How Prepiva’s practice tests are built and scored

Every question, passage, recording script and picture in Prepiva’s tests is original Prepiva material. We do not copy or closely paraphrase ETS questions.

What we take from ETS is the published structure of the 2026 test: the four sections, the twelve task types, the number of questions and the timing. Within that structure, each Prepiva test has its own topics, from science and history to campus life.

The recordings in Listening and Speaking use computer-generated voices. The pictures of speakers on our test screens are generated images, not photos of real people.

Prepiva has ten full practice tests, Practice Exam 01 to 10. Each follows the order of the 2026 test:

• Reading: 50 questions in 30 minutes, in two stages. Everyone takes the same first stage of 35 questions. Based on those answers, you get an easier or a harder second stage of 15 questions. The easier stage is written for levels A2 to B1, the harder one for B2 to C1.
• Listening: 47 questions in 29 minutes, also in two stages: a first stage of 32 questions and a second stage of 15.
• Writing: 12 tasks in 23 minutes: 10 Build a Sentence items, one email (7 minutes) and one academic discussion post (10 minutes).
• Speaking: 11 tasks in 8 minutes: 7 Listen and Repeat sentences and 4 interview questions.

As in the test itself, some Reading and Listening questions are unscored try-out questions. On every path through a Prepiva test, 35 Reading questions and 35 Listening questions count toward your band.

Reading and Listening. Each scored question is right or wrong, and the more scored questions you answer correctly, the higher your band. One fixed rule turns your share of correct answers into a band from 1 to 6, rounded to the nearest half band. If your first-stage answers send you to the easier second stage, the highest band you can reach is lower than on the harder stage.

Writing. The 10 Build a Sentence items earn one point each, and the email and the discussion post each get an AI rating from 0 to 5. We add up the points and turn the total into a band with the same kind of rule.

Speaking. Each of the 11 answers gets an AI rating from 0 to 5. We add up the points and turn the total into a band in the same way.

Overall band. Your overall band is the average of your four section bands, rounded to the nearest half band. This is how ETS calculates the overall score.

CEFR level and 0–120 range. We show the CEFR level and the comparable 0–120 range for your band from ETS’s published table. See the 1–6 band scale for the full table.

How this differs from ETS. ETS turns raw points into scores with statistical equating. In Reading and Listening it also accounts for how difficult your questions were, and it links Writing and Speaking scores across test versions. Prepiva’s method is much simpler: one fixed rule per section, with no statistical equating. We have not yet compared Prepiva bands with the ETS scores of the same people, so we cannot say how close they are. This is the main reason a Prepiva band is an estimate.

Writing: your email and your discussion post are sent to a large language model (a Qwen model, run through the OpenRouter service) together with a scoring guide for the task.

Speaking: your recording is first turned into text by a speech-recognition model (OpenAI’s Whisper, also through OpenRouter). The language model then rates that transcript.

Each answer gets a score from 0 to 5, a short reason for the score, one strength and one next step. The rating is labelled provisional on your report.

What the rating looks at:

• Write an Email: whether you cover every point the task asks for, clarity, how well your ideas connect, the right tone for the reader, grammar and vocabulary.
• Write for an Academic Discussion: a clear opinion with support, how you take part in the discussion, how well your ideas connect, grammar and vocabulary.
• Listen and Repeat: how accurately you repeat the sentence and how easy it is to understand.
• Take an Interview: whether you answer the question, how well you develop your answer, organization, grammar and vocabulary.
• Empty, off-topic or non-English answers get 0.

What it cannot judge:

• Pronunciation. The model reads a transcript, not your voice, so it cannot reliably judge pronunciation, stress or intonation. ETS’s Speaking scores do include how easy your speech is to understand.
• Its own mistakes. A language model can misread an answer or rate it too high or too low. A speech-recognition error can also change what the model sees.
• ETS’s engine. ETS rates Writing and Speaking with its own AI scoring engine, trained on human ratings and checked by human raters. Prepiva’s rating is not that engine and was not trained on ETS ratings.

• Practice estimates only. Your scores, CEFR levels and AI feedback on Prepiva are practice estimates. They are not ETS results, and they do not predict the score ETS will report.
• A simpler conversion. Bands come from the fixed rules described above, not from statistical equating.
• Provisional AI ratings. Writing and Speaking ratings can be wrong, and they do not measure pronunciation.
• Generated speech. Our recordings use computer-generated voices, which may sound different from the voices on test day.
• Review not finished. See the review status below.

Who writes. Prepiva’s practice tests, guides and web pages are written by the Prepiva team.

Who reviews. Each page on this site is reviewed by a person before it goes live, and shows the date of its last review.

Review status. Our practice tests, their timing, the AI ratings and the generated speech still need a human editorial and psychometric review. That review has not been completed. Until it is, please treat every band and rating as a rough guide, and tell us when something looks wrong. We will update this section when the review is done.

How AI is used. AI rates your Writing and Speaking answers, as described above.

Errors and corrections. If you find a mistake in a question, an answer, a score or a page, tell us on our support page with the test, the section and the question number. We reply, and when we confirm an error we fix it and update the page’s review date.

Review dates. Every page shows when it was last reviewed.

Questions about how we build or score our tests: write to us. To learn who we are, read about Prepiva. To see the scale we use, read about the 1–6 band scale, or take the free practice test and see your own report.