Four scores go into your IELTS Writing band, and each carries the same weight. Most candidates can name all four. Far fewer can say what separates a 6 from a 7 inside any single one of them, and that is the only thing that moves the number.
The band descriptors are public. The British Council's page on how IELTS is assessed states that examiners use the IELTS band descriptors to assess Writing and Speaking, and IDP, one of the test co-owners, says the examiner awards a band score for each of the four criteria. The marking is not a secret system. It is a page of text that rewards behaviours you can practise.
What follows is one criterion at a time: what the public descriptor asks for at bands 6, 7 and 8, the most common way candidates hand the band back, and a before and after pair we wrote to show the fix. The descriptor wording is copyright, so what you get here is a paraphrase of it.
#How is the IELTS Writing score calculated?
Your band for each Writing task is the average of four criterion scores, each awarded on the nine band scale, each worth exactly one quarter of that task's score. There is no fifth criterion and no bonus for length.
| Criterion | Share | What it judges |
|---|---|---|
| Task Response · Task Achievement in Task 1 | One quarter | Whether you answered the exact question asked, in full, with developed ideas |
| Coherence and Cohesion | One quarter | Whether a reader can follow your order of ideas without effort |
| Lexical Resource | One quarter | Range and precision of vocabulary, including spelling and word formation |
| Grammatical Range and Accuracy | One quarter | Variety of sentence structures, and how many sentences contain no errors |
The two tasks then combine into one Writing band, with Task 2 counting for more of it than Task 1. That is why the standard advice gives 20 minutes to Task 1 and 40 to Task 2.
The person applying the four criteria is a trained human. IELTS.org calls Writing and Speaking the examiner rated sections and credits their reliability to face to face examiner training and recertification every two years. Its Trust IELTS page adds that selected results are marked twice. So every tip you follow should trace back to a descriptor line you can point at. "Use advanced vocabulary" is not one. Precision is.
#What does Task Response actually ask for?
Task Response asks whether you answered the exact question in front of you, held a clear position on it, and developed your main ideas with support, rather than whether your essay is broadly about the right topic. It is where a strong essay can still lose the most: a strong essay answering a slightly different question is still answering the wrong one.
The public Task 2 descriptor, across the three bands that matter most:
- Band 6. All parts addressed, some more fully than others. A relevant position, though the conclusion may go unclear or repetitive. Relevant main ideas, but some underdeveloped.
- Band 7. All parts addressed. A clear position held throughout the response, not announced at the end. Main ideas extended and supported, with a tendency to over generalise still tolerated.
- Band 8. All parts covered sufficiently, and a well developed response with ideas that are relevant, extended and supported.
In Task 1 the criterion is Task Achievement. Band 6 addresses the requirements and includes an overview, but details may be inaccurate. Band 7 wants a clear overview of the main trends, differences or stages, with key features clearly highlighted. Band 8 presents, highlights and illustrates those features clearly. A General Training letter also needs a clear purpose and a consistent tone from band 7 up.
The most common way candidates lose it: writing about the topic instead of answering the question. "To what extent do you agree" asks for a degree of agreement, not a tour of both sides, and a two part prompt needs both parts at the same depth. In Task 1 the equivalent is the missing overview: every number listed, no statement of the pattern.
An opening line for the prompt "Some people think governments should spend money on public transport rather than on new roads. To what extent do you agree?"
- Before. There are both advantages and disadvantages to spending money on public transport, and this essay will discuss both sides before giving my opinion.
- After. Governments should put the larger share of their transport budget into buses and trains, because a bus lane moves more people per metre of road than a new lane for cars ever will. Road building buys a few years of relief, and then the traffic returns.
Why that is worth a band. The after version commits to a position in its first clause and attaches a reason the body can extend. That is band 7's clear position throughout, with main ideas extended and supported. The before version announces a plan instead of taking a view, and an essay that finds its position only in the conclusion sits at band 6.
The Task 1 version of the same fix:
- Before. The chart shows the number of visitors to three museums between 2010 and 2020. In 2010, the city museum had 1.2 million visitors.
- After. Overall, all three museums drew more visitors by 2020, but the growth was uneven: the city museum more than doubled, while the maritime museum finished close to where it began.
Why that is worth a band. The after version is an overview of main trends and differences, the band 7 line for Task Achievement, and it summarises without listing. The numbers come next. More in our guides to Task 1 and Task 2.
#What does coherence and cohesion mean in IELTS?
