The whole test runs about ten minutes, and every candidate sits the same one — Cambridge IGCSE speaking is untiered, so there is no Foundation or Higher paper to choose between. It is graded A*–G.

Two things surprise candidates who have looked at UK GCSE material first. There is no photo card and no reading-aloud task — 0520 does not use them. And the two topic conversations are not scored one by one: they are marked together, fifteen marks for communication and fifteen for quality of language, across both.

You get ten minutes of supervised preparation before the test, during which you are given the role play scenario on a candidate card. You may not make notes. Dictionaries are not allowed at any point.

This is worth dwelling on if you have been revising from AQA or Edexcel resources, because those boards work the opposite way: fifteen minutes of preparation with notes permitted, handed in before you finish. Advice written for a UK GCSE — “plan your answers, jot down key phrases” — is advice you cannot follow in a 0520 exam room.

What you can do in those ten minutes is think. Read the scenario, work out who you are talking to and what you need from them, and rehearse the shapes in your head rather than the words on paper.

You are also not told the topics of the two conversations in advance. That is by design: the conversations test whether you can hold an unprepared discussion, not whether you memorised one.

The app runs the sequence exactly as the exam does.

• Ten minutes of preparation first, with the role play card in front of you
• A short warm-up that isn't assessed, the same as the real thing
• The role play — about two minutes, five transactional tasks, with the AI staying in role and responding to what you actually say
• Topic conversation 1 — about four minutes
• Topic conversation 2 — about four minutes, on a different theme
• Scores across six skills, written feedback, and a full report at the end

No photo card. No reading aloud. No tier to pick.

There are 44 French IGCSE role-play scenarios in the app, set in everyday transactional situations — shopping, travel, work, accommodation and more — so you can practise repeatedly without repeating yourself. Every scenario is written in-house and follows the board's published format, timings and mark weightings.

Ten of the forty marks are on the role play, where the job is transactional: complete the five tasks. Simple language that does the job scores; elaborate language that misses the task does not.

The remaining thirty are on the two conversations, split between communication — whether you develop your answers, give opinions, offer reasons — and quality of language, meaning range and accuracy of vocabulary and structures. Both are judged across the two conversations as a whole, so a strong second conversation genuinely can rescue a nervous first one.

The most useful habit for the conversations is not vocabulary. It is length. Candidates who answer in one sentence and stop cap themselves early, however good the sentence. Answer, give a reason, add an example.

The difficulty with speaking is that it is the one component you cannot revise alone from a book. You need somebody to talk to, and in most schools individual speaking rehearsal is rationed by timetable rather than by willingness.

That is what this is for. You speak French out loud, the AI replies in French, and it scores you after every turn — pronunciation, grammar, vocabulary, fluency, interaction and confidence. Ten to fifteen minutes a day, four or five days a week, does more than an hour of silent revision the week before.

Seven sessions every 30 days are free, with no card. Paid plans start at £9.99 a month.