How repeat writers change writing-practice statistics: October 2026 study

01

Why is an essay count different from a writer count?

A person who submits ten distinct essays contributes ten observations to a response-weighted average. A person who submits one contributes one. That is appropriate when the question concerns the mix of evaluated responses, but it can mislead when the average is described as the performance of a typical writer.

Exact deduplication addresses copied text, not repeated participation. Ten distinct answers from one account remain ten answers after exact copies have been removed. We therefore report a second summary using the first retained response for each account and task cohort.

An account is the available contributor unit. We do not verify that every account represents one person, and we do not join personal identifiers across products to construct an identity dataset. Contributor-balanced sampling should be understood within that limitation.

02

How concentrated are contributions in the recent corpus?

The final research corpus contains 2,370 responses from 1,383 accounts. Cambridge contributes 549 responses and IELTS contributes 1,821 responses. Those families are kept separate in the contribution analysis.

For the top-decile statistic, we sort account contribution counts within each checker family and include the ceiling of 10% of the accounts. The statistic measures activity concentration; it is not evidence that prolific writers are more or less proficient.

The table refers to the final filtered corpus. It does not combine ScoreQwik imports with their source responses or include the high-frequency account removed during the contamination audit.

03

What changes when each account contributes once per task?

The table below compares the all-response and first-per-account summaries for every cohort that meets the primary reporting threshold. Score means, medians, response lengths and correlations can change by different amounts under this sampling rule.

For Academic IELTS tasks, the first-per-account mean estimates are lower than the corresponding all-response means. That difference cannot be called improvement: it compares two samples, not the same writers at two controlled time points.

General Training Task 2 shows a particularly large change in the length–score correlation. This illustrates why a plausible association in an essay-level table should be checked against contributor concentration before it becomes advice.

04

Why were very high-frequency accounts excluded?

The audit found 1 account contributing more than 50 distinct retained responses during the study window. The exploratory exclusion rule removed 302 responses before the published analyses.

The threshold was selected during the data audit as a conservative contamination precaution. It was not preregistered, and crossing it does not prove that an account was a bot, a staff account or an invalid learner. We publish the rule and exclusion count so its effect on scope is visible.

This series does not publish the excluded account’s scores or identifiers. A future benchmark should define contributor eligibility before examining the results and use an explicitly documented audit process for unusual activity.

05

Which sampling rule should a future report use?

A response-weighted report should describe responses. A contributor-balanced report should identify its account-level selection rule. Both can be useful when their denominators are explicit, but their averages should not be used interchangeably.

The first retained response is defined within the recent research window. It is not a learner’s first-ever essay, a pre-course baseline or a measure of unassisted ability. Exact duplicate removal can also change which response is first available for an account.

To investigate improvement, a new study would need linked task attempts, credible timing, stable evaluation methods and a design that distinguishes revision from new task selection. The current corpus does not supply that design.

The immediate reporting practice is straightforward: put response counts, contributor counts and a contribution sensitivity check next to each substantive finding. That makes the evidence easier to interpret without promising more than the records can establish.

Methods

How this report was prepared

The study uses the latest completed AI evaluation for each eligible submission in the 22 August–5 October 2026 window. Site, task, score and word-count checks precede normalized exact-text deduplication. Accounts contributing more than 50 distinct retained responses in the window are excluded by an exploratory audit rule.

Detailed primary cohorts require 100 responses and 30 accounts. Length and score bins require 20 responses and 10 accounts. The first-response-per-account-and-task analysis is a sensitivity check; it can contain fewer responses than the primary cohort. Account counts across tasks and bins can overlap.

The source does not identify every evaluator model and prompt version, verify examination conditions, or provide independent examiner scores. Lightly edited duplicates and external assistance may remain. Differences describe this practice sample and scoring system; they do not demonstrate official proficiency, causal improvement or universal task difficulty.

Computation and drafting were assisted by AI. This release has no independent human examiner validation, pedagogical review or journal peer review. Only aggregates are published; raw writing, prompts, feedback and learner identifiers are excluded.

Read the full dataset definition, formulas, exclusions and reporting thresholds.

Questions about the findings

Does deduplication make every observation independent?

No. One account can still contribute several distinct responses. The series therefore includes a first-response-per-account-and-task sensitivity check.

Do lower first-response averages prove that people improved with practice?

No. These are different sample summaries. A controlled longitudinal or revision study would be needed to investigate improvement.

Related writing research

• Dataset, methods and reporting rules→

ScoreQwik is an independent practice service. These reports are not affiliated with or endorsed by the examination organizations.