AI that assists. Teachers who decide.
Every grade the AI suggests is reviewed, overrideable, and only final when you lock it.
No credit card required | Teacher approval required | Plans start at $5/month
The Grade Harbor AI promise
- AI suggestions are never final grades.
- Teachers can edit or override every score.
- Feedback is not sent to students automatically.
- Student essays are not used to train AI models.
- Model versions are recorded in grading history.
How the AI grades
Plain-language explanation of the grading pipeline — no jargon.
Reads your rubric criteria
The AI receives your rubric — your criteria, your point values, and your anchor descriptions. It grades to your standard, not a generic one.
Scores each criterion independently
Every criterion is evaluated on its own. The AI does not produce a single holistic score — it works through each dimension of your rubric one at a time.
Writes a justification per criterion
For each score, the AI writes a plain-language explanation grounded in specific evidence from the essay. Not just a number — a reason.
Generates overall feedback
After scoring all criteria, the AI drafts a brief holistic feedback message summarizing the essay's strengths and the clearest area for improvement.
Teacher reviews and approves
You see every score and every justification. Agree with them, edit them, or override them entirely. Nothing is final until you lock the grade.
What the AI can and cannot do
What the AI can do
- Score essays against a rubric criterion-by-criterion with written reasoning
- Generate written feedback grounded in specific essay text
- Flag scores that need closer review (low confidence)
- Identify patterns across a class — which students struggle with which skills
- Suggest groupings and instructional priorities based on skill profile data
What the AI cannot do (and does not try to)
- Make a grade final without teacher review — the system prevents this
- Assess tone, intent, or context that requires knowing the student
- Evaluate content accuracy outside writing craft (e.g., whether a cited fact is true)
- Replace the teacher's professional judgment about a student's situation
- Communicate with students — only the teacher can share feedback
The human-in-the-loop guarantee
Every grade requires your review.
AI-generated scores are proposals. They cannot be shared with students or exported until you have reviewed and locked each grade.
We built it this way on purpose. The AI handles the grading volume. You make the decisions.
What happens to student essays
We know this is the question that matters most to teachers and administrators. Here is a plain-language answer.
- Essays are sent to our AI API for grading and are not used to train models.
- Essays are stored securely in our system tied to your account, accessible only to you.
- We never use student essay content to train AI models — ours or anyone else's.
- You can delete a student's data at any time. We will delete it within 30 days of your request.
- We act as a "school official" under FERPA — the same category as your gradebook vendor.
See our current FERPA Notice and Privacy Policy for compliance details.
About the AI model
- We use an AI API for grading and feedback generation.
- The model version is configurable and documented in every grade record — you can always see which model produced a given grade.
- We use the standard API — not a fine-tuned model trained on student data.
- When we change the model used for grading, we document it and retain the prior model version in historical grade records.
We do not make specific claims about model accuracy that we cannot verify. Rubric-based AI grading is assistive; it is not infallible.
Confidence scores
When the AI is less certain about a score, it says so. Scores that need closer review appear at the top of the review queue so you can look at them first. The AI tells you why it was uncertain — for example, The essay addresses the criterion but in an unconventional way.
High-confidence scores can be reviewed quickly. Scores that need closer review deserve closer attention. The teacher always decides.
What teachers should review carefully
The AI is less reliable in these situations. Spend extra time on any grade that falls into one of these categories.
- Scores for unconventional writing styles
- Essays from multilingual learners
- Very short or incomplete submissions
- Assignments requiring outside factual accuracy
- Any score marked "Needs closer review"
Glossary
- Needs closer review
- A teacher-facing label for low confidence. It means the AI found limited or conflicting evidence and wants you to review that score carefully before finalizing.
Questions and concerns
- “What if the AI is wrong?”
- Override it. The AI's score is a proposal. Your override is permanent. The system records both the AI score and your final score in the grade history.
- “Is this fair to my students?”
- You're the judge. The AI gives you a structured starting point with reasoning. You decide whether it's right for each student.
- “Does the AI have biases?”
- Rubric-based AI grading can reflect limitations in both the rubric and the model. Grade Harbor is designed to make the reasoning visible so teachers can catch issues, review patterns, and override any suggestion that does not fit their professional judgment.
- “What if I don't agree with how the AI works?”
- Contact us. We want teachers to trust the tool, and we'd rather hear your concern than lose your trust silently.
See how it works in practice
Start a free trial and run your first grading session today.