Trust Center
How scoring works
Scope: the Writing section. Last reviewed July 27, 2026.
Two different things happen
Build a Sentence is not judged. Your arrangement either matches the accepted answer or it does not, and the result is a count. No AI is involved in deciding whether a Build a Sentence item is right.
Write an Email and Academic Discussion are judged. An AI model reads your response against criteria we wrote, and produces a 0–5 rubric estimate plus written feedback. That is a judgment, and judgments can be wrong.
What produces the written feedback
Two passes. The first returns the rubric estimate and a short summary quickly. The second is slower and produces the detail — strengths and weaknesses with quotes from your own text, grammar corrections, and an improved version of your response for comparison.
The criteria the model applies are our interpretation of publicly available task descriptions. They are not ETS rubrics, and the model is not trained on official scored responses — it is given task-specific instructions.
About the numbers
The 0–5 figure is a practice estimate on a rubric scale. If you also see a figure out of 30, that is a comparison against the pre-2026 scale, kept because many study materials still use it. Neither is an official score. Official reports for tests taken on or after January 21, 2026 use a 1–6 scale in half-band increments.
What this cannot tell you
- What you would score on the real test. We have not published a study comparing our estimates to official results, so we do not claim one.
- Every problem in your response. The feedback names issues it found, which is not the same as finding everything.
- A guaranteed improvement. Acting on feedback is how practice works, but we do not promise a resulting score change.
- Whether a human grader would agree. If a teacher disagrees with a result here, that is useful information, and the teacher has seen things the model has not.
When a result looks wrong
Tell us. A result you disagree with is more useful to us than one you accept. Send it here with the task, and we will look at the response itself.