When to use it
Use this prompt when you have studied grammar and vocabulary but rarely get to use them in conversation. It suits learners preparing for a trip, a move abroad, an oral exam or a job in another language. Short daily sessions of ten to fifteen minutes work better than an occasional long one, and the low-pressure setting makes it easier to make mistakes and learn from them.
How to customize it
- [LANGUAGE] and [MY LEVEL] are the essentials. If you know your level on the CEFR scale (A1 to C2), use it; otherwise describe what you can do.
- [NATIVE LANGUAGE] is used for corrections, so explanations are easy to follow.
- [SCENARIO] should match situations you will actually face. Practising a pharmacy visit before a trip is more useful than a generic chat.
- [WHAT I WANT TO PRACTISE] lets the partner steer the conversation towards structures you need, such as questions in the past tense.
Why it works
The prompt balances fluency and accuracy. Short replies that end in a question keep you producing language, which is what most learners lack. Limiting corrections to the one or two most important mistakes per message avoids an overwhelming wall of red ink, which can make learners stop writing.
Keeping corrections in a separate block lets the conversation flow while still giving feedback. The new-word markers and the final summary turn a chat into material you can review. Large language models are generally strongest in widely used languages; quality can be lower for languages with less online text, a gap our article on AI and the language divide explains.
Tips and variations
- Repeat the same scenario a week later and compare your mistakes list; seeing an error disappear is a good motivator to keep practising.
- If your assistant has a voice mode, use it to practise listening and pronunciation as well as writing.
- Add “Speak slightly faster and use more idioms” once the scenario feels easy.
- Copy the end-of-session vocabulary into a flashcard app, or use our flashcards from your notes prompt.
- Double-check unusual corrections with a grammar reference or a native speaker; models occasionally correct things that were already right.