CELPIP guides: every task, explained with worked examples

The CELPIP-to-CLB chart is official and boring. The raw-score table you actually came looking for is reverse-engineered by prep companies, and they disagree with each other.

There is no single CELPIP score you need. Floors run from CLB 4 to CLB 7, each program sets its own, and most of them judge your weakest skill — not your average.

One version is accepted for immigration and citizenship. The other is citizenship only — and booking it for PR is not the only expensive way to get this wrong.

Two official rules decide whether you sit the test at all — the clock at the door and the single item you are allowed to carry through it. Both are published, and both are unforgiving.

Two ways to challenge a CELPIP result — sit it again, or pay for a re-mark whose fee comes back if Speaking or Writing goes up. Plus four things Paragon does not publish at all.

The cheaper CELPIP costs about a third less and IRCC accepts it for citizenship only. Here are the published fees, the date they were checked, and the mistakes that make people pay twice.

Half of CELPIP is counted by machine. The other half is judged — and on exactly who or what does the judging, three official Paragon sources give three different answers.

CLB is not a CELPIP invention — it is Canada's own scale, and your result is four separate levels. The number people average is the one that does not exist.

Paragon gives away the per-task Speaking times, the rating dimensions and the accent rule — and almost nobody links to it. A curated index, with what each file can't tell you.

Most study plans hand you a calendar. Start with a measurement instead — each component is levelled separately, and the skill that decides your result is your lowest, not your average.

Both test makers say neither test is easier. What actually decides it: IELTS Speaking is a face-to-face interview, CELPIP Speaking is a microphone and a countdown.

The two test makers jointly commissioned a real concordance study and published it in 2025. Here is the table, who built it, and the four warnings its own authors attached to it.

The audio plays once and the answer options are printed, not spoken. Every mark in CELPIP Listening turns on that pair of facts. Here are all six parts, and what each one is really testing.

Part 1 isn't one long conversation — it's three short chunks of audio with questions in between. Once you know that, your notes only have to survive ninety seconds, and the traps get easier to see.

Five questions on one short, friendly conversation — and the questions turn on who said what, which day, and what was agreed. Here is why the easy part costs marks, with a worked item.

Part 3 is not a lecture — it is two people, and one of them keeps correcting the other. The tested detail is usually the second version, not the first. Here is how to hear it.

Parts 1-3 hand you every fact twice — one speaker says it, the other confirms it. A news reader says it once and moves on. That single structural change is why scores drop here.

Part 5 is the only Listening part that plays a video, and the only one with three speakers. The questions rarely ask what was said — they ask who said it. Here is how to track that.

The last Listening task is one person talking for about three minutes — and their position moves while they talk. Here is how to hear the turns, and why the option that quotes them accurately is often wrong.

Paper and pen are provided and the audio plays once. Every word you write is a word you don't hear. Here is the notation, the side-by-side, and the things worth leaving out.

Four different problems produce the same disappointing Listening result, and each needs a different fix. Work out which one is yours before you sit another practice test.

The wrong options in CELPIP Listening are engineered, not random. Learn the four shapes they come in — and the one question that separates each of them from the right answer.

Four original CELPIP Listening items — Parts 1, 4, 5 and 6 — with the transcript, every option, and the reasoning for all four, including the three you did not choose.

Three of the four CELPIP Reading parts hide a second task inside them. Thirteen of the 38 marks aren't comprehension questions at all — and nobody practises them separately.

Part 1 hands you eleven marks and most people prepare for six. The five blanks in the reply are not checking whether your sentence is well formed — they are checking something else entirely.

Part 2 gives you no passage — the main text is a diagram, and the wrong answers are all true. Here is what the task really asks, with a full worked item.

Part 3 is the only Reading part where "this was never said" is a correct answer. Under time pressure that option feels like giving up — which is exactly how the marks go.

Ten of the 38 Reading marks, and the study pages contradict each other on what's even on the screen. The real format, plus the distractor logic in multi-opinion articles that nobody teaches.

You can read the passage. You still lose marks. Four different failures hide behind one Reading score — here is how to tell from your own practice which one is costing you, and what fixes each.

