2026 Global PTE Candidate Performance Report

Overview

One of the Largest Real-World PTE Performance Datasets

This report draws on the real activity of 80,440 students who prepared for the PTE Academic exam on the Gurully platform, generating more than one million AI-scored responses across 91 countries. It is one of the largest performance datasets in the English-proficiency category — not a survey or a sampled panel, but the actual scored attempts of learners working toward a high-stakes exam.

Three questions shape what follows: how does practice volume relate to performance, how does score performance differ across the world's major study-abroad markets, and which mistakes cost candidates the most points? The answers point to a clear, repeatable picture of how candidates move from a struggling score to a target score — and where they get stuck.

Practice volume

How mock-test frequency drives score gains

Country-wise trends

Score patterns across major study-abroad markets

Costly mistakes

Which errors cost candidates the most points

The DataSet

The Dataset

Section-level counts are drawn from distinct analytical passes and are not intended to sum to the headline total of 1,013,351 responses.

Metric	Value
Total students analysed	80,440
Total responses analysed	1,013,351
Speaking responses	520,506
Writing responses	45,531
Reading responses	220,747
Listening responses	226,567
Countries represented	91

Overview

Practice Volume vs. Score Performance

The clearest behavioural signal in the data is practice volume. Grouping students by the number of full mock tests they completed shows a strong, consistent relationship between attempts, average best score, and the share reaching a 65+ result — the threshold most universities and immigration pathways require.

Mocks Taken	Students	Avg Best Score	% Reaching 65+
1	45,970	54.2	21.1%
2	4,680	59.2	33.1%
3	3,240	60.7	35.8%
4	1,890	60.6	32.8%
5 or more	7,050	65.8	44.3%

Key Insight

2× more likely to hit 65+

Students who took 5 or more mock tests achieved a 65+ score at more than double the rate of those who only took one — 44.3% vs 21.1%.

Score Jump +11.6 points average gain

+11.6 points average gain

From a 54.2 average on the first mock to 65.8 after five or more — a clear, consistent improvement curve.

Score Tier Analysis

What Separates Score Tiers

Comparing average section scores for learners clustered around three key thresholds reveals what holds candidates back at each stage — and exposes the gap between the skill that feels hardest and the one that quietly keeps them stuck.

Score Level	Attempts	Speaking	Writing	Reading	Listening	Weakest
~50 (45–54)	35,520	46.4	51.7	48.9	54.1	Speaking
~65 (60–69)	51,280	60.8	63.5	61.6	65.6	Speaking
~79 (75–82)	14,500	76.5	74.5	77.5	77.9	Writing

At ~50 and ~65, Speaking is the lowest-scoring section. But Reading is the one that stays stuck — it barely climbs through the plateau and shadows Speaking as the persistent laggard.

The Plateau

The 50–65 Plateau: Where Candidates Get Stuck

Grouping 144,420 attempts into three broad bands exposes the structure of the plateau — the band where the largest share of candidates stall, and the precise skill that holds them there.

BAND	ATTEMPTS	OVERALL	SPEAKING	READING	LISTENING
Under 50	32,050	40.3	36.9	38.4	41.0
50–67	82,330 56% of attempts	57.6	53.8	55.0	60.2
68–79+	29,800	79.6	78.2	81.2	80.5

01

Reading leads at 68-79+

Breaking the Reading plateau is what moves a candidate from stuck to advancing. At 68+, Reading (81.2) becomes the single strongest section.

02

82,330 attempts in the plateau

More candidates sit in the 50-67 band than in the two other bands combined — making it the single most consequential zone to address.

03

Reading barely moves

Speaking feels hardest (53.8) but Reading (55.0) is the silent gatekeeper — barely improved from the band below.

Difficulty Analysis

The Hardest Question Types

Across 601,703 AI-scored Speaking, Writing, and Reading responses, each task was normalised to the percentage of its maximum possible score — isolating which tasks are genuinely hard, not just which carry low raw scores.

Rank	Question Type	Section	Responses	Avg % of Max
1	Reorder Paragraphs (hardest)	Reading	38,420	47.3%
2	Re-tell Lecture	Speaking	28,139	50.7%
3	Summarize Written Text	Writing	30,278	~57%
4	Write Essay	Writing	13,217	57.9%
5	Repeat Sentence	Speaking	169,056	~62%
6	Respond to a Situation	Speaking	30,598	~63%
7	Summarize Group Discussion	Speaking	27,416	63.5%
8	Describe Image	Speaking	80,062	~69%
9	Read Aloud (easiest)	Speaking	104,697	~73%

The easiest tasks — Read Aloud and Describe Image — are also the most heavily practised, while the hardest, highest-leverage tasks are the ones candidates avoid.

Reorder Paragraphs (47.3%) punishes weak text cohesion and logical sequencing. Re-tell Lecture (50.7%) exposes weak note-taking and content recall — both skills a structured practice tool can directly train.

Mistake Analysis

The Most Common Mistakes

Two lenses reveal where points are lost: AI-scored task averages, and a frequency analysis of specific Speaking errors across 517,644 responses.

Top Score-Reducing Patterns

Ranked by impact on overall score

Reorder Paragraphs sequencing 47.3% of max Gaps in text cohesion cost the most points
Re-tell Lecture note-taking 50.7% of max Among the hardest content tasks
Summarize Written Text ~57% One-sentence constraint trips candidates
Essay structure gaps 57.9% of max Blocks upper-band progression despite practice
Repeat Sentence memory limits ~62% 169k responses — a fluency issue
Letting Writing lag Upper band gap Blocks the jump to 79 as weakest section
Not retaking enough Behavioural The most common and most fixable mistake

Specific Speaking Errors — Ranked by Frequency

#	Mistake	Affected	Impact
1	Off-topic / doesn't match question	64,647	Severe
2	Incomplete prompt coverage	39,644+	High
3	Long pauses / hesitation	38,841	Moderate
4	Silence / no speech detected	21,863	Severe
5	Parroting the prompt	18,708	High
6	Poor articulation / word clarity	17,835	Moderate
7	Limited content coverage	11,202	High

Which mistake hurts most?

Silence scores zero across all three sub-skills

Off-topic responses, silence, and reusing prompt wording all trigger a minimum score. Silence is the most catastrophic — entirely preventable by starting to speak within three seconds.

Which mistake happens most?

Relevance, not delivery, is the biggest lever

Content mismatch — off-topic answers — is the single most common failure, affecting 64,647 responses. Candidates lose more points to not answering the question than to pronunciation.

Methodology and notes

Methodology & Data Notes June 2026

Findings are derived from real student activity on the Gurully PTE Academic platform: 80,440 students and 1,013,351 AI-scored responses across 91 countries.
Scores are reported on the PTE 10–90 scale. Section averages are computed within the stated score bands or thresholds for each table.
Question-type difficulty is normalised to the percentage of each task's maximum possible score, enabling direct comparison across tasks with different scoring ranges. Figures marked '~' are approximate.
Section-level response counts are drawn from distinct analytical passes and are not intended to sum to the headline total of 1,013,351.
Country tables include only countries with a minimum of 50 attempts.

Data sourced from Gurully PTE Academic platform · June 2026 gurully.com

The 2026 Global PTE Candidate Performance Report