Cambridge Writing Benchmark 2026: 9,158 Learner Responses | Cambridge Writing Checker
Cambridge Writing Checker Logo
Original research Snapshot through 21 Aug 2026

Cambridge Writing Benchmark 2026

Task choice, response length, and criterion patterns across 9,158 unique learner practice responses.

Report 01

Written by Lucas Weaver

Published 21 August 2026

Dataset snapshot

Completed evaluations
11,569
Provisional cleaned responses
9,158
Learners
3,691

Read this first

A picture of practice, not official exam performance

This report describes writing submitted to Cambridge Writing Checker. It covers B2 First, C1 Advanced, and C2 Proficiency practice tasks completed on the site through 21 August 2026.

The stored scores are AI-generated practice estimates. They are not official Cambridge English results, and this sample does not represent every exam candidate. The useful question is narrower: what patterns appear among people who chose to practise with this checker?

Finding 01

Essays dominate the practice corpus

C1 Advanced essays form the largest single cohort, followed by B2 First essays. Reports, reviews, proposals, articles, and letters or emails still contribute thousands of responses in total, giving the corpus useful task variety.

Cleaned unique responses by selected task and level

C1 Advanced

Essay

3,328

B2 First

Essay

1,980

C1 Advanced

Letter or email

493

C1 Advanced

Report

490

C1 Advanced

Review

458

C2 Proficiency

Essay

449

C1 Advanced

Proposal

444

B2 First

Letter or email

427

B2 First

Article

356

B2 First

Review

257

B2 First

Report

142

C2 Proficiency

Article

99

C2 Proficiency

Review

71

C2 Proficiency

Report

66

C2 Proficiency

Letter or email

57

The 15 displayed cohorts account for 9,117 responses. The remaining 41 responses belong to smaller task labels that are not separated here.

What learners can take from this

Essay guidance is easy to find because essays are practised often. Less common formats deserve deliberate preparation too: their purpose, reader, structure, and register differ.

What teachers can take from this

The dataset is strongest for B2 and C1 essays. Conclusions about smaller C2 task cohorts should be treated more cautiously.

Finding 02

Median response length changes with level

The median is the middle response after lengths are sorted. It describes this practice sample; it does not identify an ideal length or prove that writing more produces a better result.

Median word count by Cambridge writing level and task
LevelTaskMedian words
C1 AdvancedEssay259
B2 FirstEssay192
C1 AdvancedLetter or email255
C1 AdvancedReport260
C1 AdvancedReview260
C2 ProficiencyEssay285
C1 AdvancedProposal260

Practical interpretation

Treat word count as a constraint, not a target to maximize. Plan enough space to answer the task, develop the important points, and revise language that is unclear or inaccurate.

Finding 03

Language was the lowest mean stored criterion across the large task cohorts

For every large task cohort included in this comparison, the mean AI practice estimate for Language was below the corresponding means for Content, Communicative Achievement, and Organisation.

This is a consistent signal inside the cleaned data, but it is not proof that Language is the weakest criterion for all Cambridge candidates. The pattern may also reflect who used the checker and how the scoring system behaved.

Language includes more than grammar

The criterion considers vocabulary and grammar range, control, precision, flexibility, spelling, and accuracy.

Revise at sentence level

After checking task coverage and structure, review whether each sentence expresses its idea accurately and suits the intended reader.

Methodology

How the snapshot was prepared

  1. 1

    Start with completed evaluations

    The source contained 11,569 completed evaluations associated with Cambridge Writing Checker.

  2. 2

    Apply validity filters

    Records had to match the correct site and exam, contain a plausible stored score from 1 to 5, and meet the minimum word-count rule used for this analysis.

  3. 3

    Remove exact duplicate text

    If identical response text appeared more than once, it was counted once in the cleaned corpus. This removes exact copies only; lightly edited versions may remain separate.

  4. 4

    Publish aggregates

    The provisional corpus contained 9,158 unique responses from 3,691 learners after filtering. This page reports counts, medians, and a criterion-level pattern, not raw essays.

Limits and safeguards

What this report cannot establish

The sample is self-selected

These learners chose an online checker. Their task choices and writing may differ from the wider Cambridge candidate population.

Practice estimates are not official marks

The criterion scores were produced by an AI practice system, not Cambridge examiners. No official result or pass rate can be inferred.

Historical scoring versions are not stored

Historical evaluations do not reliably identify the evaluator model and prompt version. A change in stored scores over time could reflect a cohort change, a scoring-system change, or both.

Repeated writers may influence aggregate patterns

The cleaned corpus deduplicates exact text, not learners. People who submitted many distinct responses can contribute more than once.

Smaller task cohorts carry more uncertainty

C2 articles, reviews, reports, and letters or emails have much smaller samples than the main B2 and C1 essay cohorts.

Lucas Weaver

About the author

Lucas Weaver

Lucas founded Cambridge Writing Checker after teaching Cambridge exam students from more than 35 countries over nine years. Read about the checker and its founder.

Apply the findings

Check a response against the four Cambridge Writing criteria

Choose your exam, paste the complete task and response, and use the estimate to plan your next revision.

Try the writing checker