Essay grading has a math problem no teacher escapes: thirty students, ten minutes of careful reading each, and the stack is five hours deep. Assisted rubrics in Pear Assessment are built for that stack. You define (or generate) a rubric, the AI reads each response against it and proposes a score with feedback, and you review, adjust, and release — turning an evening of grading into an hour of verification.
This guide explains how the rubric machinery works, what the AI can and cannot grade, how accurate its suggestions honestly are, and how to stay firmly in charge of every score your students receive. It builds on our platform-wide overview of Pear Assessment AI features, focusing here on the grading workflow itself.
how rubrics work for grading essays in pear assessment
A rubric is a scoring grid: rows are criteria (what you are judging), columns are performance levels (how well), and each cell describes what that level looks like for that criterion. In Pear Assessment, essay and constructed-response questions can carry a rubric, and assisted rubrics add two accelerators to the classic setup.
First, generation: instead of typing every cell from a blank grid, you can have the platform draft a standards-aligned rubric from your question — criteria and level descriptors written for you, ready to edit. This matters more than it sounds. A rubric’s quality is decided in its descriptors: vague cells (“good analysis”) produce vague grading by humans and machines alike, while observable cells (“cites at least two pieces of textual evidence and explains each”) produce consistent ones. The generated draft gives you a professional starting structure; your edits aim it at your prompt, your grade level, and your students.
Second, application: with a rubric attached, grading a response means judging it criterion by criterion rather than awarding one holistic impression. That structure is what makes assisted scoring possible — the AI evaluates against the same explicit criteria you would, which is also why improving the rubric improves the AI. Whatever your own habits, keep the rubric visible to students before they write whenever the assignment allows it; a rubric is a teaching document first and a grading document second.

can the ai grade open ended questions in pear assessment
Yes — with a precise definition of “grade.” For essay and open-ended questions with an assisted rubric in place, you can enable AI auto-evaluation: the system reads each submission against the rubric and proposes a score per criterion, typically with draft feedback explaining the judgment. Those proposals come to you for review. You accept, adjust, or override any of them before scores reach students, and your edits — not the machine’s first pass — are what get recorded.
The sensible mental model is a teaching assistant who pre-reads the stack. A good TA marks the obvious strengths and gaps, flags the unusual papers, and leaves the decisions to you. Used that way, assisted grading attacks the two worst parts of essay marking: the blank-page start on every paper, and the inconsistency that creeps in by paper twenty-five, when human graders are tired and their standards quietly drift.
The workflow in practice: build or generate the rubric, attach it to the essay question, enable AI evaluation in the assignment settings, and after submissions arrive, work through the review queue — scanning the AI’s proposed scores and feedback, spot-reading the responses where the score surprises you, adjusting as needed, and releasing. Pairs naturally with the drafting tools in our guide to the Pear Assessment AI question generator: AI drafts the questions, assisted rubrics draft the grading, and you referee both.
how accurate is ai rubric grading in pear assessment
Honestly: good enough to draft, not good enough to trust blind — and its accuracy depends far more on your rubric and prompt than on any property of the AI itself. Three patterns show up consistently in AI-assisted scoring, here and everywhere:
It is most reliable at the extremes. A clearly excellent response and a clearly off-task one are easy for any rater, human or machine. Expect strong agreement there.
It is least reliable in the middle and at the margins. The B-minus/C-plus borderline, the unconventional-but-brilliant answer, the response that is technically thin but shows a real idea struggling to surface — these need a teacher’s knowledge of the student and the class’s journey. The AI has neither.
It inherits your rubric’s flaws. A vague criterion gets scored vaguely. An over-specific criterion (“mentions the word ‘furthermore'”) gets scored literally. Every systematic error teachers find in AI scores traces back, on inspection, to something the rubric said or failed to say.
So validate before you rely. The professional method takes one class set: grade ten responses yourself first, without looking at the AI’s proposals, then compare. Where you agree, note it. Where you differ, decide who was right — and fix the rubric where the rubric caused the gap. After two or three assignments of this calibration, most teachers know their error pattern precisely: where the AI runs generous, where it runs strict, and which criterion needs rewriting. That calibration, not faith, is what makes the time savings safe.
For final grades that must be adjusted after the fact — whatever the source of the first score — the mechanics are covered in our guide to how to change a grade in Pear Assessment.

