How we grade essays
The method behind every score: a fixed rubric, three grader runs, a median, and a check that every quote is really in your essay. Plus what it cannot do.
Last updated 2026-11-03.
The method in four steps
A fixed, versioned rubric
Each essay type has one rubric, with the version shown on your result. Public rubrics are summarised in our own words and linked below. For each score band we wrote example answers ourselves; they are synthetic, not real students' work.
Three runs, median score
The grader scores your essay three times with the same rubric. A run that fails is retried once, and we need at least two valid runs. We keep the median for each criterion, so one lucky or harsh run does not decide your score.
“Less certain” flags
If the runs disagree by 2 points or more on a criterion, it is marked less certain, with a striped bar. Treat that score as a range, not a number.
Quote check
Every quote in a note must appear word for word in your essay, ignoring only curly-versus-straight quote marks and extra spaces. A quote that fails the check is dropped, never shown.
Rubrics and official sources
We summarise each rubric in our own words. We do not copy the official descriptor text. For the real thing, read the source. Links read on 2026-11-03.
| Essay type | Our scale | Whose rubric | Official source |
|---|---|---|---|
| College application essay | 6 criteria, 1–6 each, total /36 | Our own rubric. Common App publishes none. | Common App 2026–27 prompts |
| Class essay (6+1 Traits) | 6 traits scored, 1–6 each (Presentation, the seventh, is not scored) | Education Northwest | 6+1 Trait rubrics |
| IELTS Writing Task 2 | 4 criteria, band 0–9 each, overall to the nearest 0.5 | IELTS | Band descriptors · Key criteria |
| TOEFL iBT writing | 3 criteria per task, 0–5 each, averaged to the nearest 0.5 (ETS scores each task as one 0–5 level) | ETS | Writing scoring guide · Writing section |
| AP Lang and Lit FRQs | Rows A, B, C: 0–1, 0–4, 0–1, total /6 | College Board | Lang 2026 · Lit 2026 |
| ACT Writing | 4 domains, 1–6 each, shown as an estimate out of 12 (ACT reports 2–12 from two readers) | ACT | Writing scores · Rubric PDF |
The SAT Essay was discontinued by College Board, so there is no SAT preset. Per-exam pages: college, IELTS, TOEFL, AP.
How we tested it
Synthetic essays, checked honestly.
Before launch we ran a calibration set of 10 fictional college essays written for testing (three intended quality levels: weak, average, strong), each graded three times by each of the two models we use. These are not real students' essays, and 10 essays is a sanity check, not a benchmark.
- Both models put all ten essays in the intended order: weak below average below strong.
- Across the three runs of one essay, the total (out of 36) varied by 0 to 4 points. That spread is why we keep the median of three and not a single run.
- Occasionally a run returns nothing usable. We retry it once and need at least two valid runs.
- Almost every quote the models produced was found word for word in the essay; the ones that were not are exactly what the quote check removes.
Run on 2026-11-03. We repeat this check when a model or rubric changes.
Which models grade your essay
Standard grades and the rubric generator use a fast MiniMax model through OpenRouter. Deep reviews use a stronger xAI Grok model. We choose models by testing them on the same rubric for consistency, speed and cost, and we can change them; the rubric version on your result stays the same unless the rubric itself changes. Both providers receive only the text needed to grade, as listed in the privacy policy.
What a score means
An estimate against a rubric.
The number is a practice estimate of how a reader using that rubric might score the draft you pasted. It is not an official score from the College Board, ETS, IELTS, ACT or any admissions office, and it does not predict a result.
We never write your essay. The grader quotes your sentences, explains each point, and rewrites one sentence at a time as an example. A request to write a full essay is refused.
Limits
- AI graders can be wrong, and can disagree with a human reader by a point or more. Ask a teacher or counselor about anything important.
- It sees text only. Handwriting, timing and sources you do not paste are invisible to it.
- Standard grading reads up to 1,000 words; deep review reads up to 1,500.
- It does not know a school's admissions priorities or your teacher's private expectations.
- We calibrate the models on synthetic test essays and check them again when we change a model or rubric.
What happens to your essay
Your essay is sent to an AI model to be graded and stored in your history until you delete it. It is not used to train models. The full list of services that touch it, and how to delete it, is in the privacy policy.
Questions about our method
Why three runs?
One model run can be lucky or harsh. The median of three is steadier, and disagreement between runs tells us when to say we are less certain.
Can I use my teacher's rubric?
Yes. Paste or upload it as a custom rubric, or build one with the rubric generator.
Do you change rubrics?
Sometimes. Each rubric has a version number shown on your result, so older results stay comparable.
Who is behind this?
Grade Your Essay is a Hueflo tool. See all Hueflo tools.
What is deep review?
A paid option that uses a stronger model and reads up to 1,500 words. You get paragraph-by-paragraph notes, a view of how a reader may take the essay, a prompt-fit check, a cliche check and a revision plan. It still never writes the essay for you.
Can I grade an essay in another language?
The rubrics and feedback are written for English essays. Other languages are not supported.