Coherence and cohesion measures two things at once: whether your ideas arrive in an order a reader can follow, which is coherence, and whether the links between sentences and paragraphs are made clearly and appropriately, which is cohesion. Candidates are usually stronger on one than the other.
- Band 6. Ideas arranged coherently with clear overall progression, but links within or between sentences can be faulty or mechanical, referencing is not always clear, and paragraphing is not always logical.
- Band 7. Logical organisation and clear progression throughout. A range of cohesive devices used appropriately, though some may be over or under used. Each paragraph has a clear central topic.
- Band 8. Ideas sequenced logically, all aspects of cohesion managed well, paragraphing sufficient and appropriate.
The most common way candidates lose it: bolting connectors onto the front of sentences that are not in the logical relationship the connector claims. "Mechanical" is the word doing the damage at band 6. A close second is the paragraph carrying three unrelated ideas, which fails the band 7 requirement for one central topic on its own.
- Before. Firstly, cities are crowded. Moreover, air pollution is increasing. Furthermore, people are unhappy with their commute. In conclusion, this is a serious problem.
- After. Cities are crowded, and that crowding is what fouls the air: more cars idle for longer in the same square kilometre. Commuters then breathe the result for two hours a day, which is why traffic comes near the top of every survey of what residents dislike about city life.
Why that is worth a band. The after version carries its links inside the sentences, through referencing ("that crowding", "the result") and one ordinary sequencer ("then"), instead of four front loaded connectors doing no logical work. Band 7 asks for cohesive devices used appropriately and flags over use, and band 6 names referencing as the thing not yet controlled. More in linking words and cohesion.
#What counts as lexical resource, and do rare words raise your band?
Lexical resource measures whether your vocabulary is wide enough and precise enough to say what you actually mean, and rare words only raise your band when they are accurate and sit in their normal collocations. A rare word used slightly wrongly is evidence against you, because the higher bands reward control, not ambition.
- Band 6. An adequate range for the task. Less common vocabulary attempted, but with some inaccuracy. Some spelling and word formation errors that do not block meaning.
- Band 7. Enough range for flexibility and precision. Less common items used with some awareness of style and collocation. Occasional errors in choice, spelling or word formation.
- Band 8. A wide range used fluently and flexibly to convey precise meaning, uncommon items used skilfully, and rare spelling or word formation errors.
Notice what is missing. The descriptor never asks for academic vocabulary or a quota of high level words. It asks for range in the service of precision, and it puts spelling and word formation in the same criterion, which makes proofreading time band scoring time.
The most common way candidates lose it: thesaurus substitution. You swap an ordinary word for a rarer one at the last minute, the collocation breaks, and you fail the band 7 line on awareness of collocation and style.
- Before. This problem is very grave, so the government must do a strong effort to abolish it.
- After. The problem is serious enough to justify emergency funding, so the government needs to make a sustained effort to reduce it.
Why that is worth a band. "Do a strong effort" and "abolish a problem" are collocation errors, and "grave" is the wrong register here. The after version keeps the collocations English actually uses and buys its precision from "emergency funding" and "sustained". That satisfies the band 7 line on collocation and style without one unusual word.
Precision is often just naming the thing. "Technology has a very big impact on people's lives" says nothing that "rota software has cut the admin load on shift managers" does not say better. The second is not more advanced. It is more specific, and specificity is what an examiner can credit. See our band 7 vocabulary guide.
#What does grammatical range and accuracy measure?
Grammatical range and accuracy measures two things at the same time: how many different sentence structures you can control, and how many of your sentences contain no errors at all. Range without accuracy stalls at band 6, and accuracy without range stalls in the same place.
- Band 6. A mix of simple and complex forms, with some grammar and punctuation errors that rarely reduce communication.
- Band 7. A variety of complex structures, frequent error free sentences, and good control of grammar and punctuation with only a few errors.
- Band 8. A wide range of structures, the majority of sentences error free, and only very occasional errors or inappropriacies.
"Frequent error free sentences" is the most countable phrase in the descriptor. Print your last essay and mark each sentence clean or not clean. If fewer than half are clean, extra complexity will not lift this criterion, because every new subordinate clause is another chance to make an error.
The most common way candidates lose it: reaching for complexity and losing accuracy on the way. In our teaching the two errors that recur most in band 6 scripts are the comma splice, where two full sentences are joined with only a comma, and the missing article. Punctuation sits inside this criterion, so a comma splice is an error the descriptor counts.
