Inside Our CEFR Scoring Model: From Raw Answers to a Level
What this article covers
Whether you are using an International English Test result for hiring, placement, or your own planning, you should be able to see exactly how it was produced. This article covers the method step by step: how answers are marked, how a percentage becomes a level, what the skill scores show, and how speaking and writing are scored.
Everything here matches the "How we score" page on our site, which is the reference for this method.
QUICK ANSWER
Multiple-choice answers are marked against a verified key, with no AI. The correct answers become a percentage of the whole test, and a fixed table turns that percentage into a CEFR level, from A1 at the bottom to C2 at the top. Speaking and writing are scored by AI models on published criteria, and the level is then calculated from the same table.
Step 1: marking the answers
The method depends on the kind of question.
Grammar, vocabulary, reading and listening use multiple-choice questions. Each has one correct answer, matched against a verified key. No AI is involved and nothing is interpreted. The same answers always produce the same score.
Speaking and writing are open-ended: there is no single correct sentence to match against. These are scored by AI language models against a fixed set of criteria (Step 4).
An automated validator checks every question in the bank against its key, for every version of the question that is served. It flags four kinds of error: a correct answer that is not among the options, two identical options, an empty or incomplete prompt, and a question with no key at all. Items that need human judgement are flagged for editorial review rather than changed automatically.
Step 2: from correct answers to a level
The correct answers are counted and expressed as a percentage of the whole test. That percentage is placed on a fixed table:
| Score | CEFR level |
|---|---|
| Below 35% | A1 |
| 35 to 49% | A2 |
| 50 to 68% | B1 |
| 69 to 78% | B2 |
| 79 to 94% | C1 |
| 95 to 100% | C2 |
There is no hidden weighting: the same number of correct answers always gives the same level.
How the table is used depends on the test:
| Test | How the level is given |
|---|---|
| English Level Test | The band your score falls into |
| Level-specific test (for example, a B2 test) | Certifies that level with a 70% pass mark |
The levels come from the CEFR, published by the Council of Europe. For what each level means in practice, see our guide to the English levels.
Step 3: the skill scores
Each question is tagged to a skill: grammar, vocabulary, reading, or listening. Alongside the overall level, the result shows a percentage for each skill.
The overall level comes from the total percentage, not from the skill scores. The skill percentages show where you are stronger and weaker. A strong reader with weaker listening sees exactly that.
Step 4: speaking and writing
Speaking and writing are scored by AI language models from Google (Gemini) and Anthropic (Claude), on these criteria:
| Skill | Criteria |
|---|---|
| Speaking | Fluency, coherence, grammar, vocabulary, pronunciation |
| Writing | Grammar, vocabulary, content, organisation |
For speaking, the recorded audio is evaluated directly, with no separate speech-to-text step. Each criterion is scored on its own, from 0 to 100. The scores are averaged with equal weight, and the CEFR level is then calculated in code from the same threshold table as above. The model returns a score for each criterion; it never chooses the level.
We explain why we publish these criteria in why we publish our AI scoring criteria.
What the results look like across all test takers
The table below counts every English Level Test submitted on our site over 91 days. It counts tests, not people: someone who took the test twice appears twice.
| Result | Share of tests (9 Aug to 7 Nov 2026, 279,371 tests) |
|---|---|
| A1 | 7.6% |
| A2 | 12.2% |
| B1 | 24.3% |
| B2 | 17.6% |
| C1 | 32.4% |
| C2 | 5.9% |
These are the people who chose to take our test, not a random sample of English learners, so the shares describe our test takers and nothing wider.
Who stands behind the method
International English Test is operated by International Test Limited, a UK company, and is listed as an Associate Member on the ALTE associate members page. ALTE is the Association of Language Testers in Europe. Associate membership is the status held; this is not described as accreditation.
What to check as a buyer
If you are assessing this method for an organisation, three questions are worth asking of any test:
- Can the marking be repeated? For the multiple-choice skills here, yes: the same answers always give the same score.
- Is the score-to-level rule written down? Here it is the fixed table in Step 2.
- Are the criteria for open-ended answers stated? Here they are the five speaking and four writing criteria in Step 4.
For the wider question of what makes a result hold up under scrutiny, see what makes an English test result defensible. For using the test in a company, see English assessment tests for companies. To see the method from the candidate's side, take the English level test.
Key points
- Multiple-choice questions are marked against a verified key, with no AI.
- The percentage of correct answers is placed on a fixed table, from A1 at the bottom to C2 at the top.
- The overall level comes from the total percentage; skill percentages are reported alongside it.
- Speaking and writing are scored by AI models on published criteria, and the level is calculated in code from the same table.
Frequently Asked Questions
International English Test Editorial Team
ALTE Associate Member · UK English assessment provider · Est. 2023
Found this helpful? Share it:
