Oral Practice Report 2026 · Published 11 September 2026 · Updated 20 September 2026
Asked for a personal example, four in ten English oral practice answers gave none.
AI read 12,459 first answers given by 1,290 children practising PSLE English oral and PSLE Chinese oral on PSLEPrep between 1 June and 10 September 2026, and labelled what each one contained. This page is the report: the numbers, how they were measured and checked, and what they do not show.
12,459
first answers read
1,290
children behind those answers
2
languages, English and Chinese
120
labels re-checked by a second AI system
For journalists: download the chart · how we measured it · media contact
The short version
- Asked “tell me about a time…”, four in ten English answers gave no personal example.
- Almost every English opinion answer gives a reason: 96 per cent. Three in four give no example from the child’s own life, which an opinion answer does not have to.
- Chinese answers to the same kind of question gave a personal example far more often than English ones: 78 per cent against 59.
- 8pm is the busiest practice hour. One in three sessions starts between 7pm and 10pm; the median child practised ten times.
- Among English-speaking children with three sessions, the share of “tell me about a time” answers with a personal example rose from 51 to 62 per cent between the first and third. The data cannot say why.
- Answers are not short: a typical first answer runs to about 100 words, roughly a minute of speech.
Finding 1
The missing personal example
Some questions ask the child about their own life: “Have you ever…”, “Tell me about a time…”. Four in ten English answers to those questions gave no personal example.
41%
of English answers to “have you ever…” and “tell me about a time…” gave no personal example.
The same question, answered two ways
Illustrative examples written by PSLEPrep, not children’s transcripts.
Examiner: “Do you enjoy eating at hawker centres? Tell me about a time you went.”
General · no personal example
“Yes, I enjoy hawker centres because the food is cheap and tasty.”
Specific · a personal example
“Yes, because there is so much choice. Last month my grandmother took me to the Bedok hawker centre and I tried chee cheong fun for the first time. I did not like the sauce, so she laughed and bought me chicken rice instead.”
Both answers give a reason. Only the second describes a particular visit.
Share of first answers, before any follow-up question. Children practising on PSLEPrep, 1 June to 10 September 2026.
Gave a personal example when asked “have you ever…”
Gave a reason when asked for an opinion
The numbers behind this chart
| Feature | English | n | Chinese | n |
|---|---|---|---|---|
| Gave a personal example when asked “have you ever…” | 59% | 1,999 | 78.4% | 1,497 |
| Gave a reason when asked for an opinion | 96.3% | 1,956 | 89.9% | 795 |
Get 10 free PSLE Oral practice sessions by email.
Finding 2
Reasons, yes. Examples, less.
96%
of English opinion answers gave a reason.
77%
of the same opinion answers gave no example from the child’s own life.
The answers were not short, and they were not bare assertions. A typical first answer ran to about 100 words, roughly a minute of speech, and only one answer in seventy was a one-liner. What stood out was how often an answer gave a reason but no personal example. The SEAB syllabus lists expressing personal opinions, ideas and experiences among the oral paper’s assessment objectives; it does not say every opinion answer needs an example, and SEAB has said pupils are not expected to reproduce pre-prepared responses. The pattern matters most on the questions that do ask for a personal example, and those are the answers where it most often went missing.
Finding 3
English and Chinese answers are built differently
On the same kind of question, “have you ever…”, 78 per cent of Chinese answers gave a personal example, against 59 per cent of English answers. English answers were more likely to carry a reason. We compare the two languages only within a question type, because the questions themselves differ: the Chinese picture question says “describe what happened”, while the English one asks “what do you think is happening, and why”. The answers also come from different question banks, and only some children practised in both languages, so this describes the answers we saw; it does not establish that the language itself causes the difference.
Finding 4
When families practise
Practice happens after dinner, and it happens repeatedly.
Share of all practice sessions by hour of the day, Singapore time, 1 June to 10 September 2026. Every kind of practice, reading aloud and conversation, both languages.
The numbers behind this chart
| Hour (SGT) | Share of sessions |
|---|---|
| 00:00 | 0.3% |
| 01:00 | 0.1% |
| 02:00 | 0.0% |
| 03:00 | 0.0% |
| 04:00 | 0.0% |
| 05:00 | 0.0% |
| 06:00 | 0.2% |
| 07:00 | 0.4% |
| 08:00 | 1.4% |
| 09:00 | 2.2% |
| 10:00 | 2.6% |
| 11:00 | 3.2% |
| 12:00 | 3.2% |
| 13:00 | 3.2% |
| 14:00 | 5.8% |
| 15:00 | 8.5% |
| 16:00 | 9.4% |
| 17:00 | 8.9% |
| 18:00 | 7.6% |
| 19:00 | 8.6% |
| 20:00 | 14.1% |
| 21:00 | 12.3% |
| 22:00 | 6.3% |
| 23:00 | 1.5% |
35%
of sessions start between 7pm and 10pm.
10
sessions is the median per child.
83%
of conversation sessions answered all three questions.
Practice also bunched up before the exam. More than a quarter of all sessions in the period (27 per cent) began in the four days before the oral examinations started on 12 August.
Finding 5
Three sessions later
For English-speaking children with at least three conversation sessions in the period, we compared each child’s first session with their third. First answers were a fifth longer by the third session for two children in three, and the share of “tell me about a time” answers that gave a personal example rose from 51 to 62 per cent, measured within the same children. This comparison cannot show what caused the change, and a longer answer is not automatically a better one.
One number moved the other way and we do not yet understand it: Chinese answers to “have you ever…” gave a personal example less often by the third session. It may be which questions later sessions ask. We are not drawing a conclusion from it.