- Before. Many students are studying abroad. They want better job. It is expensive. Their parents must pay a lot, this is a problem for them.
- After. Many students study abroad because a foreign degree opens doors at home, even though their parents often carry a cost that takes years to recover.
Why that is worth a band. The after version supplies the missing article, removes the comma splice, and subordinates two ideas with "because" and "even though" inside one clean sentence. Band 7 wants a variety of complex structures and frequent error free sentences. The before version offers neither: no range, and two of four sentences carry errors. Build range safely with our band 7 structures.
#What does a 6.5 in Writing actually look like?
A 6.5 in Writing never comes from four scores of 6.5, because each criterion is awarded a whole band, and the mix tells you what to do next. Three candidates, same overall band:
| Profile | Task Response | Coherence | Lexis | Grammar | Average |
|---|---|---|---|---|---|
| A | 7 | 7 | 6 | 6 | 6.5 |
| B | 6 | 6 | 7 | 7 | 6.5 |
| C | 7 | 7 | 7 | 5 | 6.5 |
Same band, opposite homework. A has ideas and structure and is held back by language. B writes accurate English about the wrong question. C is the instructive one: three criteria at 7 and one at 5 still averages 6.5, so lifting that 5 to a 7 makes the whole score a 7, while polishing the three 7s without lifting any of them a whole band changes nothing. Our band calculator and guide to IELTS band scores cover the rest of the sums.
#Which criteria can AI feedback be trusted on?
AI feedback is on its firmest ground with lexical resource and grammatical range and accuracy, because the evidence for those sits on the page, and on its weakest ground with task response and coherence and cohesion, because those turn on judgement about ideas. In its May 2026 insight article on what automarking means for language test validity and integrity, IELTS names higher order skills among the limitations of automarking, and lists organisation, idea development and the nuances of human communication.
That article is widely misquoted, so note two things. It never says IELTS uses automarking on the live test, and it argues that a hybrid model with a high degree of human oversight is currently essential, citing the UK regulator Ofqual's rule that AI cannot be the sole determinant of results in high stakes qualifications. It also makes the distinction that matters at home: automarking in low stakes practice has very different implications from a high stakes entry test.
IDP is blunter. Its guidance on using AI to prepare tells candidates to refrain from using LLM tools to grade their essays, because the grade will not match how a human examiner marks against the official band descriptors, and it warns that AI tends towards overwhelmingly positive feedback. Keep its constructive rule: use AI for correction and feedback, not for creation, and write your first draft yourself. The longer answer is in can ChatGPT score my IELTS essay.
Here is what we claim for our own scorer. It returns a band for each of the four criteria alongside the overall band, and the first full score for a piece of text is sealed: resubmit the same essay and you get the same band, the same criterion scores and the same feedback, byte for byte. That is a consistency property, not an accuracy one, and it matters because any change you see after a rewrite came from the rewrite. On accuracy we publish a figure every Saturday at /quality, from an automated run through the live scoring endpoint with the sealed score cache bypassed. Read the caveat with it. The calibration set holds 24 scripts, every label written in house by the two of us against the public descriptors, and not one has been marked by a certified IELTS examiner. The figure covers Writing, and /quality breaks it down criterion by criterion as well as overall. We publish no equivalent for speaking.
#How to use this on Monday morning
Pick the one criterion costing you the most and work only on that for two weeks. Lifting all four at once is how candidates put in months of work and move nothing, because the four pull against each other in the short run: rarer vocabulary damages accuracy, and complex sentences damage coherence.
- Get four numbers, not one. An overall band hides where the loss is. Score a full timed essay and insist on a band per criterion.
- Take the lowest. If two are level, take Task Response first, because a fix there usually improves coherence as a side effect.
- Find the exact line you fail. Open the public descriptor at the band above yours and read the clauses for that criterion. One will describe you accurately. That clause is the target, not the whole band.
- Drill one task type. Write three responses to the same question type in a row, fixing only that clause. Changing task type every session hides whether the fix worked.
- Re-score criterion by criterion. The overall band moves in half steps, so it is a poor progress signal over two weeks. The criterion score is the signal.
Our free diagnostic gives you that criterion breakdown, and the free tier includes three AI writing scores. Every writing task on the site was written by us rather than reproduced from the real exam.
One last thing. Print the descriptor and mark your own last essay against it, criterion by criterion, before you look at a score from any tool, human or machine. Learning to predict your own four criterion scores is the real goal, because it means you have internalised the descriptor and not just the feedback.