CELPIP Listening Part 3 (Listening for Information): Two Speakers, Two and a Half Minutes, Six Questions

What Part 3 actually is

Part 3 is a conversation, not a lecture. A lot of preparation material treats "Listening for Information" as one person reading out facts while you copy them down, and people practise accordingly — pen ready, waiting for numbers.

The official study material describes something else: two speakers, in a single audio clip of about two to two and a half minutes, followed by six questions. That is the whole task.

That difference is not cosmetic. Information that comes out of two people arrives in a different shape from information that comes out of one. It gets asked for, misremembered, hedged, checked against a screen, and revised — inside the clip, out loud. The answer to a Part 3 question is very often the second thing said about a detail, not the first.

And the clip plays once. Official sources are flat about this: "All audio clips are played one time only." There is no rewind to check which version you wrote down.

The failure: first-mention lock-in

Here is how the marks actually go.

You hear a date. You write it. Your pen moving is a small act of closure — the detail is now handled, and your attention moves on to the next thing. Eleven seconds later one speaker says "sorry, that's the wrong line" and gives a different date. You half-register it. Six questions later you answer from your notes.

Call it first-mention lock-in: you commit to a fact at the moment you first hear it, and the conversation then edits that fact while your attention is elsewhere.

It is a listening failure that punishes attentive people specifically. The candidate who writes nothing down often does better on these items, because nothing got frozen. Part 3 rewards holding detail in a provisional state until the conversation has finished with it.

Worked example (original AlmiWorld practice item)

Devon: Hi — I registered online for the intermediate pottery class, and I just wanted to confirm the start date.

Maya: Sure, let me pull it up. Intermediate pottery... okay, I have you on the Thursday evening group, starting the fourth of October.

Devon: The fourth. Good, that matches my confirmation email.

Maya: Hold on — sorry, that's the beginners' group on my screen. The beginners start on the fourth. Yours is the Thursday intermediate, and that one got pushed back a week when the kiln replacement was scheduled. You start on the eleventh.

Devon: The eleventh. And it's still eight weeks?

Maya: Eight sessions, finishing on the twenty-ninth of November. There's no class the week of the twenty-second — the studio's closed for the long weekend — so on the calendar it stretches across nine weeks.

Question: When does Devon's class begin?

• A. 4 October
• B. 11 October
• C. 22 November
• D. 29 November

The answer is B.

Notice what every wrong option has in common: it is a real date, said out loud, correctly heard. None of them is a mishearing. This is what distractors in Part 3 are actually made of.

A — 4 October. The pre-correction value. It is the first date in the clip, it is said twice, and Devon confirms it against his email — so it gets triple reinforcement before Maya withdraws it. It is also a genuine start date, just for the beginners' group. If you wrote "Oct 4" and moved on, this option will look right and feel right.

C — 22 November. Accurate and irrelevant. That is the week with no class. It is in the clip because holiday closures are exactly the kind of detail these conversations carry, and because a listener tracking "weeks" rather than "start date" will have it circled in their notes.

D — 29 November. The end date. Wrong question, right conversation. This one catches people who retained the last number they heard about the course rather than the number attached to begin.

The item is not testing whether you can hear "the eleventh". It is testing whether you were still listening to the start date after you thought you had it.

The shape that works

1. Read the options before you commit. In the Listening test you read the answer choices — they are not read to you ("Test takers will be required to read answer choices in the Listening Test"). Your eyes and ears are doing different jobs, so give the options a real look rather than scanning for the digits you wrote.
2. Write the detail, leave room beside it. Paper and pen are provided at the test centre. Note start Oct 4 with a gap after it, not a boxed final answer. The gap is where → 11 goes.
3. Track who said it. Two speakers means one of them usually has the authority — the one with the screen, the schedule, the file. When versions conflict, the informed speaker wins.
4. Let numbers repeat before you settle. Anything genuinely important in a two-person exchange gets said more than once, because that is how people confirm things to each other. A number said only once and never echoed is often scenery.
5. Answer at the pace the test sets. The study material gives about 30 seconds to hear and answer each question in Parts 1–3, and the screen moves on by itself. Six questions here. Do not spend your thirty seconds reconstructing the clip.

Holding a detail loosely

This is the specific move Part 3 asks for, and it is trainable.

The habit to build is a two-stage grip: record, then wait for release. A detail is not finished when you hear it — it is finished when the conversation moves past it without touching it again. Until then it stays open.

In practice, that means learning the sound of a fact being revised. These phrases are your cue to un-commit:

• Hold on / wait / sorry —
• Actually, that's the...
• No, that's the other one.
• Let me check that. / Let me pull it up.
• It was, but they changed it.
• That's what it says here, but...
• I think so — I'd have to confirm.
• It used to be X. Now it's Y.

And the phrases that tell you a detail is now settled, and safe to close:

• Right, so the eleventh. (the other speaker repeating it back)
• That's confirmed.
• Same as last time.

Train on the hedges too. "I think", "should be", "as far as I know" mark a value that the speaker has not yet stood behind. A hedged number is an invitation to keep listening.

Notice, in the example above, the second small revision: eight sessions but nine weeks on the calendar. Nobody flagged it as a correction. A question asking how long the course runs has two defensible-sounding answers sitting right there, and only one of them answers what was asked.

What actually costs you the mark

• Writing a number down and treating the pen stroke as the end of that fact.
• Note-taking so dense that transcribing eats the attention you needed for the correction.
• Choosing the option that matches your notes instead of the option that answers the question.
• Answering from the loudest or most repeated detail — repetition in this part is often the setup for the correction, not proof of the answer.
• Reading the options at speed and picking the first one you recognise. They will all be recognisable.
• Carrying an unresolved question forward. The clip plays once and you cannot return to a previous part — a half-answered item is a spent item. Choose, and let it go.

Try one

[Try Listening Part 3 →] — one full task, marked, no card required.