CELPIP Listening format: the questions come after the audio, and it plays once

Search this and you get the same table everywhere: six parts, a question count, a minute count. The table is worth about ninety seconds of your preparation, and after that it tells you nothing you can act on.

Here is the thing that is missing from it, taken from Paragon's own Listening Pro study pack:

"The questions for each Listening part come after the audio." And in Parts 1 to 3, you hear the question and read only the four answer choices — the question itself is never printed on your screen.

Read that twice. In Parts 1 to 3 of CELPIP Listening, the clip ends, and then a voice asks you a question you cannot re-read, about content you cannot replay, and you have thirty seconds. That is not a comprehension task in the ordinary sense. It is a working-memory task with a comprehension task inside it, and almost every piece of advice written about this section is written as if it were the other thing.

This is written from the marking side — we build the engine that scores CELPIP practice on Hilingo, and the listening items are where a scorer's job is easiest to describe honestly, because every answer is right or wrong against a key. The format facts below are all from Paragon's own pages, not from another blog's table.

The format, as Paragon currently publishes it

CELPIP-General and CELPIP-General LS use the identical Listening component: same six parts, same question counts, same time allotment. If you are sitting LS for citizenship, everything on this page applies to you unchanged.

Part	Questions	What it is
1 — Listening to Problem Solving	8	Two speakers; the conversation is split into three separate audio sections, with questions after each
2 — Listening to a Daily Life Conversation	5	Two speakers who know each other, one clip
3 — Listening for Information	6	Two speakers, one with expertise the other wants
4 — Listening to a News Item	5	One speaker, a short radio-style report
5 — Listening to a Discussion	8	Three speakers, delivered as video
6 — Listening to Viewpoints	6	One speaker, the longest clip, a prepared talk

38 questions. 46 to 55 minutes allotted. (CELPIP Test Format and the July 2026 Guidebook for Test Takers.)

Two notes on that table, because this is where competing pages go wrong.

There is no per-part timing on Paragon's current format page. It publishes one number for the whole component. Paragon's Listening Pro study pack does give approximate per-part timings — roughly 8, 5, 6, 5, 9 and 8 minutes — and they sum to about 41, with the balance going to instructions. That study pack is a 2022 document, so treat those as the shape of the section rather than a stopwatch. If you have seen a per-part timing table presented as current official fact, it was copied from somewhere like that and not checked. One widely-shared page currently lists Part 5 as having 2 questions. It has 8, in both Paragon's current format page and the July 2026 guidebook — a third of a part's marks, missing from a table people are planning around.

Some of the questions you answer will not count. Paragon states that unscored items used for test development can appear anywhere in Listening and Reading, in the same format as scored items, and that you cannot tell them apart. The time they take is inside the total. The practical consequence is the only one available: there is no such thing as a question worth skipping, because you cannot know which one you just skipped.

The mechanic that should reorganise your preparation

Four facts, all from Paragon's material, and they compound:

1. Every audio clip is played one time only. Paragon's own tips and resources page says it plainly: "Each audio file can be heard only once." Clips start automatically and cannot be paused.
2. The questions arrive after the audio, not alongside it.
3. In Parts 1 to 3 the question is spoken, one at a time, with only the four answer choices on screen, and you must answer them in the order presented.
4. You cannot return to a previous part once its time is up.

Now put that next to IELTS, which is the comparison most candidates are carrying in their heads. IELTS listening also plays once — that part is the same. What is not the same is that in IELTS the questions sit in front of you while the recording runs, so the audio is a search task: you know the target before it arrives, and you can re-read the question three times while waiting for it.

CELPIP removes that. For most of the section you do not know what you will be asked until after the thing has been said. So you cannot listen selectively for a target, because there is no target yet. You have to hold a structured version of the whole clip and then be interrogated about it.

That is a genuinely different skill, and it is why candidates with good English and IELTS habits sometimes underperform here. They are running a search strategy on a test that does not permit one.

