Glossary term

AI Scoring

AI scoring is Pearson’s automated marking of every PTE item. The engine is trained on responses rated by human experts to every speaking and writing question, scores each response trait by trait against published descriptors, and is backed by human review of Content on the seven longest tasks. It is why results arrive within days rather than weeks.

Updated 23 September 2026

In plain language

Every item is scored automatically, human experts review the long tasks, and the marking follows published descriptors you can read. Preparing against those descriptors is the whole strategy.

Where you meet it

The report shows only the overall and four skill scores. Pearson’s score guide describes the scoring: automated for all items, with human expert review on named tasks and a second human to resolve disagreements. EdKnot practice is scored entirely automatically; no person reviews practice responses.

How the marking works

Pearson’s score guide describes two layers. Human expert raters score responses to every speaking and writing question, and the automated system is trained to reproduce those standards. Live responses are then scored automatically, trait by trait: a Describe Image response gets a Content score, an Oral fluency score and a Pronunciation score, each on its published scale, and the three are added.

On the seven long tasks a human expert reviews the Content score before it is final, and on Write Essay two further traits. Where the human and the machine disagree, a second human decides. Reading items, most listening items and the short speaking tasks are scored automatically without review.

What automated marking means for preparation

The descriptors are public, so the target is specific. Speaking traits count hesitations, repetitions and false starts, and judge intelligibility rather than accent. Writing traits count words, sentences and spelling errors exactly, and the Content descriptors reward paraphrase and specificity while naming copied and generic language as failures.

Two consequences. First, mechanical accuracy on the short tasks is fully rewarded: every word in Read Aloud, every word in sequence in Repeat Sentence, every correctly spelled word in dictation. Second, on the long tasks there is a reader in the loop, so a template with empty slots or a memorised paragraph is scored as what it is.

The numbers that matter

Items scored by AI	All of them, in both tests
Human review of Content	Describe Image, Retell Lecture, Respond to a Situation, Summarize Group Discussion, Summarize Written Text, Write Essay, Summarize Spoken Text
Training data	Human raters score responses to every speaking and writing question; the engine replicates those standards
Results	Typically within 48 hours; never later than 5 business days
Challenge route	Rescore of the whole test, requested within 14 days, US$160, refunded if any score changes

Worked example

The same Read Aloud, two ways

A learner reads the text with a strong regional accent, every word clear, in phrases, no restarts. The engine recognises every word: full Content, top bands on fluency and pronunciation. Another reads with a “neutral” accent, stops twice to correct a word, and pauses mid-phrase. Fewer Content errors are not enough: the restarts and pauses put fluency in the Intermediate band.

What it shows: The engine scores what the descriptors describe. Accent is not in them; restarts are.

Tasks where it applies

Frequently asked questions

Is PTE marked by a computer or by people?

Every item is scored automatically. Human experts rate the training responses the engine learns from, and a human reviews the Content score on the seven long tasks before results are released. If you disagree with a result you can pay for a rescore of the whole test.

Can AI scoring be fooled by templates?

Not reliably. The Content descriptors on the long tasks penalise generic and copied language, and a human reviews those scores. Templates that fix the order of moves help; memorised text that fits any prompt scores as generic.

Does the AI understand my accent?

The Pronunciation descriptors are about intelligibility to regular speakers, with natural connected-speech features expected. Any accent that leaves words recognisable scores in the top bands. Consistent distortion of particular sounds or wrong word stress is what costs marks.