teaching, not from data
What follows is teaching judgement from two DELTA qualified teachers with 13 years of classroom experience between them. It is not a measurement, and we are labelling it that way on purpose.
- Fewer essays, marked and then rewritten, beat more essays written once and filed away. The rewrite is where the correction gets applied rather than read.
- Plans are cheap. You can plan six essays in the time it takes to write one, and Task Response is largely decided in the plan.
- Frequency beats bulk. Three short sessions across a week hold better than one long Sunday.
- Full timed papers, both tasks in one sitting, belong near the end. Early on they mostly measure stamina you have not built yet.
#How should you plan your writing practice, starting today?
Work criterion by criterion rather than by essay count. Here is the loop.
- Get one baseline essay marked, under time. Not a draft you polished over an evening. Our free diagnostic is built for this, and the free tier includes three AI writing scores.
- Read the four criterion scores, not the overall band. The band tells you where you are. The criteria tell you what to do on Tuesday.
- Pick the lowest criterion and work on that one only. For Task Response, write plans rather than essays and check each answers every part of the question. For Coherence and Cohesion, rewrite body paragraphs you already have, one idea each, stated in the first sentence. For Lexical Resource, collect phrases from reading on recurring topics. For Grammatical Range and Accuracy, keep an error log and re-edit old essays for those errors only.
- Rewrite before you move on. One marked essay plus one rewrite beats two fresh essays.
- Re-measure on a fresh prompt after every fourth or fifth marked piece. Comparing two consecutive essays tells you little, for the reason Koraishi's outliers illustrate.
- Stop drilling a criterion when it is no longer your lowest. Then repeat with the new one.
One detail makes the loop measurable on our platform. The first full score we give a piece of text is sealed, and identical text returns the identical result on every later submission: same band, same criterion scores, same feedback, byte for byte. A check in our build fails if that stops being true. It is a consistency property, not an accuracy one. It matters here because resubmitting the same essay cannot hand you a different band, so a rescore is only worth reading once you have actually changed the text.
#When will BandNine publish a number?
We will publish a figure for how many marked essays it takes to move a band when we have enough scored attempts to support one, and we will publish the sample size beside it. Our own sample is far too small today, so printing a number now would be marketing dressed as evidence.
When it goes up it will carry the count of scored attempts, the date range, the starting band distribution, the definition of a "move" we used, and a plain statement that a change in our scored band is not an official IELTS result. It will sit on our quality page, beside the weekly scoring run.
That page is also where our accuracy figure lives. Every Saturday at 06:00 UTC an automated run sends our calibration scripts through the live production scoring endpoint, with the sealed score cache bypassed, and records the mean absolute error against our labels. Those labels are written in house by the two of us against the public IELTS band descriptors. The set is small, and not one script in it has been marked by a certified IELTS examiner, which the raw JSON reports openly. Our internal pass mark is 0.50. That figure covers Writing. We do not publish one for Speaking, Listening or Reading. This week's number is on /quality, and the method is set out in how accurate AI IELTS scoring really is.
#The one thing that is not in doubt
Feedback you act on beats volume you do not. IDP reaches that from one direction, by telling candidates to use AI for correction rather than creation. We reach it from the classroom, by watching what changes a script between one draft and the next.
If you want a starting shape rather than a target count, use this one. Get one essay marked this week. Fix the lowest criterion. Rewrite it. Then let the second score tell you how many more you need, instead of a number somebody put in a headline.