Two more facts that follow from the same design:

The answer will not repeat the words you heard. Paragon's study pack is explicit: you should never expect the correct choice, or any choice, to be worded as it was in the clip. Every part of the section paraphrases. Its own example is a speaker saying she was too tired from a night shift to stay awake during a lecture, and the correct option reading that she almost fell asleep in class. Keyword-matching does not merely underperform on CELPIP; it is the mechanism the distractors are built to punish.

Parts get harder as you go. Paragon states the audio and questions increase in difficulty from Part 1 to Part 6, and describes Part 6 as a formal, prepared talk of around three minutes using low-frequency and specialised language. So your energy budget is upside down by default — the hardest listening arrives when you are most tired, roughly forty minutes into the section, in a test that will not pause for any reason.

What the questions are actually asking

Paragon names three question types, and they run across the whole section:

Type	What it wants	How it is usually worded
General meaning	An idea assembled from a large stretch — several sentences, more than one speaker, or the whole clip	"The two speakers are talking about…", "The three speakers all agree that…"
Specific information	One retrievable fact: a name, a date, a place, an amount	"The man offers the woman…", "The woman's job is…"
Inference	A conclusion that was never stated — a feeling, an attitude, what happens next	"The man sounds ___ about next week's schedule", "Ms. Wilson will probably…"

This taxonomy is more useful than it looks, because the three types fail for different reasons and therefore have different fixes.

Specific-information questions fail when the fact was not written down. That is a note-taking problem, and it is the cheapest one to fix.

General-meaning questions fail when your notes are a list of details with no shape — you captured everything and understood nothing. That is a structure problem.

Inference questions fail when you recorded what was said but not how. Paragon points out that speakers rarely state feelings directly; you get "I can't believe my bus is late again" rather than "I am frustrated." If your notes carry the content but not the attitude, an inference question has nothing to work from, and you will end up guessing between two options that both look factually reasonable.

The clean way to see this is to mark your own practice by type. If your inference questions are at 50 per cent while your specific-information questions sit at 90, the fix is not more listening — it is changing what your pen records.

A note page that follows from the constraint

Paragon supplies paper and a pen at the test centre, and says you can ask for more. Generic advice stops there. If the questions come after the audio and arrive roughly in the order the answers appeared — which Paragon also states — then your page is not a memory aid. It is the index the questions will be looked up in. Build it that way.

Before the clip starts. You get an introductory statement setting the scene. Use it to draw the page, not to relax. Part 1 gets three blocks, because the conversation arrives in three sections with questions after each. Part 5 gets three columns, labelled by where each speaker sits on screen as they are introduced. Parts 4 and 6 get a single column with room down the left for names.

During the clip, record only what you cannot reconstruct. Full sentences are the commonest error — you cannot write at speech pace and listen at the same time, so a candidate writing sentences is a candidate who stopped listening. What you cannot rebuild afterwards is precise: last names, numbers, times, places, and the direction of each opinion. A plus, a minus or a question mark beside a speaker costs you a quarter of a second and answers inference questions that an hour of vocabulary study will not.

Keep it strictly in order down the page. Because the questions track the audio order, a sequential page turns a memory search into a scan. A page of scattered marginalia does not.

In Parts 1 to 3, write the question's keyword as you hear it. The question is spoken once. The moment it lands, put down the two or three words that carry it, then look at the choices. Paragon's own after-listening strategy for these parts is to note keywords from the questions, and the reason is exactly this: if you go straight to reading four options, you will have lost the question by the third one.

Answer in your own words before you read the options. This is the single highest-value habit in the section, and it falls directly out of the paraphrase rule. Say the answer silently to yourself first, then look for the option that means it. If you read the four options cold, you are comparing them against a half-remembered clip, and the distractor that reuses the speaker's actual words will feel right. It is designed to.

In Parts 4 to 6, work out of order deliberately. These parts use sentence completion, put every question on one screen, and let you answer in any sequence with one shared timer. Take the ones you are certain of first and bank them, then spend the remainder on the rest. Paragon's own Part 6 guidance is to watch the clock and answer every question, guessing where you have to — which is the right instruction for any section where an unanswered item and a wrong item are worth the same.

