CELPIP Writing Score: Levels 3–12, CLB Chart and Rating Categories

CELPIP level to CLB and CEFR chart

CELPIP level	CLB level	CEFR	Official descriptor
12	12	C2	Expert proficiency in high-stakes social, educational, or workplace contexts
11	11	C1	Advanced proficiency in high-stakes social, educational, or workplace contexts
10	10	C1	Highly effective proficiency in some high-stakes social, educational, or workplace contexts
9	9	B2	Effective proficiency in some high-stakes social, educational, or workplace contexts
8	8	B2	Good proficiency in more demanding social, educational, or workplace contexts
7	7	B2	Adequate proficiency in somewhat demanding social, educational, or workplace contexts
6	6	B1	Developing proficiency in everyday social, educational, or workplace contexts
5	5	B1	Acquiring proficiency in everyday social, educational, or workplace contexts
4	4	A2	Adequate proficiency for daily life activities
3	3	A2	Some proficiency in limited contexts
M	—	—	Minimal proficiency: insufficient information to assess at level 3 or above

Source: celpip.ca, "Test Format & Scoring", read 20 October 2026. Scores are valid for two years from the date of the score report.

Which writing level do you need?

Level 4: Minimum for Canadian citizenship (CLB 4 in speaking and listening; writing is not tested for citizenship).
Level 5: Federal Skilled Trades Program minimum for writing (CLB 5).
Level 7: Canadian Experience Class (TEER 0 or 1) and Federal Skilled Worker minimum (CLB 7 in each ability).
Level 9: CLB 9 in all four abilities unlocks the higher skill-transferability points in Express Entry.
Level 10: CLB 10 gives the maximum first-language points in the Comprehensive Ranking System.

Requirements change and differ by stream; confirm the current CLB requirement for your program on the IRCC website before you book a test. To check all four of your abilities at once, with the CRS points and the program minimums they meet, use the CELPIP to CLB calculator.

How the raters score a task

Each of the two writing tasks is rated by two trained raters. A rater gives four category ratings, one for each of the categories below, and those are combined into a task score. Task 1 and Task 2 are each worth 50 percent of the writing score. The official performance standards list these factors for each category:

Category	Factors (from the official standards)
Content / Coherence	number of ideas, quality of ideas, organization of ideas, examples and supporting details (how well are the ideas organized and developed?)
Vocabulary	word choice, suitable use of words and phrases, range of words and phrases, precision and accuracy (what is the range of the vocabulary and is it used naturally?)
Readability	format and paragraphing, connectors and transitions, grammar and sentence structure, spelling and punctuation (how easy is it to read and understand the response?)
Task Fulfillment	relevance, completeness, tone, word count (how well did the test taker follow the instructions, and is the tone appropriate?)

Two consequences follow. First, the four categories carry equal weight, so a missing bullet point (Task Fulfillment) costs as much as frequent grammar errors (Readability). Second, each category is rated on the whole text: one brilliant sentence does not lift Vocabulary if the rest is repetitive, and one serious error does not sink Readability if the rest is clean.

What each level looks like in each category

The official standards describe performance in prose; the table below is our condensed reading of them, written to match the samples on this site. Use it to place your own writing before you run the scorer.

Level	Content/Coherence	Vocabulary	Readability	Task Fulfillment
Level 10–12	Ideas are developed with specific, convincing support; nuance (concession, comparison, consequence); organisation feels effortless.	Precise and often idiomatic; natural collocations; no reaching for words.	Complex structures with full control; errors rare and never distracting; clean paragraphing and transitions.	Everything asked for, in a tone that fits the reader exactly; length in range.
Level 9	Well organised; each reason developed with a relevant, specific example or detail.	Good range used mostly precisely; some natural collocations; occasional slight imprecision.	Varied sentence structures with good control; few, minor errors.	All parts fully addressed; tone right throughout.
Level 7–8	Clear paragraphs; ideas developed with some support; may stay general.	Adequate for the task; some repetition or imprecise choices.	Mix of simple and complex sentences; errors noticeable but rarely obscure meaning.	Every part addressed; tone generally appropriate.
Level 5–6	Ideas stated rather than developed; basic organisation.	Limited, repetitive; some wrong word choices.	Mainly simple and compound sentences; errors frequent, meaning usually clear.	Task addressed briefly; tone may slip.
Level 3–4	Few ideas, little organisation.	Very limited.	Short simple sentences; frequent basic errors that often obscure meaning.	Task only partly addressed.

Two patterns from the samples are worth remembering. A correct but simple text (short sentences, generic ideas, basic connectors) stays at 7 even with few errors, because Content and Vocabulary do not reach 9. And a text with developed ideas but errors in most sentences is held at 6 by Readability. The quickest route from 6 to 8 is accuracy; the route from 8 to 9 is specific support.

How the AI scorer estimates your level

The scorer on this site is not an official rater and does not see your whole test. It works in three steps, built to be conservative rather than flattering:

A language model (gpt-oss-120b on Cloudflare Workers AI) rates the four categories on the 3–12 scale, picks an overall level, lists the errors with the exact passage and a rule, and quotes the passages that show level-9 features (precise wording, controlled complex structures, cohesion, nuance, specific examples).
Code verifies every quoted error and every quoted piece of evidence against your text, and discards anything that is not actually there.
Code applies the rules: frequent verified errors cap the level (at 8, 6 or 4 depending on the density); a missing part of the task caps it at 7 (two missing parts: 5); a very short text caps it at 8 or 6; and level 10 or above is only kept when at least four verified pieces of evidence from at least two categories are present and errors are rare.

Calibration. We wrote 28 original answers at known levels (the sample answers on this site plus eight more at levels 4, 6, 8 and 9) and ran them through the scorer in its launch configuration on 20 October 2026: 25 of 28 estimates (89%) were within one level of ours, 7 were exact, and the average bias was −0.07 levels (mean absolute error 0.86). All three misses were high answers: two level-11 answers and one level-10 answer estimated at 8–9. Both level-4 answers were estimated at 4 and the four level-9 answers at 8–9. The full log is on the about page. In short: levels 10 and above are estimated conservatively by design, so a 9 from the scorer may be a 10 or 11 on the real test; a 6 from the scorer is unlikely to be an 8.

If you have an official score report, you can tell us your official level and the level the scorer gave; we use these pairs to recalibrate and never publish them.

Frequently asked questions

Is CELPIP writing level the same as CLB?

Yes. CELPIP levels 3 to 12 correspond directly to CLB 3 to 12. A writing level 9 on your score report is CLB 9 for immigration purposes.

How are the two writing tasks combined?

Each task is worth 50 percent of the writing score. Two raters rate each task on four categories, the category ratings are combined into a task score, and the two task scores give the writing level. The exact arithmetic is not published.

Can a word count outside 150–200 fail the task?

No automatic fail. Word count is one factor inside Task Fulfillment. Far too short means there is less to reward in every category; slightly over is not penalised in itself.

What is level M?

M stands for minimal proficiency: the response gives the raters too little to place it at level 3 or above (for example an empty, off-topic or very short answer).

How is the AI estimate on this site produced?

A language model rates the same four categories and quotes evidence from your text; code then verifies every quoted error and every piece of evidence against the text, caps the level when errors are frequent or part of the task is missing, and only allows level 10 and above with verified evidence. The calibration results are on this page.