Independent · Not affiliated with NAATI
How our AI marking works, and its limits
With Premium, CCL Mate's AI marks each answer you record in a mock test and gives you an estimated score out of 90. It is a practice estimate to help you find and fix your errors, not a NAATI result: only NAATI's examiners give official results.
What happens to your answer
- You record your answer to a segment of the mock test. It is sent over an encrypted connection to our server for marking.
- The AI writes down what you said, word for word, with your hesitations and self-corrections. It does this without seeing the expected answer, so it cannot “hear” words you did not say. For some languages, where our own tests found more listening errors, a stronger listening model is used.
- The AI checks the meaning. It compares what you said with the original segment and a model answer, splits the original into small pieces of meaning (who, what, when, how much, why, and under what conditions) and marks each one as conveyed, partly conveyed, left out or distorted. It also lists anything you added, language issues (grammar, a formality that does not match the speaker, a wrong term, English words left in your language, unnatural phrasing, or speaking in the third person), and counts your self-corrections and hesitations.
- Fixed rules turn these notes into marks. You see the estimate for each segment and dialogue, what you left out or changed, short feedback and, where your answer can be improved, a better version of it.
How the estimate is worked out
NAATI describes its own marking as mark deduction: each test begins with 90 marks, and marks are taken off for errors, more or fewer depending on how much each error affects the communication. Our estimate works the same way. Each dialogue starts at 45 marks, and each segment is worth its share of the 45, by length.
- A key detail (a number, date, time, name, negative, dose or condition) counts more than other details, and a changed detail costs more than one left out.
- Additions, language issues, a late start (more than 5 seconds after the chime), repeated self-corrections, many hesitations and long silences cost smaller amounts.
- Each repeat after the first in a dialogue costs a mark. If nearly a whole dialogue is in the third person (“he says that…”), that dialogue scores 0, because NAATI may invalidate a test for it.
The two dialogues add up to the estimate out of 90, which you can compare with the pass standard: 63, with at least 29 in each dialogue. NAATI publishes its method and what it marks, but not how many marks each error costs, so these amounts are our own judgement.
What it cannot do
- It can mishear you. Speech recognition makes more mistakes in some languages and accents, and with background noise or a quiet microphone. Your results show what it heard, so you can check.
- It does not replace an examiner. It can misjudge a nuance or the level of formality, or mark down a correct answer worded in an unusual way.
- It has not been compared with NAATI's results. We have not measured how close its estimates come to official NAATI results, so we publish no accuracy figure, and a good estimate is no promise of a pass.
- It marks mock tests only. In word and sentence practice your recordings stay on your device and are not sent for marking.
Your recording
Each recording goes over an encrypted connection to our server (Google Cloud Functions for Firebase, in Australia) and is processed by Google's Gemini AI through Google Cloud Vertex AI. Google may do this processing outside Australia. We do not keep recordings on our server after marking, we do not listen to them, and Google Cloud does not use them to train its AI models. Your marks, feedback and a written transcript of each answer are saved on your device and, if you are signed in, in your account. Details are in our privacy policy.
When NAATI's assessed practice test is the better choice
NAATI's paid assessed practice test (A$165) is taken in the same conditions as the real test, uses previously used test materials, and is marked by NAATI examiners with the same marking rubrics and scores; you get feedback from the examiners. If you want an examiner's view of your readiness, it is the better choice. Its results take 4 to 6 weeks, so book it well before your test; NAATI says practice results cannot be used to claim migration points. See NAATI's CCL page. CCL Mate's AI marking is for practising often, between such checks.
AI marking comes with Premium, a subscription from A$19 every 2 weeks: see pricing, or start with the free mock test, which you mark yourself.
Questions
Is the AI score my real CCL result?
No. It is a practice estimate out of 90, against the pass mark of 63 with at least 29 in each dialogue. Only NAATI gives official results, and CCL Mate is not affiliated with NAATI.
How accurate is CCL Mate's AI marking?
We do not publish an accuracy figure, because we have not compared our estimates with official NAATI results. Use the estimate and the notes to find your errors and to follow your progress over several mock tests, not to predict your result.
Does a person listen to my recording?
No. Your recording is marked by AI. We do not listen to it, and we do not keep it on our server after marking. A copy stays on your device so you can replay your answers.
How many answers can I have marked?
With Premium, up to 150 answers a day, about 5 full mock tests; the limit restarts at midnight Sydney time. The free version has no AI marking: you mark yourself against the model answers.
Which languages does the AI marking cover?
All 15 CCL Mate languages, in both directions: English into your language, and your language into English.
NAATI's marking method: NAATI, Credentialed Community Language test (checked 26 October 2026). CCL Mate is not affiliated with NAATI.