Our method

Who makes civicoach

civicoach is made by Mario Corea, a native Italian speaker, in Milan. It is a small independent project: not a language school and not an exam body, and we don’t claim teaching qualifications we don’t have.

We are not affiliated with CILS, CELI, PLIDA or CERT.IT. We use their names only to say which exam you are practising for.

How the exercises are written

We write every exercise ourselves. None is copied from an official exam. Each one follows the format the certifier publishes for its B1 exam: the parts, the task types, the time limits, the word counts and how each part is scored. These are the documents we follow:

• CILS B1 Cittadinanza: evaluation criteria (Università per Stranieri di Siena, PDF)
• CELI 2 i: exam description (Università per Stranieri di Perugia, PDF)
• PLIDA B1: Quaderno delle specifiche (Società Dante Alighieri, PDF)
• CERT.IT B1: level page (Università Roma Tre)

Where a certifier’s documents are unclear or contradict each other, our exam guides say so instead of guessing. The listening recordings are read by AI-generated Italian voices; we check every recording against its script with speech recognition.

How the AI corrects your writing and speaking

Your text, or the transcript of what you said, goes to an AI model together with the criteria of your exam (for example task completion, vocabulary, grammar and spelling) and our scoring guide. The AI gives a score for each criterion, marks the mistakes and explains them in your language.

Every feedback screen says that it comes from AI. Your mistakes go into your deck, so the plan brings them back until you fix them.

How we checked the AI

Before launch we prepared a set of practice answers at different levels and graded each one against the published criteria, before the AI saw them. These are our own reference grades, prepared in-house with the help of AI tools. They are not grades from official examiners, and no real candidate’s exam paper was used.

Then the AI graded the same answers. It agreed with our pass or fail decision on 97% of the writing answers (33 answers) and 94% of the speaking answers (17 answers), in October 2026.

What this shows: the AI applies the criteria the way we read them, consistently. It does not prove that an official examiner would give you the same score.

What the AI cannot judge

For speaking, your phone turns your voice into text and only the text is scored. So the AI cannot hear your pronunciation or intonation. Where an exam scores pronunciation, your feedback marks it as not assessed. What the AI judges is what you said: vocabulary, grammar and how you handled the task.

Speech recognition can mishear a word, and that can look like a mistake you didn’t make. Practise out loud anyway, and if you can, ask an Italian speaker to listen to you.

The AI can also be wrong. If a correction looks odd, check the lesson and the rule, and write to us: we read every message.

Practice estimates, not official results

Every score in civicoach is a practice estimate, not an official result. It shows you what to work on. Only the certifier’s exam decides whether you pass, and we never promise that you will.

Explanations in your language

The lessons and explanations are written in Italian and English and translated into the other languages with AI. Native speakers have not yet reviewed every language. If something reads wrong, tell us at support@civicoach.app.