It Proposes. You Decide. And It Shows You Why.
GradingPal never applies a mark on its own. It suggests one against the criterion you wrote, attaches the sentence from the student’s work that it judged, and waits. Everything below is a description of how that is built — including the things we have decided not to build, and why.
No claim on this page is one you have to take on trust. Every one of them is checkable in the product.
Five Things That Are Always True
Not principles. Mechanisms — the ones you can go and check.
- Every suggested score has the evidence attached
- When GradingPal proposes a mark against a criterion, the sentence or passage it judged is attached to that criterion. You are never asked to accept a number on its own. This is the one to look at first, because everything else on this page rests on it.
- It proposes. It never applies
- Nothing reaches a student, a gradebook or a report until a teacher has approved it. There is no setting that turns this off, because there is no version of the product in which a mark is applied without a person.
- Your rubric is the standard, not ours
- GradingPal judges work against the criteria you wrote, in your wording. It does not hold an opinion about what good writing is; it holds your criteria and looks for them.
- Student work is not used to train models
- What your students write is not used to train AI models, ours or anyone’s. How the data is handled is set out in full on our security page.
- You can always see what changed
- When you override a suggestion, the record keeps both — what was proposed and what you decided. Nothing is quietly replaced.
Four Things We Have Decided Not to Build
Some of these get asked for often. Here is why the answer is no rather than not yet.
AI-writing detection
We do not detect whether a student used AI to write something, and we are not going to. The tools that claim to are unreliable in a specific and damaging way: they are wrong more often about students who write in a second language, and about students who write plainly. A false accusation of cheating is a serious thing to do to a sixteen-year-old on the strength of a probability score.
What we can do is the thing underneath the question. If you suspect a piece of work is not a student’s own, the useful move is not a verdict — it is a conversation about the work itself. GradingPal gives you the criterion, the passage, and what the writing did and did not do, which is a better basis for that conversation than a number neither of you can interrogate.
- Predicting test scores
- We do not predict how a student will do on a state test, an exam or a board assessment. What we show is what a student demonstrated on the work you set. A prediction is a different question with a different method, and it is not one we answer.
- Marking without a rubric
- GradingPal will not grade work against nothing. If there is no criterion, there is no judgement to make and no evidence to attach — so it asks for one rather than inventing a standard.
- Grading a student rather than the work
- Nothing here forms a view about a student. It reads one piece of work against one set of criteria, each time, and the picture that builds is a record of work rather than a judgement about a person.
The Number Your Department Reports Is Arithmetic
Standards mastery — the figure that goes into a report, a review or a conversation with a parent — is not produced by a model. It is a calculation over marks a teacher has already approved, using a rule you can read and change. Run it twice and you get the same answer, because it is arithmetic rather than inference.
That distinction matters more than it sounds. It means the number a department reports upward is reproducible, explicable to somebody who does not trust AI at all, and unchanged by anything we do to the model. The roll-up rules are written down, and you choose which one applies.
When It Can’t Read Something, It Says So
Handwriting that defeats it, a photograph taken at an angle, a page half out of frame, a diagram where the answer is in the drawing — these happen, and the honest thing to do with them is not to guess. Work it cannot read properly is flagged for you rather than marked anyway.
A tool that quietly produces a low mark for an unreadable page is worse than one that admits the limit, because the mark looks like a judgement about the student and is actually a judgement about the photograph.
The Mark Is Yours. So Is the Responsibility.
We are not going to tell you that the software is accountable for a grade, because it cannot be. A mark that goes on a report was given by a teacher who approved it, and if a parent asks why, the answer has to come from a person who can explain it.
What the product owes you is the material for that explanation: the criterion, the evidence, and a record of what you changed. When a grade is questioned, you should be reading back what happened rather than reconstructing it.
Don’t Take Our Word for It
Everything above is checkable, and here is where to check it. How the marking performs — the testing, the method, and what it is measured against — is on our research page. What happens to student data, including the training question and our compliance position, is on our security page.
And the fastest check is the product itself: grade one assignment and look at what it attached to the mark.
Frequently Asked Questions
Does GradingPal detect AI-written work?
No, and it is not something we intend to build. Detection tools are unreliable in a way that falls hardest on students writing in a second language, and a false accusation of cheating is a serious thing to do on the strength of a probability score. What GradingPal gives you instead is the criterion, the passage and what the writing did — which is a better basis for a conversation than a verdict.
Can AI grading be trusted with real grades?
Only in the sense that a teacher can be. GradingPal proposes a mark against the criterion you wrote and attaches the sentence it judged; nothing reaches a student until you approve it. The grade is given by a person who can explain it — the software’s job is to make that explanation easy to produce.
How do I know why it gave a particular score?
Because the evidence is attached to it. Every suggestion carries the passage from the student’s work that it was based on, against the criterion it was judged by. If the evidence does not support the mark, you change the mark — and the record keeps both.
Is student work used to train AI models?
No. What your students write is not used to train AI models, ours or anyone else’s. Our security page sets out the data handling in full, including our compliance position.
Does it remove bias from grading?
No tool can claim that, and we do not. What GradingPal does is narrower and checkable: it applies the same written criteria to every piece of work in the same wording, and it shows you the evidence behind each judgement so you can see when one looks wrong. Consistency is not the same thing as fairness, and a teacher still decides.
Will it replace teachers?
It cannot, because it does not do the part that requires a teacher. It reads work against criteria and proposes; it does not decide, does not know the student, and does not carry the responsibility for the grade. What it removes is the part of marking that is transcription rather than judgement.
What does GradingPal refuse to do?
Four things: detect AI-written work, predict how a student will do on a test, grade against no rubric, and form a view about a student rather than a piece of work. Each is a decision rather than a gap, and the reasons are on this page.
Grade One Assignment Free
Mark something you have already marked, and look at what it attached to each suggestion.