When you miss something, drop it. Paragon says this too, and it is worth more than it sounds. Spending four seconds recovering a lost word costs you the sentence that carried the next answer. The candidates who lose whole parts usually lose them to one missed phrase they refused to let go of.

What your listening level has to be

CELPIP levels map one to one onto the Canadian Language Benchmarks — a CELPIP 7 is CLB 7 — which is confirmed in Paragon's own guidebook chart alongside the CEFR equivalents. It also means a listening level is directly readable as immigration points, with no conversion step for you to get wrong.

Express Entry scores language per ability, on that ability's own benchmark, while your eligibility is decided by your lowest skill. So listening does two separate jobs on your profile:

CELPIP Listening	CLB	CRS language points from this one ability	What it does for you
10–12	10+	34	Ceiling. A 12 earns exactly what a 10 does
9	9	31	The level most competitive candidates target
8	8	23
7	7	17	Federal Skilled Worker minimum, in all four skills
6	6	9
5	5	6	Federal Skilled Trades minimum for listening and speaking
4	4	6	Canadian citizenship, via CELPIP-General LS

Points shown are the single-applicant grid; the with-a-spouse grid is slightly lower. Programme minimums are IRCC's and change — confirm against IRCC's language requirements before you file. The full CELPIP-to-IELTS-to-PTE-Core conversion lives in our CELPIP score chart.

Two things fall out of that table that are worth more than the table.

The step from 8 to 9 is worth 8 points; the step from 9 to 10 is worth 3. If listening is your weakest skill and it is sitting at 8, that is where your study hours belong. If it is at 10, they do not.

You cannot convert a raw listening score to a level yourself, and neither can anyone else. Paragon does not publish a raw-to-level table, because forms are scaled so a level means the same thing across different sittings. Any practice product telling you that 33 out of 38 "is a 9" has invented it. What a raw score honestly gives you is a direction and a per-part profile, which is the thing worth tracking anyway.

While we are here: a page currently ranking for this topic states that CELPIP 7 corresponds to CLB 9. It does not — CELPIP 7 is CLB 7, and the same page also asserts a CELPIP-to-IELTS equivalence that neither test owner publishes. If you are planning a PR application, that single error is the difference between booking a test you are ready for and booking one you are not.

Honest about the week before

Some of this section responds to a few days of work. Most of it does not, and pretending otherwise wastes the days you have.

What genuinely moves inside a week:

• Your note page. Layout, abbreviations, the plus-minus-question-mark for attitude. This is a habit, and habits install in days.
• Answering before you read the options. Also a habit, and it directly counters how the distractors are built.
• Clock discipline in Parts 4 to 6. Bank the certain ones first, answer everything, leave nothing blank.
• Letting go of a missed word. One rehearsed decision, worth several questions.
• Knowing the shape of each part, so Part 5's three speakers and Part 6's three-minute talk are not a surprise at minute forty.

What does not move inside a week, and you should stop trying:

• Processing speed at natural Canadian conversational pace. Paragon states all speakers are native English speakers with Canadian accents. Comfort at that pace is built from months of exposure, not a weekend.
• Vocabulary breadth for Part 6, which Paragon describes as using low-frequency and specialised language. You cannot close that gap in a week. You can practise inferring from context, which is the skill that survives not knowing the word.
• Sustained concentration across 50 minutes. If you fade at minute thirty-five in practice, you will fade at minute thirty-five on test day. The fix is full-length sittings over weeks, not a technique.

If your test is in five days, spend them on the first list. If it is in five weeks, the second list is where the level is.

Why practising this alone stalls

Listening is the section self-study helps least, and the reason is structural rather than motivational.

You finish a set, you get a score, and you look at question 14 — which you got wrong. You now know that C was right and you chose B. What you do not know is the only thing that matters: was B wrong because you misheard the clip, because you matched a keyword the speaker actually said, or because you had the meaning and picked the option that paraphrased it less well? Those are three different problems with three different fixes, and a score of 27 out of 38 distinguishes none of them. Replaying the clip does not help either, because the second hearing is not the test — you already know the answer.

