How Are English Tests Scored? AI and Human Raters

Most test takers look at their score without asking how it was produced. That's worth asking, because the method affects how consistent the score is, how quickly you receive it, and how much weight universities, employers, and visa authorities give it.

Three ways to score a test

Answer-key marking. Multiple-choice and other closed questions have one correct answer, so they are marked by matching against a key. The same answers always produce the same score.

Examiner scoring. For open-ended answers, such as an essay or a spoken response, trained examiners apply published criteria. Their strength is judgement. Their weakness is that two examiners may not always agree.

Automated scoring. AI systems score open-ended answers against defined criteria and apply the same rules every time, but their quality depends on how they were built and checked.

Many tests combine these methods.

How five tests are scored

The table summarises what each provider says on its own pages. Details change, so check the provider's current information before you book.

IELTS

IELTS publishes its criteria. For Writing: task achievement (Task 1) or task response (Task 2), coherence and cohesion, lexical resource, and grammatical range and accuracy. For Speaking: fluency and coherence, lexical resource, grammatical range and accuracy, and pronunciation. The criteria are weighted equally. IELTS appears in many visa and immigration requirements, including routes in the UK and Australia, but each route keeps its own list of accepted tests.

TOEFL iBT

ETS reports four section scores and an overall score on a 1 to 6 scale. The overall score is the average of the four sections, rounded to the nearest half band. ETS says scores are available in your account three days after the test.

PTE Academic

PTE Academic is taken on a computer, without a face-to-face examiner. Pearson describes its scoring as AI technology combined with human expertise, built on a large set of examiner responses.

Duolingo English Test

The Duolingo English Test reports four subscores (speaking, writing, reading and listening) that average to the overall score on a 10 to 160 scale. For how the test is scored, see the provider's own scoring and research pages.

The International English Test

The method is set out on our "How we score" page:

• Grammar, vocabulary, reading and listening are multiple-choice and marked against a verified key, with no AI.
• Speaking is scored by AI language models (Google Gemini, Anthropic Claude) on fluency, coherence, grammar, vocabulary and pronunciation; writing on grammar, vocabulary, content and organisation.
• Each criterion is scored from 0 to 100, the scores are averaged with equal weight, and the level is calculated in code from a fixed table. The model does not choose the level.

The English level test uses the multiple-choice part only, and you see your level at the end. If you need proof of your level, see the English certificate.

Reliability and validity

Two terms come up whenever scoring is discussed. The Standards for Educational and Psychological Testing (open access), published by AERA, APA and NCME, define them:

• Validity is "the degree to which evidence and theory support the interpretations of test scores for proposed uses of tests". In plain terms: does the score mean what it is used to mean?
• Reliability refers to the consistency of scores across replications of a testing procedure. Would the same performance get the same score from another rater, or on another day?

With answer-key marking, the question answers itself: the key either matches or it doesn't. For examiner and AI scoring, consistency has to be shown with evidence, such as agreement between raters or between a model and human raters. Providers differ in how much of that evidence they publish.

What can go wrong with automated scoring of speech

Any system that scores recorded speech depends on the recording. A noisy room, a poor microphone, or a weak connection can lower the quality of the evidence. If you are taking a test with a speaking section, use a quiet room and check your microphone before you start. On our speaking test, the recorded audio is evaluated directly, without a separate speech-to-text step.

Choosing a test by its scoring method

• Immigration and high-stakes university entry: the requirement usually names specific tests, often IELTS or TOEFL. Check the list for your route or course before you book.
• Work: many employers ask for a CEFR level rather than a named test. A result that states a CEFR level and can be checked online is easy for an employer to read.
• Your own planning: any test with skill-by-skill results helps you see what to study next.

Common mistakes about scoring

• Assuming two tests are equivalent because both mention the CEFR. Conversion tables are estimates.
• Ignoring sub-scores. An overall result can hide a weak skill.
• Choosing by speed alone. A fast result is only useful if the organisation you are applying to accepts it.
• Not checking review policies. Many providers offer a score review or remark. Check the policy before you assume a score is final.

Key points

• Closed questions are usually marked against an answer key; open-ended answers are scored by examiners, AI systems or both.
• IELTS publishes four equally weighted criteria each for Writing and Speaking; TOEFL iBT now reports on a 1 to 6 scale.
• Reliability is about consistency; validity is about whether a score supports the use it is put to.
• For immigration and university entry, the requirement usually names specific tests, so check it before you book.