On six of the eight Speaking tasks you get thirty seconds of preparation. Not two minutes — thirty seconds. The component is 15 minutes for 8 tasks — under two minutes each, preparation included.
Almost every Speaking problem people describe as "my English isn't good enough under pressure" comes down to that number. And it is fixable without improving your English at all.
The failure this page prevents
Treating a low Speaking level as a verdict on your English. It usually isn't. It is a verdict on four recognisable behaviours in front of a microphone, each with a different fix. Practising more without knowing which is yours is how people sit the test three times and score the same. Record yourself, then work out which of the four you are doing.
The clock, task by task
The official per-task times. They come from the Speaking study packs, not the test-format page, which gives the task names and the component length but not the timings — which is why almost nobody quotes them.
| Task | Preparation | Speaking |
|---|---|---|
| 1 Giving Advice | 30 s | 90 s |
| 2 Talking about a Personal Experience | 30 s | 60 s |
| 3 Describing a Scene | 30 s | 60 s |
| 4 Making Predictions | 30 s | 60 s |
| 5 Comparing and Persuading | two parts — see below | |
| 6 Dealing with a Difficult Situation | 60 s | 60 s |
| 7 Expressing Opinions | 30 s | 90 s |
| 8 Describing an Unusual Situation | 30 s | 60 s |
Task 5 has a two-part structure that surprises people. You get 60 seconds of preparation with no speaking at all, then a further 60 seconds of preparation and 60 seconds of speaking. Expect to talk at the end of the first minute and the task will throw you.
Speaking is rated on four dimensions: Content/Coherence, Vocabulary, Listenability and Task Fulfillment. Listenability is officially defined as "How easy is it to listen to and understand the response?" Nothing official publishes how they are weighted.
Failure mode 1: deciding while the clock runs
Recognise it: record yourself and listen to the first fifteen seconds. If you are still circling — "well, that's an interesting situation, I think there are many things to consider..." — you have spent a quarter of a sixty-second answer arriving.
What it is: with 30 seconds of preparation you cannot plan an answer, only choose one. Spend the window hunting for the best answer and you arrive at the microphone still deciding — and the deciding gets recorded.
The fix: a first sentence that commits inside the first three seconds — the advice, the choice, the opinion, stated flatly. "I'd tell her to take the job." "The second apartment is better." There is no correct answer to find. The remaining fifty-odd seconds support the thing you already said, which is what Content/Coherence looks at.
Failure mode 2: the silence
Recognise it: your recordings have gaps of three or four seconds while you searched for a word. You will not remember them — they are on the tape.
What it is: there is no examiner in the room. Nobody nods, nobody prompts, nobody fills the gap while you retrieve the word you wanted. Silence is what gets recorded — and Listenability is about how easy the response is to listen to.
The fix: a small set of honest continuation moves while your brain catches up — restate what you just said in different words, give an example, or say the simpler word. "A place where they fix cars" costs you nothing next to four seconds of nothing. Stopping for the perfect word trades a Vocabulary point you might not win for a Listenability problem you definitely created.
Failure mode 3: answering a nearby question
Recognise it: read the prompt again after you record, and check whether you did the verb it asked for. Advise. Predict. Compare and persuade. Decide. Describe a scene to someone who cannot see it.
What it is: the commonest way marks are lost, across all eight tasks. A fluent, well-organised minute about the topic is not the same as doing what the task asked, and Task Fulfillment is a named dimension. Behind it sits the floor rule, which outranks the rest of the rubric:
A response giving insufficient information to assess scores 0 or 1.
Levels 0 and 1 map to no CLB level. That is not reserved for blank recordings — it is what happens when a rater is not given enough of the task to judge.
The fix: spend three of your thirty preparation seconds on the verb, not the topic, and make your first sentence perform it.
Failure mode 4: running out early, or being cut off
Recognise it: time your recordings against the limits in the table.
What it is: the tasks are not all the same length. Tasks 1 and 7 give you 90 seconds. Most give 60. A 45-second answer to a 90-second task is not concise, it is unfinished — half the evidence a rater needed. At the other end, four planned points in a 60-second task means being cut mid-sentence with the strongest one unspoken.
The fix: knowing which two tasks are long. Practise 1 and 7 at 90 seconds until the extra half-minute stops feeling like an abyss.
What is not published
- No weighting for the four dimensions. Anyone ranking them for you is guessing.
- Whether Speaking is AI-scored is unknown. Paragon's FAQ describes an AI-human hybrid system for the Writing Test specifically. Nothing official says either way about Speaking. Official pages also differ on rater counts — the results page says three to five speaking raters, the Score Guidance PDF at least three certified raters.
- Whether templates or memorised material are penalised is unknown. CELPIP publishes no statement on it, whatever you have read elsewhere.
How to practise so it diagnoses
Recording is not the practice. Listening back with a pen is. Mark three things: the second you committed to an answer, the number of silences over two seconds, and whether you did the prompt's verb. Three recordings is enough to see a pattern — one of the four modes will be yours, most likely the same one every time.
What actually costs you
- Arriving at the microphone still choosing your answer.
- Stopping dead to retrieve one word instead of saying the simpler one.
- A good answer to a question next door to the one asked.
- Treating every task as sixty seconds when two of them are ninety.
- Expecting to speak at the end of Task 5's first preparation minute.
- Practising without listening back, which trains the habit instead of finding it.
- Chasing "better English" when the problem is the clock.
Try one
[Try a Speaking task →] — one full task, marked, no card required.