This is why every objective answer in a Hilingo CELPIP mock comes back with the reason it was wrong plus the evidence sentence quoted from the transcript. You see the line the answer came from, next to what you picked and why it does not follow from that line. That converts a wrong answer into a diagnosis — misheard, keyword-trapped, or paraphrase-missed — which is the distinction you cannot make for yourself. Transcripts open after the attempt, never during it, because a transcript during the attempt is a different test.

Your first full scored mock is free, with no card. There are 29 approved CELPIP-General mocks and 28 CELPIP-General LS mocks, each running under the real section timers, so a full sitting feels like a full sitting including the Part 6 fatigue. The AI teacher answers questions from your own result rather than in the abstract, the study centre points at whichever skill is currently capping your profile, and any question can be translated into one of 50 languages while you are learning the format.

One caveat, stated rather than buried: Paragon does not publish how raw listening marks become a level, so we do not claim to reproduce it. Any scorer of a scaled test is estimating a conversion it cannot see. Our per-item marking is exact — the key is the key — and the level is an estimate. Use the per-part accuracy to find the leak; use the level as a direction.

Take a free scored mock

• How CELPIP is scored — the 12-level scale, the four dimensions, and what a 9 needs
• CELPIP score chart — CELPIP level to CLB, IELTS and PTE Core, both directions
• CELPIP vocabulary — why recognition has to be instant when the audio plays once
• CELPIP vs IELTS vs PTE Core — which test is hardest in your weakest skill
• PTE Core vs CELPIP — how the two Canadian PR routes differ in marking

Frequently asked questions

How long is the CELPIP Listening test?

46 to 55 minutes for the whole component, which is what Paragon publishes on its current test format page. It is the same allotment on CELPIP-General and CELPIP-General LS. Per-part timings are not on that page; Paragon's Listening Pro study pack gives approximate ones summing to about 41 minutes, with the balance going to instructions, so treat any part-by-part table you see elsewhere as approximate rather than official.

How many questions are in CELPIP Listening?

38, across six parts: 8, 5, 6, 5, 8 and 6. Some questions in the section may be unscored items used for test development — Paragon says these can appear anywhere in Listening and Reading and are indistinguishable from scored ones, which is the reason to answer every single question properly.

Can you replay the audio in CELPIP Listening?

No. Paragon states that each audio file can be heard only once. Clips start automatically, cannot be paused, and once a part's time is up you cannot return to it. That single constraint is why note-taking matters more on CELPIP than on tests where the question paper sits in front of you during the recording.

Do you see the questions before the audio in CELPIP?

No, and this is the biggest practical difference from IELTS. The questions for each part come after the audio. In Parts 1 to 3 the question is also spoken rather than printed, with only the four answer choices on screen, so you cannot re-read it. In Parts 4 to 6 the questions are sentence completions that all appear together on one screen and can be answered in any order.

Can you take notes during the CELPIP Listening test?

Yes. Paper and a pen are supplied at the test centre and you can ask for more, and the paper is handed back at the end. You cannot bring your own. Since the questions arrive after the clip and roughly in the order the answers appeared in it, notes kept in sequence work as an index rather than a memory aid — which is what makes them worth practising rather than improvising.

What CELPIP Listening score do I need for Canadian PR?

CELPIP levels equal CLB levels, so the Federal Skilled Worker minimum of CLB 7 means a CELPIP 7 in listening — and in the other three skills, since IRCC reads your lowest. Most competitive Express Entry candidates aim for 9, because CLB 9 is where the points step up sharply. Canadian citizenship is CLB 4 in listening and speaking, which is what CELPIP-General LS is for.

Why do I keep choosing the wrong answer when I understood the audio?

Almost always paraphrase. Paragon says outright that you should never expect the correct choice to use the same words as the clip, so the option that repeats what the speaker actually said is usually the trap. The fix is a habit rather than more study: answer the question in your own words before you read any of the four options, then look for the one that means what you just said.