PTE AI scoring: practice and inspect real feedback
See how RealPTE marks a practice answer: listen to the recording, compare the recognised words, and inspect saved rubric feedback in 5 saved examples. A free account gets 25 AI scores a day. These are practice estimates, not Pearson scores or a guaranteed prediction of your exam result.
At a glance: how each task type is scored
RealPTE practice scoring is different for each kind of task. This table separates objective rules from AI judgements and shows what you can check yourself.
| Task types | How it is marked | Evidence you can inspect | Not measured | Practise |
|---|---|---|---|---|
| Read Aloud, Repeat Sentence | Recognised words lined up against the text; pauses over 0.8 s, pace and fillers for fluency; pronunciation estimated from how clearly each word was recognised | Saved example #010001: recording, recognised words and rubric points Saved example #020002: recording, recognised words and rubric points | Sound-by-sound pronunciation; agreement with Pearson scores | Read Aloud · Repeat Sentence |
| Write from Dictation | Objective: one point per correctly spelled word in the sentence | Saved example #201007: recording, recognised words and rubric points | Nothing beyond the words and their spelling | Write from Dictation |
| Describe Image, Retell Lecture | An AI language model judges content against the key points; fluency and pronunciation as for Read Aloud | No saved public example yet No saved public example yet | Agreement of content judgements with Pearson scores | Describe Image · Retell Lecture |
| Summarize Group Discussion, Respond to a Situation | An AI language model judges the content (speakers' views, the request and register); fluency from timing | Saved example #SGD1-001: recording, recognised words and rubric points (pronunciation not available) Saved example #RTS1-001: recording, recognised words and rubric points (pronunciation not available) | Agreement of content judgements with Pearson scores | SGD · RTS |
| Answer Short Question | Your words matched against accepted answers; an AI check only when they do not match exactly | No saved public example yet | Pronunciation and fluency (only whether the answer is correct counts) | Answer Short Question |
| Summarize Written Text, Write Essay, Summarize Spoken Text | Word count and sentence rules checked exactly; content, grammar and vocabulary judged by an AI language model with fixed rubric instructions | Essay worked example · SWT worked example · SST worked example | Agreement with Pearson's writing scores | Free checkers, no login |
| Reading and listening (multiple choice, fill in the blanks, reorder, highlight) | Objective: marked against the answer key; multiple-answer questions take one point off per wrong choice, never below zero | The marking rule itself; no AI is involved | Nothing: the key decides | All task types |
Pearson's own scoring and AI practice
Pearson scores the real PTE Academic test with its own automated system, and it sells its own Official PTE Academic AI Practice (checked 20 October 2026). RealPTE is independent: our practice scores come from our own code and are not Pearson's engine or Pearson results.
Scoring example
Scored: · #201007
Question from our bank
A person's educational level is closely associated with their social and economic background.
Sample answer we recorded / wrote for this example
Written by RealPTE for this example; not a student submission.
A person's educationel level is closely asociated with their social and economic background.
Actual scoring output
- Correct words
- 11 / 13 points
Reference sentence
Word comparison: matched · different / unclear · missing / replaced
- educationel → educational
- asociated → associated
These are practice estimates, not official Pearson scores.
Each correctly spelled word in order earns one point in this practice result. Correct the two marked spellings, then listen and type the full sentence again.
Scoring example
Scored: · #020002
Question from our bank
A full bibliography is needed at the end of all assignments.
Sample answer we recorded / wrote for this example
Created by RealPTE for this example; not a student submission. Speaking audio is synthetic.
A bibliography is needed at the end of assignments.
Listen to the sample answer
Words recognised from the recording
A bibliography is needed at the end of assignments.
Actual scoring output
- Content
- 2 / 3 points
- Pronunciation estimate
- 5.0 / 5 points
- Fluency estimate
- 5.0 / 5 points
Reference sentence
Word comparison: matched · different / unclear · missing / replaced
- Missing: full
- Missing: all
Recognised words per minute: 231
This saved example uses local speech recognition and the real RealPTE scoring code. Live practice uses a different recognition service, so its transcript and estimates may differ. Pronunciation here is estimated from word matching, not a sound-by-sound assessment.
These are practice estimates, not official Pearson scores. A synthetic voice does not measure a learner's speaking ability.
Compare the marked words with the reference, then practise the full sentence again. For speaking, listen before treating a recognition difference as a pronunciation error.
Revised answer to practise (not scored)
A full bibliography is needed at the end of all assignments.
Restore “full” and “all”, which the saved word comparison marks as missing. Say the complete reference sentence; its revised recording has not been scored.
Scoring example
Scored: · #010001
Question from our bank
The Grand Canyon is 277 miles long, up to 18 miles wide and attains a depth of over a mile. While the specific geologic processes and timing that formed the Grand Canyon are the subject of debate by geologists, recent evidence suggests the Colorado River established its course through the canyon at least 17 million years ago.
Sample answer we recorded / wrote for this example
Created by RealPTE for this example; not a student submission. Speaking audio is synthetic.
The Grand Canyon is 277 miles long, up to 18 miles wide and attains a depth of over a mile. While the geologic processes and timing that formed the Grand Canyon are the subject of debate by geologists, evidence suggests the Colorado River established its course through the canyon at least 17 million years ago.
Listen to the sample answer
Words recognised from the recording
The Grand Canyon is 277 miles long, up to 18 miles wide, and attains a depth of over a mile. While the geologic processes and timing that formed the Grand Canyon are the subject of debate by geologists, evidence suggests the Colorado River established its course through the canyon at least 17 million years ago.
Actual scoring output
- Content
- 0.965 / 1 points
- Pronunciation estimate
- 5.0 / 5 points
- Fluency estimate
- 5.0 / 5 points
Reference sentence
Word comparison: matched · different / unclear · missing / replaced
- Missing: specific
- Missing: recent
Recognised words per minute: 154
Read Aloud content here is the code's normalised word-match ratio, on a 0–1 scale; it is not a Pearson rubric mark.
This saved example uses local speech recognition and the real RealPTE scoring code. Live practice uses a different recognition service, so its transcript and estimates may differ. Pronunciation here is estimated from word matching, not a sound-by-sound assessment.
These are practice estimates, not official Pearson scores. A synthetic voice does not measure a learner's speaking ability.
Compare the marked words with the reference, then practise the full sentence again. For speaking, listen before treating a recognition difference as a pronunciation error.
Revised answer to practise (not scored)
The Grand Canyon is 277 miles long, up to 18 miles wide and attains a depth of over a mile. While the specific geologic processes and timing that formed the Grand Canyon are the subject of debate by geologists, recent evidence suggests the Colorado River established its course through the canyon at least 17 million years ago.
Restore “specific” and “recent”, which the saved comparison marks as missing. The reference below is the complete passage to read; it has not been rescored.
Speaking
Your recording is turned into text word by word, with the time of each word and how clearly it was heard.
For Read Aloud and Repeat Sentence the words are lined up against the text, so you see which words you missed, which you said differently and which were not clearly heard. Numbers are matched however you say them.
Fluency comes from your timing: a gap longer than 0.8 seconds between words counts as a hesitation (1.2 seconds after a comma, 1.5 after a full stop), along with your pace and filler sounds such as "um".
For open answers (Describe Image, Retell Lecture, Respond to a Situation, Summarize Group Discussion) an AI language model judges the content against the key points of the image, lecture or situation.
Writing
Word counts, sentence count and the required length are checked exactly, as Pearson's form rules are exact.
Content, grammar and vocabulary of Summarize Written Text, Write Essay and Summarize Spoken Text are judged by an AI language model using short, fixed rubric instructions. Individual AI judgements can vary.
Write from Dictation is marked word by word against the sentence, one point per correct word.
Reading and listening
Multiple choice, fill in the blanks, reorder paragraphs and highlight incorrect words are marked against the answer keys. Questions with several answers give a point for each right choice and take one off for each wrong one, never below zero, as the real test does.
What it cannot do
It is not Pearson's scoring engine, and every score is an estimate. Pronunciation is estimated from how clearly each word was recognised, not sound by sound. If your recording is silent or cannot be understood, you get no score rather than a made-up one.
Free accounts get 25 AI scores a day on every task type; reading and listening answers are always marked. The full mock test report uses the same scoring: free PTE mock test with a score report.
Choosing a practice tool? Compare PTE practice websites and apps.
Follow one answer through the scoring chain
Listen to the synthetic recording below, read our complete original answer, then compare the recognised words. The saved rubric points and feedback show what the scoring code actually returned. A recognition difference is not proof of a pronunciation error.
The saved examples contain a missing-word Repeat Sentence and a missing-word Read Aloud. There is no saved complete-answer or long-pause counterpart for the same question in this example dataset, so a three-way score comparison is unavailable. No scores for those missing variants are invented.
Open-task feedback may identify covered ideas without detecting every missing detail. For example, the situation answer omits the two-hour deadline, while the saved feedback still credits the presentation request. Listen and check the prompt as well as the score. Synthetic speech and local recognition are not matched learner/exam validation.
Questions students ask
Are practice AI scores the same as Pearson scores?
No. RealPTE scores are practice estimates on the same 10 to 90 scale from our own code, to show what to fix. They are not calibrated against official results and are not Pearson's engine.
What can word matching tell me?
It shows which words the recogniser heard, missed or heard differently. A recognition difference is a reason to listen again, not independent proof of a pronunciation error.
Can I inspect scoring before signing up?
Yes. The 5 saved examples linked in the table above, and the worked examples on the three free writing checkers, can be read and played without an account.
How does RealPTE score speaking?
It turns your recording into text word by word, compares it with the text or the key points, and measures your pauses, pace and how clearly each word was heard.
Why can PTE AI scoring seem unfair?
Speech recognition can miss a word you actually said, and a content judgement can overlook a detail. Listen to your recording, compare the transcript and inspect the feedback before treating one practice estimate as a verdict.
How many AI scores are free?
25 a day on every task type for a free account, plus one full mock test.
Which task types are scored?
All 22 PTE Academic task types, including Respond to a Situation and Summarize Group Discussion.
Sign up free: 25 AI scores a day on every task type.
Practise with AI scoringRelated guides: PTE practice for students in India · PTE practice for students in Nepal · PTE practice with Hindi, Nepali and Urdu support · PTE new question types: SGD and Respond to a Situation