Set up an AI study partner and establish your starting point

Set up a simple workspace

Use an assistant you already have access to and whose data settings you understand. Text is enough to begin. Create a new conversation and a separate document called Study Log. Keep four sections: goals, unassisted samples, corrections to practise, and weekly reviews. Use fictional workplace examples instead of actual client material.

If voice is available, test the microphone with an invented sentence and check the transcript. If it is not, record yourself locally and use a consenting human partner for spoken feedback when possible. A typed dialogue is a valid interaction exercise, but it is not a pronunciation assessment.

Paste this starting instruction:

Be my English practice partner, not an examiner.

My working goal is to move from roughly B2 tasks toward C1-like tasks;

my level is not verified. Ask me to attempt before showing a model answer.

Use English by default. Explain in Russian only after my attempt if I ask.

Ask one question at a time. At the end, quote at most three useful points

from my actual response, separate errors from optional style changes,

and ask me to repair them. Do not assign a certified CEFR level.

Do not infer pronunciation from a transcript. State uncertainty.

Try it with one short answer. If the assistant immediately rewrites everything or announces your level, ask it to follow the boundaries. Keep a copy of the instruction in your log so you can restart without depending on memory.

What B2 to C1 means here

CEFR describes abilities across different language activities. Treat the target as a profile: handling more demanding meaning, expressing ideas with greater flexibility and producing well-organized language. Read the official self-assessment grid rather than treating a single chatbot score as your identity. Your reading, listening, speaking and writing may develop differently.

The British Council level test explicitly offers an approximate indication and focuses on grammar, vocabulary and phrasing. At the time checked it required login or registration. It does not replace separate evidence of your speaking or writing. A recognized assessment is the appropriate route when certification is needed.

Make a baseline before accepting help

Complete these original tasks without AI, dictionaries or a model answer. Record the conditions and date; these are Academy exercises, not validated CEFR tests.

Write 180–220 words: “Your fictional team wants to replace every weekly meeting with written updates. Recommend a policy, acknowledge a drawback and explain an exception.” Take 20 minutes as an exercise constraint, not an exam standard.

Speak for two minutes on the same decision without reading your text. Then answer: “What would change your recommendation?” Save the recording if you are comfortable doing so. For reading and listening, select separate official practice materials with answer keys from Cambridge's preparation resources. Follow their own instructions; do not ask AI for answers first.

Now request feedback on the writing using this prompt:

Here is my unassisted sample. For task fulfilment, organization,

wording and grammar, give one observable strength and one specific

development need. Quote my words. Do not convert this to a CEFR level

or a percentage. Identify uncertain judgments. Give me two repair tasks

instead of rewriting the whole answer.

Turn feedback into a study decision

A fictional observation might be: “The recommendation is clear, but the exception is missing.” That suggests practising qualification and exceptions. “Use more advanced vocabulary” is too vague; ask for the exact phrase and why an alternative serves the meaning better. Check disputed corrections in a reliable dictionary or grammar reference and keep unresolved items marked.

Your output should be a profile such as: “I can state a recommendation; I need to explain exceptions and respond to challenges. Listening details need separate checking.” This is useful even without a level label.