How accurate is the writing score?

The headline figures: Task 2

Each essay was scored three times, because the scorer varies slightly between runs, and every scoring counts.

Two examiners marking the same essay give exactly the same band only about half the time, so the honest test is within half a band, not an exact match.

What it means if you are aiming for band 7

On Task 2, a 7 here is very likely a 7

None of the essays an examiner put at 6.5 or below came out at 7 or above. It does not over-promise at the band most people need.

A 6.5 here may already be a 7

Of the essays an examiner put at 7 or above, 73% came out below 7. The score is cautious with strong writing: read a 6.5 as “possibly a 7, not yet certain”.

An essay's band here is the middle of its three scorings.

What we measured against

• Task 2 essays from official IELTS preparation books, each with the band an IELTS examiner gave it. None of them had been used to build or adjust the scorer, and they were kept sealed until the check.
• The figures above are for essays of 250 words or more, the length the task asks for, as decided before the check. Counting shorter essays too, 67% of scores were within half a band.
• Task 1: Academic answers with an examiner's band, from the same kinds of book, each scored three times: 80% of scores within half a band. General Training letters: 77%.

How the score is produced

1. The essay is scored from 0 to 9 on each of the four criteria against the public IELTS band descriptors: Task Response (Task Achievement for Task 1), Coherence and Cohesion, Lexical Resource, and Grammatical Range and Accuracy. The word count is supplied, because under-length answers lose marks. For Academic Task 1, the scorer also sees the chart.
2. The task band is the average of the four, rounded down to the half band, exactly as examiners do. 7, 7, 7 and 6 is 6.5, and every criterion must reach 7 before the essay is a 7. Nothing is added afterwards: the band you see is that average.

Where it is weakest

• Strong writing. On average the score sits a little below the examiner's. The hardest essay is one with clear ideas and organisation but a handful of grammar slips. Examiners themselves put such essays anywhere from 6 to 8. The scorer usually says 6.5, and most of its misses sit there.
• Task 1 misses 7s too: 44% of scores for answers an examiner put at 7 or above came out below 7. And unlike Task 2, it sometimes over-promises: 17% of scores for weaker answers came out at 7 or above.
• Very weak or under-length answers are sometimes scored half a band to a band too high.
• The essays are published in books, so the models behind the scorer may have met some of them before. Many were marked under the band descriptors used before May 2023.
• Speaking has not been measured yet.
• The test sets are small. Figures like these move by a few points when re-measured, so read them as a guide, not to the decimal.