In Examplary, the AI suggests a grade for every open answer: points for each criterion in your rubric, with a short explanation and feedback for the student. You review the suggestion and accept or change it, so the judgment stays yours. Closed questions are graded by Examplary itself, without AI.
Grading open questions takes time. With AI grading, you start each answer from a suggestion you can check, since you can see why the AI made its choice for each criterion.
Which answers get a suggestion
| Type of question | Who grades it | Then |
|---|---|---|
| Closed questions, like multiple choice or true/false | Examplary, using the correct answer you set | Graded right away |
| Open questions scored with Exact values | Examplary, by comparing the text exactly | Graded right away |
| Other open questions | The AI suggests a grade | (auto-graded) until you review it |
| Question types without AI grading, like Javascript | You | Needs review |
The AI starts on its own as soon as a student saves an answer, so there's nothing to click. If the student changes the answer, grading starts again. While the AI is working, you'll see Auto-grading…. Scanned answers from a paper test go the same way.
What does the AI look at?
For every answer, the AI is given:
- the question, with the instructions that come with its question type;
- the rubric, the model answer and any grading guidance;
- the facts from your source material that belong to this question;
- the level and language of the test;
- your Grading and feedback AI instructions from the test settings, and the test creator's AI preferences, such as the tone and length of feedback;
- up to eight answers to the same question that you've already graded;
- the student's answer, without their name.
What the AI suggests
For each criterion, the AI suggests points, with a short explanation and how confident it is. Usually that's full points or none, sometimes half. It also writes feedback for the student: what was wrong, why, and how to improve.
The confidence shows as a percentage next to each criterion. Hover over it to read the AI's reasoning and its confidence score. It's the AI's own estimate, and Examplary never shows more than 95%. A low percentage is a reason to look more closely. Nothing else happens with it: Examplary never accepts a suggestion on its own.
Does the AI learn from my grading?
Yes, but only within one question. Every answer you grade yourself or accept becomes an example for the same question in the same test. From then on, whenever the AI suggests a grade for that question, it gets up to eight of those examples, with high and low scores. That's how it picks up how strict you are and how you give partial credit.
This helps with answers that come in later or are graded again. Suggestions that are already there don't change by themselves. No AI model is trained on your examples.
Grading answers again
Examplary never overwrites what you graded yourself. You can have auto-graded answers graded again in three ways:
- Change the rubric with Edit rubric in Grade test.
- Change your instructions in Settings → Grading & feedback and click Re-grade AI scores.
- Change a single answer with Edit answer and turn on Auto-grade again.
What you put on the AI context tab is used when the AI writes questions, not when it grades. Grading instructions go in Grading and feedback AI instructions. Providing feedback instructions explains how to write them.
What students see
During a test, students only see a grade if Give feedback after each question is on, as in a practice test. They then see the AI's suggestion straight away, before you've looked at it. After the test, they see nothing until you publish the results.
If you publish results before everything is reviewed, students see the AI's suggestions as if they were final. Only the final grade waits until every answer is graded. See Reviewing answers and publishing results.
What the AI doesn't do
- Grade closed questions and exact values. Examplary does that without AI.
- Change your rubric, accept suggestions or publish results.
- Calculate the grade. Your test's grade calculation does that.
- Learn from your tests or your students' answers outside your own test. No model is trained on any data.
The EU AI Act counts AI that helps assess students as high-risk, and high-risk systems need human oversight. That's why the AI in Examplary only suggests, and you decide. EU AI Act compliance explains how Examplary works toward compliance, and Data residency and backups covers where the AI processing happens.