CELPIP Writing Checker
Paste your Task 1 email or Task 2 survey response. In under a minute you get an estimated CELPIP level (and CLB), a rating on each of the four official categories, every error corrected and explained, your strengths and priorities, and a level-10 model answer on the same prompt.
Renovation noise from the apartment upstairs
The tenant in the apartment above yours started a renovation three weeks ago. The work begins at 7 a.m., including on weekends, and you work evening shifts. Write an email to the building manager.
In about 150–200 words, write an email to the building manager. Your email should do the following things:
- Describe the problem and when it happens
- Explain how it is affecting you
- Suggest what the manager could do about it
Draft saved in this browser.
Free: 2 full reports per day per device, no sign-up. Your text is scored by an AI model on the four official categories; the level is an estimate.
- Checking the task and the word count
- Rating the four categories
- Finding and explaining errors
- Writing a model answer
What the checker rates
The same four categories as the two official raters: Content/Coherence (ideas, organisation, examples), Vocabulary (range, precision, natural use), Readability (grammar, sentence structure, spelling, punctuation, paragraphing, connectors) and Task Fulfillment (relevance, completeness, tone, word count). The scoring guide explains each level in each category.
How it keeps the estimate honest
- Every error in the report is quoted from your text and verified by code; anything the model quotes that is not in your text is dropped.
- Frequent verified errors cap the level, however good the ideas are, because Readability is one quarter of the task.
- A missing bullet point or an unstated choice caps the level, because Task Fulfillment is another quarter.
- Level 10 and above is only kept when the model quotes at least four verified passages showing level-9 features (precise wording, controlled complex structures, cohesion, nuance, specific examples) and errors are rare.
The model is gpt-oss-120b (OpenAI), Cloudflare Workers AI. On 28 level-labelled answers, 25 estimates (89%) were within one level of ours (average bias −0.07); details on the scoring page.
What to do with the report
- Read the category ratings first: the lowest one is your priority for the next attempt.
- Fix every error in the corrections list in your own text, then compare your version with the model answer paragraph by paragraph.
- Rewrite and score again. One rewrite scored twice teaches more than three new prompts.
Not sure what to write? Use the practice prompts, read the samples with level breakdowns, or start with the Task 1 and Task 2 guides.
Frequently asked questions
Is the CELPIP writing checker free?
Yes. Every visitor gets two full reports a day, with no account. Each report includes the estimated CELPIP level, ratings on the four official categories, every error with a correction and a rule, strengths, priorities and a level-10 model answer on the same prompt.
What does it check?
It rates Content/Coherence, Vocabulary, Readability and Task Fulfillment, the four categories official raters use, and checks the task itself: whether each bullet point (Task 1) or the choice, reasons and conclusion (Task 2) is there, the tone, and the word count. Every error it reports is verified against your text.
Is the level accurate?
It is an estimate. The checker is built to be conservative: level 10 and above is only given with verified evidence in the text, and frequent errors cap the level. Calibration results on our own level-labelled answers are published on the scoring page.
Is my text stored?
Your text and its report are stored for up to 12 months so that the report link works, then deleted. No account, no email, no name is required. See the privacy page.
Can I check an answer to a prompt from a book or a course?
Yes. Click "I answered a different prompt", paste the situation and bullet points or the survey options, then your answer.