CELPIP Writing and Speaking Scoring Criteria: What It Takes to Reach Level 9

The four dimensions in Writing and Speaking

Writing

Speaking

CELPIP levels 12 to 3, dimension by dimension

Each row is PrepEx’s own summary of CELPIP’s published level descriptors. For CELPIP’s own wording, open the official score comparison chart.

Writing

Speaking

What moves a response up a level

Where CELPIP describes two levels the same way

At a few boundaries, CELPIP’s descriptors use the same words for two neighbouring levels. That line can’t tell you which level a response is, so decide on the dimensions that do change.

The descriptors describe a typical test taker, so a real response can be stronger in one dimension, weaker in another and land on the same level. They aren’t the whole rubric either: raters use an unpublished scale with five performance levels per dimension. The current versions are CELPIP’s May 2026 Level Descriptors PDF and its Score Comparison Chart, with two rated responses and an analysis per level. Search results still surface 2019 and 2020 study packs; the October 2026 Target 9+ packs replace them.

How CELPIP Writing and Speaking are marked

Each Speaking performance, meaning all eight responses, gets three to five raters. Each Writing performance, both tasks, gets four to six. Those numbers come from CELPIP’s test results page. A 2021 CELPIP podcast described two raters per Writing task plus a benchmark rater when they disagree. That fits four to six, but CELPIP doesn’t spell out the arithmetic.

• Independent and blind. Tests are assigned at random online. Raters can’t see each other’s ratings, your name, your other scores or the level you need. A Writing rater sees one task only.
• A benchmark rater for disagreements. If the ratings don’t agree, an experienced benchmark rater who hasn’t seen them rates the performance.
• A level per dimension, from evidence. CELPIP converts those ratings into a CELPIP level with rules set by language experts.
• The whole response. Raters don’t count errors, and one impressive word doesn’t lift a score.
• No accent category. CELPIP staff have called the test “accent agnostic” (Official CELPIP Podcast #2). An accent matters only if it makes something unclear.
• Re-evaluation never lowers a score, and raters aren’t pushed to mark low.

Is CELPIP Writing checked by AI?

Partly. CELPIP’s FAQ calls its Writing scoring an “AI-human hybrid”: AI working alongside its human panel, with the rating team reviewing and verifying every Writing score. The FAQ says nothing about AI in Speaking, and the results page, Score Guide and Performance Standards don’t mention AI. Many prep pages and search summaries still say CELPIP uses no AI. For Writing, CELPIP’s own FAQ says otherwise. It doesn’t say when the system started or how the work is split.

What CELPIP doesn’t publish

Whether the four dimensions count equally, or how the tasks are weighted. For Speaking, CELPIP says each of the eight responses is rated on all four dimensions, so there are many ways to reach the same level. Any percentage split you see online is a guess.

The CLB 9 check: what level 9 shows in each dimension

Many Canada-bound test takers need CLB 7 and aim for CLB 9, which earns more Express Entry points. So the useful question is does this reach 9?, asked one dimension at a time. Here is what a Level 9 response shows, from CELPIP’s descriptors, Level 9 profiles and Target 9+ study packs, and what usually holds a Level 8 back.

Rules that change a mark: word count, templates, off-topic answers and the timer

Word count

Both Writing tasks ask for about 150–200 words, with a counter on screen. CELPIP allows a little room either side, but a former rater said on CELPIP’s podcast that a response far over or under loses marks even when it’s well written. No cut-off or penalty size is published. By PrepEx’s count, CELPIP’s rated samples at levels 8 to 12 run from about 145 to 200 words. Spend spare minutes revising, not adding.

Memorized and templated answers

CELPIP counts memorizing test answers as cheating and can cancel responses that aren’t original or spontaneous. The former rater said memorized general sentences, usually in openings and closings, are obvious to raters and earn no credit. A plan you fill with new sentences is fine.

