IELTS.international

Original data · 2026

IELTS Writing mistakes report 2026

We looked at the AI feedback on 667 IELTS Writing Task 2 practice essays that 484 learners typed on IELTS.international between 13 May 2026 and 28 September 2026: which errors show up, what would raise each essay most, and where the band estimates sit.

By Oleksii Vasylenko · Published 28 September 2026 · Data updated 29 September 2026

Headline findings

  1. Small grammar words go wrong in most essays

    84% of essays had articles errors, 79% subject–verb agreement errors and 69% word form errors. articles was named the single priority error in 35% of essays.

  2. The most useful fix: add specific, concrete examples

    It was part of the "single biggest improvement" in 37% of essays. 65% of essays did not back a single claim with a specific example, and reasoning was rated weak or absent in 78%.

  3. 68% paraphrased the question by swapping synonyms

    Only 28% restructured the idea in their introduction, which is what the Lexical Resource descriptors reward.

  4. Grammatical Range and Accuracy was the weakest criterion

    Mean 4.75, against 5.08 for Task Response. A gap of only 0.33 band: the essays are held back by several habits at once rather than one weak skill.

  5. The median estimated band was 5.0

    98% of essays were estimated below Band 6.5. Counting only each learner's first essay gives the same picture: median 5.0 across 484 learners. These are AI estimates; see the limitations below.

  6. 81% slipped out of academic register

    39% were rated too informal and 43% mixed. The most flagged phrase was "a lot / lots of" (22% of essays).

Where the band estimates sit

Overall band is the mean of the four criterion bands, rounded to the nearest half band. The mean was 5.02; 1% of essays reached Band 7 or higher.

Estimated overall band, share of essays
BandEssaysShare
3.5 or below152%
4.013721%
4.58513%
5.015123%
5.511017%
6.015423%
6.5 or above152%
667 essays. Bands below 4.0 and from 6.5 up are pooled because each holds only a handful of essays. The lighter column is Band 6.5 or above.
Mean band by criterion (scale 0–9)
CriterionMean band
Task Response5.08
Coherence and Cohesion4.88
Lexical Resource5.06
Grammatical Range and Accuracy4.75

The grammar errors in most essays

Share of essays in which the AI evaluation found at least one error of each type. The median essay had 7.1 errors per 100 words.

Essays with at least one error of this type
ItemShare of essays
Articles (a / an / the)84%557 essays
Subject–verb agreement79%525 essays
Word form (economy / economic)69%460 essays
Punctuation64%424 essays
Prepositions62%411 essays
Sentence fragments and run-ons32%212 essays
Spelling31%208 essays
Verb tense23%151 essays
The one error type named as the priority to fix
ItemShare of essays
Articles (a / an / the)35%232 essays
Punctuation17%113 essays
Spelling13%89 essays
Subject–verb agreement11%74 essays
Word form (economy / economic)10%67 essays
Other8%54 essays
Sentence fragments and run-ons4%23 essays
Each essay has one priority error. Types named in fewer than 20 essays are left out.

The single biggest improvement

Every evaluation ends with the one change that would raise the essay most. Grouped by theme, the most common was add specific, concrete examples. Themes named in fewer than 20 essays are left out.

Themes in the “single biggest improvement” note
ItemShare of essays
Add specific, concrete examples37%248 essays
Develop and explain ideas further33%218 essays
Organise into clear paragraphs27%177 essays
Write a proper conclusion21%142 essays
State a clear position or thesis21%138 essays
Fix basic grammar accuracy14%97 essays
Answer every part of the question12%77 essays
Reach the 250-word minimum3%22 essays
A note can match more than one theme, so shares do not add up to 100%.

Paraphrase and argument

How the introduction paraphrased the question
  • Swapped synonyms only68%
  • Restructured the idea28%
  • Copied the question or no paraphrase4%
  • Other or mixed1%
Quality of the reasoning chain
  • Weak70%
  • Moderate21%
  • Absent8%
  • Strong2%

645 essays with this rating.

