IELTS Academic Writing Samples: Reading the Four Numbers Under the Band

Most collections of IELTS Academic writing samples give you an answer and one number. That number is an average of four, and the four do not move together. Two answers can both sit at overall 7 and need completely different work.

Reading a sample answer properly means reading it four times, once per criterion. This post shows how, using answers published with their four numbers visible, and three real sentences rewritten against the exact descriptor bullet each one lands on.

What an Academic writing sample contains

The Academic writing paper has two tasks and you get 60 minutes for both.

• Task 1 gives you a chart, table, graph, map or process diagram. You summarise it in at least 150 words, in about 20 minutes. You describe what the data shows. You do not give your opinion.
• Task 2 gives you a statement or question. You write an argument in at least 250 words, in about 40 minutes.

Task 2 carries roughly two thirds of your writing band and Task 1 roughly one third. So a complete sample set shows you both, and a page offering only Task 2 essays shows you the larger half of the paper.

A useful sample has four parts: the question, the answer, the four criterion bands, and the reason for each band. Most collections publish the first two. The third and fourth are where the learning is.

The four criteria, and why one number hides them

Both tasks are marked on four criteria of equal weight.

The four whole bands are averaged and the result is rounded to the nearest half band. The band score table sets out which sets of four produce which overall figure, and it is worth seeing once, because it explains why one weak criterion costs you so little and two cost you so much.

This is the first thing to do with any sample answer you find. Before you read the essay, write down the four numbers. If the page does not publish them, you are reading an essay with a label on it, not a marked script.

Where the published samples come from

There are three kinds of sample answer in circulation, and they are not equally reliable.

1. Officially published candidate responses. The test owners publish a small number of real candidate scripts with the band awarded and a written comment from the marker. These are the benchmark. There are not many of them, and that is the point: each one took a qualified marker to produce.
2. Teacher written model answers. A teacher writes an answer at a target level. These are clean and useful for structure, but the band on them is the author's intention, not a mark.
3. Machine estimated answers. An automated scorer gives four bands and a reason. This is what the Bandly sample library publishes, and the numbers on those pages are estimates. They have not been checked against a qualified human examiner, so treat them as a reading of the four criteria rather than as a mark.

Knowing which kind you are reading changes what you take from it. From the first kind, take the band. From the second, take the structure. From the third, take the reasoning: which sentence moved which criterion, and why.

What the numbers look like across 25 published answers

Here is the figure that no collection of samples prints, counted from the 25 answers currently published in the Bandly library. Each page shows its own four numbers, so you can check this by opening them.

Two things follow. First, the content criterion sits high and moves very little, so it discriminates badly: an answer that covers the question is quickly at the top of that column and cannot go further. Our own engine is generous here, which is worth saying plainly rather than hiding. Second, the column that actually varies is Lexical Resource, and it spans three whole bands.

On the four Academic Task 1 answers specifically, Task Achievement was estimated at 8 or 9 on all four, while Lexical Resource came back at 6 on three of them. Coherence and Cohesion never came back above 7 on any of the 25.

So when you compare your own Task 1 answer against a sample, the difference is probably not coverage. You almost certainly described the chart. The gap is in the words you used to do it and in how the sentences join.

Three sentences from one published answer, rewritten

This is a line graph answer about household electricity use in four countries. It was estimated at overall 7, with Task Achievement 8, Coherence and Cohesion 6, Lexical Resource 6 and Grammatical Range and Accuracy 7. The full answer and its reasoning are on the Academic Task 1 line graph page.

Look at what held the vocabulary number at 6.

Original: Average household electricity consumption in Germany decreased in 4100 to 3000 kWh per year between 2000 and 2020.

Rewritten: Average household electricity consumption in Germany fell from 4100 to 3000 kWh per year between 2000 and 2020.

The verb was fine. The preposition after it was wrong, and "decreased in" is not a collocation. Band 6 Lexical Resource says the candidate "attempts to use less common vocabulary but with some inaccuracy". That is this exact sentence.

Original: Meanwhile, India experienced a steady grow from a very low base.

Rewritten: Meanwhile, India saw steady growth from a very low base.

"Grow" is a verb doing a noun's job. The same band 6 bullet covers it: "makes some errors in spelling and/or word formation, but they do not impede communication". The meaning survives, and the band does not.

Original: while the figures for India shows a signifcant percentage increase over the two decades.

Rewritten: The figures for India show a significant percentage increase over the two decades.

Three faults in one clause. A spelling error, a subject that does not agree with its verb, and a clause tacked onto a sentence that was already finished. The first two sit under Lexical Resource and Grammatical Range and Accuracy. The third is why the Coherence and Cohesion estimate came back at 6, with the comment that cohesion between sentences was mechanical.

None of these three fixes needs a wider vocabulary. They need the words already there to be the right form, spelled correctly, in a sentence that ends where it should. That is the realistic route from 6 to 7 on an academic answer, and it is invisible if you only read the overall band.

How to use a sample without copying it

Copying a sample answer does not transfer. The marker is reading your sentences, and a memorised passage scores zero on Task Response if it does not answer the question in front of you.

Use each sample like this.

1. Write your own answer to the sample's question first, under time. 20 minutes for Task 1, 40 for Task 2.
2. Then read the sample and mark the four criteria on your own answer, one at a time, with the descriptor for that criterion open.
3. Find the first sentence in your answer that you would change for each criterion. One sentence per criterion, four in total.
4. Rewrite those four, and name the criterion each rewrite serves.
5. Keep a list of the error types that keep coming back. Three repeated word form errors are one habit, not three mistakes.

For Task 1, work through the question types one at a time rather than mixing them, because the structure of the overview changes with the data. What is different about Academic Task 1 covers that difference against General Training, and Academic Task 2 prompts covers what the essay questions assume you can already do. A worked Task 2 sample answer with the band explained shows the same criterion by criterion reading on an essay rather than a chart.

What to do today

Pick one Academic Task 1 question and one Task 2 question. Write both under exam time. Then mark each of your two answers on the four criteria separately, writing the four numbers down before you compare anything to a sample.

If you want a second reading of those four numbers on your own answer, Bandly estimates each criterion separately and shows the sentences behind each one, which is the part a sample answer with a single label can never give you.