Co&Grading

How the feedback works

Co&Grading is an independent practice tool. It uses the official IELTS Writing criteria to estimate performance and suggest revisions. It does not issue official IELTS results.

Four criteria, equally weighted

The app averages four whole-band criterion marks and rounds the practice estimate to the nearest half band, with ties rounded upwards. When both tasks have been assessed, Task 2 receives twice the weight of Task 1. Your target band guides advice and never changes the assessment score.

Reference examples and validation

Our initial development collection contains ten official examiner-scored Task 2 responses across four questions. A matching question group is excluded from the examples supplied when that question is assessed. New validation examples are kept separate from calibration.

On 20 November 2026, the selected API configuration had an average absolute error of 0.85 band across ten development scripts, with 5 of 10 estimates within half a band. It tended to grade low by 0.55 band on average; one strong essay was undergraded by 2.0 bands. Median API assessment time was about 17 seconds. These are single-run development measurements, not a guarantee for new writing.

Separate validation

Two Task 2 essays kept out of calibration scored 6.0 and 7.0 against examiner scores of 5.5 and 7.5. Two Task 1 responses scored 7.0 and 6.0 against examiner scores of 8.5 and 7.0. This is a small validation set, and high-band undergrading remains unresolved. Access is limited to invited trial accounts while grading accuracy is improved.

What the estimate cannot promise

AI feedback can misinterpret a question, overlook language strengths or make an incorrect deduction. Handwritten reference scripts can also introduce transcription uncertainty. Use the feedback to guide practice and discuss important assessment decisions with a qualified teacher.

Official sources

Co&Grading is not affiliated with or endorsed by IELTS, the British Council or IDP.

Open writing studio