For a child preparing now
What to do with this
- Help the child recall real experiences and talk about them aloud: a time they helped someone, a time something went wrong, a time they tried something new. Then practise choosing the one that fits the question. The aim is recall, not a script.
- Ask “when did that actually happen?” “Last year”, “in Primary 4”, “one Saturday”. It is the difference between a habit and a personal example.
- Keep “because”. Reasons are already there in almost every answer.
- On the picture question, answer the second half. One English answer in six described the photograph and never reached “why do you think…”.
The stimulus-based conversation guide covers the answer moves, and picture conversation practice runs the real question sequence with an AI oral examiner.
Method
How we measured it, and what it does not show
Data. From all conversation practice sessions on PSLEPrep between 1 June and 10 September 2026 we drew 6,000, 3,000 in each language: every child with three or more conversation sessions contributes their first three, up to 600 children per language, and the rest is an evenly spaced sample across the whole period. The 12,459 first answers in those sessions come from 1,290 children. A first answer is what the child said to each examiner question before any follow-up. The time-of-day figures use every kind of practice session on the platform in the same period, reading aloud and conversation in both languages, and are reported as shares.
Who is counted. Percentages are shares of answers, so children who practised more count more: by design, about 60 per cent of the answers come from children with three or more sessions, and the ten per cent most active children contribute a quarter of the “have you ever” answers. Counted one answer per child instead, the headline figure is 44 per cent rather than 41; on each child’s first session only, 45. The Chinese example figure and the opinion figures are the same to within a point either way.
Labels. Each answer was read against its question and labelled on three features anyone can check in a transcript: did it give a personal example, did it give a reason, and did it address the question asked. A language model applied the labels. A second AI system, able to read both languages, then re-read 120 of the labelled answers one by one, 60 in each language, with agreement of 93 to 100 per cent depending on the feature. A check of the English labels by a person is under way, and this page will say so when it is done. Answers the transcription could not capture, about 2 per cent, were excluded.
Privacy. The analysis worked from de-identified transcripts, not audio, and reports only aggregated findings. No name, quote or identifier from any child appears in this report.
What is not here. Nothing about pronunciation, marks, or how a child would score in the real examination: the AI examiner’s scoring was not used. A transcript can show that an answer gave a personal example, not whether it happened. SEAB has said its examiners will prompt a pupil who needs support, without penalty; this report counts what children said before any prompt. The families are self-selected, so these are findings about children practising on PSLEPrep, not about Singapore children as a whole. The period includes the run-up to the 2026 oral, when practice is heaviest, so the time-of-day shares describe that period.
For journalists and teachers
Use this data
The figures and the chart on this page may be reproduced free of charge with attribution to the PSLEPrep Oral Practice Report 2026 and a link to this page. Please keep the wording “children practising on PSLEPrep”, and “answers” rather than “children” for the percentages.
Download the chart as a PNG (about 1,800 pixels wide). Suggested citation: PSLEPrep, Oral Practice Report 2026, published 11 September 2026, updated 20 September 2026, psleprep.sg/findings.
Questions about the method, or a cut of the data we have not published: paul@psleprep.sg.
FAQ
Questions people ask about these numbers
Is this a national sample of Singapore children?View answerHide
No. Every number describes children practising on PSLEPrep between 1 June and 10 September 2026, a self-selected group of families who chose to practise. It is a large sample of that group, not a survey of all Primary 5 and 6 pupils, and we do not present it as one.
Are the percentages shares of children or shares of answers?View answerHide
Shares of answers, unless the text says children. A child with three sessions contributes up to three answers to a question type. Counted one answer per child instead, the headline figure is 44 per cent rather than 41; counted on each child’s first session only, it is 45 per cent.
How were the answers labelled, and who checked the labels?View answerHide
Each first answer was read against the question it answered and labelled on features anyone can check in a transcript: whether it gave a personal example, whether it gave a reason, and whether it addressed the question asked. A language model applied the labels. A second AI system, able to read both languages, then re-read 120 of the labelled answers one by one, 60 in each language; agreement was 93 to 100 per cent depending on the feature. A check of the English labels by a person is under way and this page will say so when it is done. The AI examiner’s marks were not used anywhere.
What counts as a personal example?View answerHide
A specific thing that happened to the child, or to their family or friends: “last year my brother…”, “there was a time when…”. Habitual statements such as “we usually go to the park” do not count, and neither does a story about the people in the photograph. A transcript can show that an answer gave a personal example; it cannot show whether the event happened.
Why only first answers, when a real examiner would prompt?View answerHide
Because that is what the data can show cleanly. The Singapore Examinations and Assessment Board said in August 2026 that oral examiners will give additional prompts to a pupil who needs support, without penalty, and that pupils are not expected to reproduce pre-prepared responses. PSLEPrep’s examiner also asks follow-up questions. Counting only the first answer shows what children say before anyone prompts them; it is not a mark, and it does not say what the child would have said once prompted.
Is any child identifiable in this report?View answerHide
No. The analysis worked from de-identified transcripts, not audio, and reports only aggregated findings. No name, quote or identifier from any child appears here.
Can I reuse the chart or the numbers?View answerHide
Yes. The chart and the figures on this page may be reproduced free of charge with attribution to the PSLEPrep Oral Practice Report 2026 and a link to this page. Please keep the wording “children practising on PSLEPrep” rather than “Singapore children”, and “answers” rather than “children” for the percentages.
See what your child’s first answer contains.
One scored PSLE oral practice session with the AI oral examiner, English or Chinese. Ten free sessions, no card.