can i see why the ai gave a particular rubric score
Yes — the “assisted” design assumes you will ask exactly that. AI evaluations come with their reasoning attached: the proposed score is presented per criterion, with feedback text tying the judgment to the rubric’s language, so you can see which descriptor the system believed the response matched and why. When a proposed score looks wrong, that per-criterion breakdown is where you look first: usually the AI has latched onto something real (the evidence is there but unexplained, the claim is present but unsupported) and weighted it differently than you would — a judgment call you are free to reverse.
This transparency serves a second purpose: student-facing feedback. The AI’s draft comments, after your edit, become the feedback students read. Starting from a rubric-anchored draft (“Your response states a claim but does not yet explain how the evidence supports it”) is faster than composing thirty comments from nothing, and because the language mirrors the rubric students saw before writing, the feedback loop closes neatly. Do edit for voice and knowledge of the student — the AI cannot write “this is the paragraph structure we practiced Tuesday, and you nailed it,” and those are the comments students remember.
Pear’s own product pages describe the intent behind these tools — AI that accelerates a teacher’s judgment rather than replacing it — in their overview of Pear Assessment’s AI capabilities, which is worth comparing against your experience after a few assignments.
Improving a weak rubric (the AI will thank you)
Because the AI applies your rubric literally, assisted grading exposes rubric flaws faster than any workshop. Watch for these three weak-cell patterns and their fixes.
The adjective cell. “Excellent analysis / good analysis / weak analysis” tells no rater — human or machine — anything. Rewrite with observables: what does the top level do that the next level does not? Count evidence, name the move (“explains how the evidence supports the claim”), describe the error pattern of the low level.
The everything criterion. A row trying to judge ideas, evidence, organization, and grammar at once forces a single score to mean four things; the AI will average invisibly and you will argue with it constantly. Split overloaded criteria into their own rows, even if you later report one combined score.
The absent floor. Many rubrics describe excellence lovingly and the bottom level as “missing or poor.” The AI needs the floor described too — off-topic, restates the prompt, no evidence attempted — because a large share of early submissions live there, and that is where first-pass accuracy matters most for feedback quality.
A practical editing pass: take three real (anonymized) responses from last year — strong, middle, weak — and score them against your draft rubric yourself. Any cell where your judgment and the words disagree gets rewritten before the rubric meets either students or the AI.

When not to lean on assisted grading
Keep human-first grading for the work where the score is the relationship: personal narratives, reflective writing, anything a student wrote about their own life, and high-stakes pieces where a grade will be contested. Use assisted rubrics where volume and consistency dominate — routine constructed responses, exit essays, practice drafts, reading responses across many sections. And whatever you automate, tell students the arrangement: “AI drafts, I decide” is a sentence that builds trust; discovering it later does the opposite.
Also keep a hand on department or school policy. Some schools welcome AI-assisted scoring with teacher review; others restrict it for particular assessments or require disclosure in course documents. The workflow above satisfies the strictest common version of those policies — human review before release — but your local rules are the ones that bind you.
FAQ
Do students see the AI’s original scores?
No — students see what you release after review. The AI’s proposals live in your grading view until you accept or adjust them.
Can assisted rubrics score handwriting or uploaded work?
The workflow is built around responses written in the platform’s text tools. Material submitted other ways should be graded with the rubric manually — the rubric still keeps your criteria consistent.
How many criteria and levels should a rubric have?
Three to five criteria and three to four levels cover most classroom writing. More rows than that and neither you nor the AI can hold the distinctions honestly; fewer and the feedback becomes too coarse to teach from. Single-point rubrics — one column describing proficiency, with space for notes on strengths and gaps — also work well with assisted feedback, since the AI drafts comments against the proficiency description.
Will using AI grading make my grades less defensible?
The opposite, if you calibrate: every score you release is yours, backed by a written rubric and feedback. Defensibility comes from the rubric and your review, not from who drafted the first pass.

Conclusion
Assisted rubrics in Pear Assessment pair the two things essay grading needs most: a well-written rubric and a fast first reader. Generate or build the rubric carefully, let the AI draft scores and feedback against it, calibrate its tendencies on your own graded sample, and keep final judgment — every time, on every paper — with the teacher. That division of labor is what turns a five-hour stack into an hour of professional review — and gives students their feedback while the writing is still warm.