Thirteen of the 38 Reading marks are blanks in a second short text, and they look like a grammar exercise. They aren't: every option fits the sentence, only one fits the passage.

Four original CELPIP Reading items, one per part, with the passage, every option, and the reasoning for all four — including the near-miss you were probably about to choose.

One missed instruction can score the whole task 0 or 1, however good your English is. Here is what CELPIP's Task 1 email actually asks for, and a complete worked answer with the reasoning shown.

Task 2 gives you two options and asks which you prefer. Neither is correct. What is marked is how you defend the one you chose — and the commonest failure is refusing to choose.

A response that gives insufficient information to assess scores 0 or 1. Miss one bullet of the prompt and your beautiful email isn't an 8 with a deduction — it's at the bottom of the scale.

Four ways people lose CELPIP Writing marks that better grammar cannot buy back — and how to spot each one in answers you have already written, using only a clock and a count.

Paragon names the four Writing dimensions in public. Most preparation goes into two of them — and those two cannot rescue the one with a 0-or-1 floor underneath it.

A full 184-word Task 1 email, and the habit that separates a mid answer from a strong one: counting the demands in the bullets before you write a sentence.

Task 2 asks which option you prefer. There is no correct one — and the fastest way to lose marks is to weigh both fairly and never choose. One full response, read against the rubric.

There is no examiner. You talk to a microphone, and 15 minutes covers all eight tasks — under two minutes each, including your thinking time. Here is every task, what it asks, and where each one is lost.

Task 1 gives you one named person with one specific problem. The commonest failure is advice so general it would suit anyone. Here is what separates a strong answer, with a full worked response.

Task 2 asks you to talk about something that happened to you. Nobody is checking whether it did. What is marked is whether the story goes somewhere — and most answers are events with no point.

Task 3 shows you an image and asks you to describe it. Almost everyone names things — a man, a table, a dog. What is marked is whether someone who cannot see the picture can picture it.

Task 4 shows you the same image as Task 3 and asks what happens next. Almost everyone describes it again. The task rewards committing to a future that has not happened yet.

Task 5 asks you to compare two options and then persuade a specific person to take one. Most answers compare beautifully and never persuade — and the person you are persuading was described to you for a reason.

Task 6 puts you in an awkward situation with two uncomfortable choices and asks you to tell someone what you have decided. Politeness is not the skill being tested. Choosing, and being understood, is.

Task 7 asks what you think and why. Most answers state a view and then say it again in different words. What is marked is the reasoning underneath — and a reason nobody could disagree with is not a reason.

Task 8 shows you a strange image and asks you to describe it to someone who cannot see it. The instruction people skip is the one in the prompt: the listener is blind to the picture, and the picture is deliberately odd.

Paragon says raters do not consider your accent. Online that sentence quietly turns into "pronunciation doesn't matter" — which is false, and the rubric says exactly why.

Six of the eight Speaking tasks give you thirty seconds to prepare. What people call nerves is usually that number — and it is fixable without better English.

A complete 90-second Task 1 answer, written out and read against the four published rating dimensions — plus why you cannot reuse a single line of it.

A sixty-second story about forgotten laundry, written out in full and read against CELPIP's four published Speaking dimensions — and why the boring memory teaches you more than the dramatic one.

A complete 60-second answer to a Describing a Scene picture, written out and read against the four published dimensions — and why a perfect list of nouns is still a weak answer.

A complete 60-second Making Predictions answer in which not one sentence describes the picture — plus the reason most Task 4 answers are really Task 3 answers wearing a future tense.

Task 5 hands you two options and one person. The marks are in the person — and most answers compare the options beautifully, then forget to persuade anybody of anything.

Task 6 gives you a full minute to prepare — twice what most tasks get. That minute is not for wording. It is for choosing which person you are going to disappoint.

One test separates a real Task 7 answer from ninety seconds of noise: could a reasonable person disagree with what you just said? A full 90-second answer, read against the four dimensions.

The listener cannot see the picture. One full 60-second answer to an odd photograph, written out, then read against the four published rating dimensions — and no template.