No claim backed by a specific example
65%
Topic sentences unclear or inconsistent
75%
Conclusions that sum up the main points
13%

Linking words and register

42% of essays leaned too hard on at least one linking word, and 93% used at least one phrase the evaluation flagged as informal. Only words and phrases flagged in 20 or more essays are listed.

Register of the essay
  • Mixed43%
  • Too informal39%
  • Mostly academic16%
  • Consistently academic2%

Most flagged words and phrases

  • and overused as a linker7%
  • also overused as a linker5%
  • a lot / lots of flagged as informal22%
  • kids flagged as informal8%

What to do about it

Three habits cover most of what this report found: check articles, subject–verb agreement and word form before you submit; act on your own single biggest improvement (add specific, concrete examples was the most common); and paraphrase the question by changing its structure, not just its words.

Methodology and limitations

Source: IELTS.international production database, table writing_submissions: Task 2 practice essays and the stored AI evaluation of each. We read 1,574 Task 2 essays and kept 667 (907 excluded). Median length 258 words; median writing time 27.2 minutes. Snapshot generated 29 September 2026.

Inclusion rules

  • IELTS Writing Task 2 essays with a complete AI evaluation (four criterion bands plus the error analysis).
  • Typed in the timed essay editor, with a recorded writing time.
  • Written at 60 words per minute or slower.
  • At least 150 words.
  • Real learner accounts only: test and staff accounts removed.
  • A 2026-09-28 college lab cohort removed until the college approves publication.
  • Overall band = mean of the four criterion bands, rounded to the nearest half band (.25 and .75 round up).
  • Word and phrase lists show only items flagged in at least 20 essays.
Exclusions, applied in order
RuleEssays
No complete AI evaluation (missing one of the four criterion bands or the error analysis).119
A 2026-09-28 college lab cohort, excluded until the college approves publication (its email domains, or accounts created 2026-09-28 02:00–08:30 UTC).274
Submitted in paste mode rather than typed in the timed editor.432
No recorded writing time, so the typing-speed check cannot be applied.25
Written faster than 60 words per minute (word count ÷ minutes spent): likely pasted or prepared in advance.57

Typing speed: across every timed Task 2 essay, 120 arrived faster than 60 words a minute; 57 of those were still in the sample when that rule was applied and were removed. 274 essays from a college lab session were left out until the college approves publication. Omitted: the platform does not store a country for most learners, so no country cut meets the 50-essay minimum per group.

  • AI-estimated practice bands, not official scores

    Every band and error label here comes from IELTS.international's AI evaluation of practice essays. They are not IELTS results and are not endorsed by the British Council, IDP or Cambridge. The AI can mislabel an error or miss one.

  • Few high bands

    Only 2% of essays were estimated at Band 6.5 or above. Part of that is who practises on a free checker; part may be the scorer, which our scoring study found tends to mark strong essays too low. Read the headline share below 6.5 with that in mind.

  • Who is in the sample

    Self-selected learners practising on IELTS.international, 484 of them, between 13 May 2026 and 28 September 2026. Most wrote one essay (median 1). They are not a random sample of IELTS candidates, and we do not know their country or test date, so there is no country breakdown.

  • Essays, not people

    Shares count essays. A learner with several essays counts several times; the first-essay-per-learner figures above check that this does not change the band picture.

  • Themes are keyword-matched

    The "single biggest improvement" is free text written by the AI. We grouped it into themes by keyword rules (published in the code), and one note can match several themes, so theme shares do not add up to 100%.

How the scores are produced: how scoring works. How close the AI gets to reference bands: AI scoring accuracy study. Our publication standard: research methodology.

Cite this report

APA

IELTS.international. (2026). IELTS Writing mistakes report 2026: Errors in 667 AI-evaluated IELTS Writing Task 2 practice essays [Data set and report]. https://www.ielts.international/research/ielts-writing-mistakes-2026

Permalink: https://www.ielts.international/research/ielts-writing-mistakes-2026. Data licensed CC BY-NC 4.0: free to quote and reuse with credit to IELTS.international.