Tool · Teachers
AI grader for essays and papers, on your rubric
Paste your rubric one criterion per line, choose the point scale, and add the class set — pasted, or uploaded as files. The AI grader scores every submission on every criterion, returns one table for the whole set, shows the level behind each point, and exports to CSV.
0 submissions
How the AI grader works
Each line of the rubric becomes one question, asked about each paper. For a criterion called Evidence with the description “supports the claim with relevant, specific evidence”, the model is asked, verbatim: Rubric criterion "Evidence": supports the claim with relevant, specific evidence. Which level describes this student's work on this criterion? It answers with a probability for every level on the scale, and the nearest level becomes the points.
The levels come from the point scale you pick. These are the ladders it uses:
| Scale | Levels, lowest first |
|---|---|
| 1–3 | Does not meet the criterion → Meets the criterion → Exceeds the criterion |
| 1–4 | Beginning → Developing → Proficient → Exemplary |
| 1–5 | Beginning → Developing → Proficient → Accomplished → Exemplary |
| 1–6 | Minimal → Limited → Developing → Proficient → Strong → Exemplary |
Every submission is graded in its own call, up to thirty per run and five at a time. That is a design choice worth knowing about: when one long prompt grades a whole class at once, the papers read early can shift the standard for the papers read late. Here, a paper's points depend only on the paper, the rubric and the assignment.
A test set, graded
We graded three essays we wrote at known levels — strong, average and weak — on a four-criterion, 4-point argument rubric, on 26 September 2026. All three came back in half a second.
| Student | Claim | Evidence | Reasoning | Conventions | Total |
|---|---|---|---|---|---|
| A — strong | 4 | 3 | 4 | 4 | 15/16 |
| B — average | 3 | 2 | 2 | 2 | 9/16 |
| C — weak | 3 | 2 | 2 | 2 | 9/16 |
B and C tie at 9 in whole points, yet they are not the same paper, and the detail view shows it. On Reasoning, the model placed B at 1.35 and C at 0.60 on a ladder running from 0 to 3; both round to a 2, but one is on its way to Proficient and the other has barely left Beginning. Those in-between positions are where a borderline point is decided, and they are exactly where your judgment belongs.
Papers as well as essays
As a paper grader it works the same way for research papers, lab reports and short-answer responses, as long as the rubric describes what you are grading. A paper can run to 20,000 characters, about 3,500 words; for anything longer, grade the sections separately — introduction and methods against one rubric, discussion against another — so each criterion reads the part it is about. The assignment field matters more for papers than for essays: paste the question or the lab's goal, and relevance and completeness are judged against it.
Reading the results table
One row per paper, one column per criterion, then the total. Click a row to open its report: the level wording for every criterion, a bar showing where between two levels the paper sat, and the length, paragraph and repetition checks the essay grader runs. The CSV holds student, criterion points, total, maximum and percent, ready to paste into a gradebook.
Writing a rubric the grader reads well
- Describe something observable: “supports the claim with specific evidence”, not “good evidence”.
- Put the standard in the description; the name is a label.
- One quality per criterion. “Organization and grammar” forces one score onto two things.
- Paste the assignment prompt, so relevance is judged against what was actually asked.
- Match the scale to the one you grade on, so a 3 here means what a 3 means in your gradebook.
- Before running a whole set, grade two papers you have already marked by hand and compare.
Where it helps, and where you decide
A first pass, not the last word
The AI grader is fastest as a first read: sorting a stack into likely levels, spotting a paper whose criteria disagree with each other, or checking that two sections were marked to the same standard. It does not know your class, the conversations behind an assignment, or what a particular student is working on, and it does not detect plagiarism or AI-written text. Review before any grade reaches a student, and look hardest at papers whose bars sit near the middle of a level.
Students grading their own drafts will find the essay grader more useful: one essay, a letter grade, and a conversation for follow-up questions.
Student work and privacy
Submissions are graded and discarded; nothing is written to storage, and the CSV is built in your browser. The text does go to the model's API to be scored, so follow your district's policy on online tools, and label papers with initials or ID numbers rather than full names.
Questions
Can I use my district's rubric?
Yes. Paste each row as one line, the criterion's name then its description, and pick the matching point scale. Up to eight criteria and 3 to 6 points are supported.
How many papers can the AI grader take at once?
Thirty per run. Each is graded separately and in parallel, so a full set usually returns in a few seconds, and a paper's grade never depends on which papers came before it.
Which files can I upload?
PDFs with a text layer, DOCX and plain text. Each file becomes one submission named after the file. A scanned PDF has no text to read and is refused with a note, rather than graded as blank.
Do students see the grades?
No. There are no student accounts: the grades exist on your screen and in the CSV you download, and nothing is sent to anyone. What reaches students, and when, stays your decision.
Does it detect plagiarism or AI-written work?
No. It scores the writing against your criteria and nothing else. Use your school's plagiarism tool for that, and your own reading for anything the grader flags as unusual.