Where marks are lost

This is a study of thousands of automatically scored answers submitted on EdKnot over seven weeks in July and August 2026: about a third from self-paced practice and two thirds from complete mock tests, across 25 PTE question types. Every answer was marked by the same scoring engine, trait by trait, which makes a question answerable that score reports alone cannot: when candidates lose marks, which trait actually takes them?

• Same engine throughout: practice and mock answers are scored identically, trait by trait.
• Aggregates only: no individual answer was read for this study, and internal and test accounts are excluded.
• Early data: this is a young dataset and we say so where it matters. Task types with fewer than 30 attempts are reported as indicative only. We will rerun the study as volume grows.

One figure per row, each with the population it describes. Quote them with the date; the study is refreshed and older figures are superseded.

Most candidates worry about their accent. The data says they are worrying about the wrong trait: across every speaking task, in both practice and mock exams, pronunciation was the weakest of the three speaking traits in just 0.4–14% of attempts. The other 86–99% of the time, the limiting trait was content or fluency: what you said, and whether you kept saying it.

• Read Aloud: fluency is the weakest trait in 70% of attempts; the words themselves are read correctly 94.6% of the time.
• Describe Image: fluency again is the most common weakest trait, while pronunciation is effectively never the bottleneck.
• What to do: stop drilling accent and drill pace and continuity with the speaking templates, which exist precisely to keep you talking.

Two speaking tasks turn out to be listening-comprehension tasks wearing a microphone. In Retell Lecture, content was the weakest trait in about 8 of every 10 attempts: candidates speak fluently about too little of the lecture. And in Repeat Sentence, errors do not cluster at the end of the sentence the way a memory problem would predict. Accuracy is a flat 62–70% at every word position, and words are almost twice as likely to be substituted (21%) as omitted (12%). Candidates are not forgetting the sentence; they are mishearing it.

• Retell Lecture: content lowest in 78–82% of attempts. The fix is note-taking method, not speaking drills.
• Repeat Sentence: a flat error curve plus substitution-heavy errors point at comprehension, so train it with listening volume, not memorisation tricks.
• Both feed two skills: these tasks score into Listening as well as Speaking, which is exactly why they are worth fixing first.

PTE scores writing structure mechanically: word counts and the one-sentence rule are checked by machine, and a violation zeroes the form trait outright. In this dataset roughly one in three writing attempts did exactly that. Between 36% and 39% of Summarize Written Text attempts scored zero on form, and about 38% of Write Essay attempts did the same. These are marks lost to instructions, not to English.

• The rules are absolute: one sentence for Summarize Written Text, 200–300 words for the essay, and no partial credit on form.
• Templates make form violations impossible: a frame with the right shape satisfies the rules by construction.
• Grammar and vocabulary rarely limit scores: in the same attempts those traits averaged 77–95% of their maximum.

Word-level scoring shows where inside a response the marks disappear. In Read Aloud the loss is concentrated on a single word: the last one, read correctly 89.7% of the time against 96.8% for the first. Candidates trail off or stop the recording early. Write From Dictation shows the same shape, with accuracy falling from the opening words to the closing ones; we hold the exact figures back until a scoring change that affects how word order is credited has been re-measured.

• Read Aloud: finish the final word at full volume before you stop. A seven-point cliff lives there.
• Write From Dictation: the ending is where answers die; if you must choose, secure the end of the sentence.
• A caveat we owe you: dictation is scored word by word against the sentence, which amplifies, but does not create, the decay pattern.

The same question types score about 16 points lower in a full mock test than in self-paced practice, on 15 of the 17 types with enough data to compare. Speaking falls hardest: Repeat Sentence drops 33 points, Respond to a Situation 32, Retell Lecture 23. And the zeros tell their own story. Most zero-scored mock speaking answers are not bad English but near-silence: 91% of Read Aloud zeros came from recordings under one and a half seconds. In practice, where you can re-record, the speaking zero rate was 0.2%.

• Practice scores flatter you: treat them as a ceiling. Only a complete mock measures exam-day reality.
• The silence problem is a conditions problem: nerves, timing and microphone discipline. Rehearse the recording rhythm, not just the answers.
• Plan with the gap in mind: if your target is 79, practising at 79 is not enough. See the 79+ plan.

Taken together, the data argues for a different allocation of practice time than most candidates choose: less accent work and more fluency and note-taking method; absolute discipline on mechanical form rules; deliberate attention to endings; and mock tests treated as the measurement instrument, with the preparation roadmap structuring the weeks between them. We would rather show early, honest data than none, and we will republish as the dataset grows.

• Fluency and content over accent: that is where 86–99% of speaking marks are lost.
• Form rules first in writing: a third of attempts are donating those marks today.
• Mock tests are the truth: practice minus 16 points is your working estimate until a mock says otherwise.

How the figures were produced, in four lines. Every number on this page also appears on the PTE statistics reference with its check date.

Any figure may be quoted with this attribution and the study date. If you need a number that is not here, or a breakdown by task, ask through the contact page.