Original practice-data studies
Writing research you can inspect
We study submitted writing to understand response length, criterion profiles and patterns in stored practice estimates. Each report includes its evidence tables, methods, limitations and downloadable aggregate data.
October series: submissions from 22 August–5 October 2026. These are self-selected practice samples with AI estimates, not official examination results. Several reports analyze the same cohorts; their response counts must not be added together.
October 2026 studies
4 reports on this site-
5 October 2026 · 134 responses in scope
B2 First essay word count and practice scores: October 2026 study
Among 134 B2 First Essay responses from 129 accounts, the median length was 187 words. The length–score rank correlation was 0.093; this association does not establish a benefit from adding words.
Read the study and data -
5 October 2026 · 230 responses in scope
C1 Advanced essay word count and practice scores: October 2026 study
Among 230 C1 Advanced Essay responses from 212 accounts, the median length was 256 words. The length–score rank correlation was 0.135; this association does not establish a benefit from adding words.
Read the study and data -
5 October 2026 · 364 responses in scope
Cambridge essay criterion profiles: Language, ties and score gaps
Language had the lowest mean stored estimate in both studied cohorts. Shared-lowest and uniquely-lowest counts show how often that pattern applied to individual practice responses.
Read the study and data -
5 October 2026 · 364 responses in scope
B2 First and C1 Advanced essay practice-score distributions
The median stored overall estimates were 4 and 4 in the two studied cohorts. These distributions describe AI-evaluated practice responses, with task cohorts and repeat-writer sensitivity reported separately.
Read the study and data
Shared methods and related research
The series spans IELTS Writing Checker, Cambridge Writing Checker and ScoreQwik. Each report has one canonical publication page.
Earlier benchmark
The August publication describes a separate historical cohort. Its counts and cleaning rules differ from the October series; comparing the two does not establish learner improvement.
Cambridge Writing Benchmark 2026 — published 21 August 2026