TEF Canada writing practice: AI correction and an estimated NCLC level

Yes: this page, the TEF Canada tool of TCF Canada Coach. You submit a section A or section B text and get an estimated NCLC and CEFR level, the matching score band out of 699 and your explained corrections. It is an independent practice tool: the level is an estimate, and only CCI Paris Île-de-France issues official results.

Here, the preview is free 2 times a day per device: estimated level, score band and your first 2 corrections. The first preview asks for nothing, the second for an email address. The full report (all corrections, a score per criterion, a model answer) is paid, from US$9.90 for 10 corrections. Other free resources are listed in our comparison of TEF Canada preparation websites.

An AI tool can give you an estimate based on the criteria CCI publishes, but it does not replace the examiners. At the TEF Canada, two examiners assess each paper independently, with both sections in front of them; our tool assesses one section at a time. Above B2 the estimate is conservative: a C1 text is often estimated at B2. Use it to find your mistakes and track your progress, not to predict your exact result.

The grader on this page handles both sections, one after the other: pick A or B in the list. For a test taken since 11 December 2023, NCLC 7 in writing is 428–471 out of 699. For an estimate on both sections together, against the clock, take our writing mock test. Either way it is an estimate: only your official certificate counts.

We can only speak for ours, and we publish no accuracy figure: we do not yet have enough official results from users to compare. Treat it as an estimate for practice, conservative above B2, where a C1 text is often estimated at B2. Watch the trend over several texts and the mistakes that keep coming back. Once you get your official result, you can report it from your report page (optional, no name asked): it helps us improve the estimate.

Yes for the criteria, no for the marks: the full report rates your text on the seven criteria from CCI’s sample tests, but each mark out of 20 is our AI’s estimate, not a CCI mark. The seven are fit to the situation, text organisation, clarity of the story or argument, developing and nuancing ideas, vocabulary, syntax (tenses and moods) and spelling. These marks and their comments are paid, from US$9.90 for 10 corrections, like all your corrections (up to 12), the point-by-point check of the instructions, your strengths and 3 priorities, and a C1 model answer. The free preview gives the estimated level, the score band, the word count and your first 2 corrections. The overall level is not an average of the seven marks: the AI judges the level, then our code checks it.

Yes, both, and the free preview shows the result too. Word count: our code counts your words as you type and on the report, with the rule France Éducation international publishes for the TCF (whatever stands between two spaces is one word), which we also apply to the TEF. Below the minimum (80 words in section A, 200 in section B), the report shows a red warning; our code does not lower the level for that, but the AI is told the count and can. There is no maximum at the TEF Canada, and a text under 15 words is not assessed. Off topic: the AI judges whether your text answers the prompt, the one shown or the one you paste. If it says no, you get a red warning with a comment and our code caps the estimate below NCLC 4. The AI can get this wrong.

Partly: it shows the level of the text you submit and what to fix first, but it does not know your target and writes no list specifically for NCLC 7. In our estimate, NCLC 5 and 6 are B1 and NCLC 7 is the start of B2, so the step is from B1 to B2. To give B2, the AI is told to look for several of these: varied linking words, correct complex sentences, developed ideas, precise vocabulary, only occasional errors. These are our grader’s rules, not a CCI grid. Our code also holds a text at B1 or below if errors are frequent, or if a text of 100 words or more uses no varied linking word, and the report says so, even in the free preview. The full report adds a mark per criterion and your 3 priorities. This covers one text in one section: for both sections together, take our writing mock test.

They can help you find mistakes, but no AI can promise your official score. Our tool is an AI too, and we have no data showing it comes closer to official results than they do. What we can show is what our code does around the AI: it counts your words, it converts the AI’s level to NCLC with IRCC’s table for the TCF (our approximation for the TEF) and to the matching band out of 699 from CCI’s table, it drops any correction quoting a passage that is not in your text, and it brings C1 or C2 back to B2 if the AI’s quotes do not support it. That keeps the numbers consistent; it does not show they are closer to your official result. The section rules and CCI’s criteria are built in, so you do not have to write them into a prompt. ChatGPT and Claude, on the other hand, let you ask follow-up questions; our report does not.

Use AI correction to practise often and a teacher for explanations and follow-up; you can also combine them. AI correction needs no appointment and, here, has a free preview and then starts at US$9.90 for 10 corrections, so it suits writing a lot and spotting the mistakes that come back. A teacher can do what our report cannot: answer your questions, work out why a mistake keeps coming back, adapt the work to you and follow you over time. Think about a teacher if the same errors return over several texts, if your estimated level stops moving, if an explanation stays unclear, or if you are aiming above B2, where our estimate is conservative. You can also show your report to a teacher and ask them to check it. Neither gives an official result: only CCI Paris Île-de-France does.