Off-topic and incomplete answers

Relevance and completeness count, but gaps don’t automatically sink a response. CELPIP’s rated samples include a Level 9 email that skipped the third bullet and spent a paragraph on the wrong day, an unfinished Level 10 survey response, a Level 10 Task 2 whose first idea wandered off topic, and a Level 8 Task 2 that never chose an option. CELPIP doesn’t say what happens to a response that’s off topic throughout.

Running out of time in Speaking

Only what’s recorded is rated. CELPIP’s samples rated 8, 9 and 12 were cut off mid-idea, so being cut off isn’t fatal. Make the main point early.

Sarcasm, humour and tone

Humour isn’t on the rating scale, and sarcasm or rudeness can make the tone seem dismissive or the content off topic. Invented names and details are fine.

Spell-check

The Writing test has an automatic spell-checker. It won’t catch a correctly spelled wrong word or a slip on the punctuation keys, which the former rater listed among common problems. In Speaking, talk at a normal volume: CELPIP’s study pack says too loud or too quiet affects Listenability.

CELPIP marking myths, checked against the evidence

How to mark CELPIP Writing and Speaking with these criteria

CELPIP says plainly that no one in a classroom is a CELPIP rater, the teacher included, so its process can’t be recreated there. It also advises against asking a class to guess a response’s level. Mark against a target instead, with evidence, and call the result an estimate.

1. Fix the target first, usually 7 or 9, and ask for each dimension whether the response reaches it.
2. Read or listen to the whole response first.
3. Note evidence per dimension: one phrase for a yes, one for a no. Use the factors, not an error count.
4. At the identical boundaries (Speaking 7/8 and 8/9, Writing Readability 8/9), decide on the dimensions that change.
5. Keep rule problems in Task Fulfillment. Word count, a missed bullet or an option never chosen shouldn’t also come off Readability.
6. Report an estimate: yes or no per dimension, or a range such as 8–9, plus one next step. Start with what works.
7. Mark Speaking from the audio. CELPIP’s raters never use transcripts, and pauses don’t show up on paper.

CELPIP’s post on using sample responses in the classroom adds: use responses at or below your learners’ level (a class around 7 learns most from 5 to 7), set a 7 beside a 9 and have learners name the differences, and use the Performance Standards for peer editing. Never have learners memorize a model answer.

To mark your own Speaking, CELPIP suggests recording a response, transcribing it and using the Performance Standards to find strengths as well as weaknesses.

For CELPIP instructors: training, the Teacher Support Pack and PrepEx

CELPIP offers two free courses for teachers. Both use rated samples to show how responses are marked.

• Instructor Training Level 1: a recorded course of about five hours for teachers with at least a year of experience, covering the Writing and Speaking criteria and the Performance Standards. Finish within ten days for a certificate and a PDF Instructor Training Manual. Groups start monthly.
• Instructor Training Level 2: about ten hours, self-directed, for teachers with CLB 10 English who have done Level 1. It goes deeper into rating and includes six rated Writing responses, in two sets of three, for class use. The certificate needs full marks on every quiz.
• Teacher Support Pack: a free .zip behind a short form (name, email, country). Its landing page lists ideas for using practice tests in class, with schedules for three course formats, printable copies of the two free tests, the Performance Standards, checklists for students in both skills, a vocabulary journal template and classroom activities.

Prep providers can also join the CELPIP Network (co-branding, referral incentives, discounted study materials in bulk). Details are on CELPIP’s instructors page.

What PrepEx gives a CELPIP class

You can invite students one by one or in a batch, assign any of the 20 CELPIP task families with a due date, open each student’s saved answer with PrepEx’s estimate for every dimension, track progress and send PDF reports that parents can read. There are no full CELPIP tests to assign, teachers can’t change PrepEx’s estimates, and teachers can’t play back students’ Speaking recordings. Access is arranged with PrepEx, usually starting with a short complimentary evaluation for the school. Set up PrepEx for your CELPIP class.