AI Marking You Can See, Set and Correct: Short Answers in Moodle and Canvas

A student comes to you after class with their phone out. Question 7 on the Module 4 quiz asked for the function of the mitochondria; they wrote “it makes the cell's energy”, the gradebook says zero, and their friend who wrote “produces ATP” got the mark. They want to know why. You open the attempt and look at it, and you cannot actually tell them: was that answer compared word for word, judged by the AI, or never really checked at all?

Automatic marking is only as trustworthy as your ability to explain it and correct it. A mark you cannot account for is worse than no automatic mark at all, because it spends a student's trust and your afternoon at the same time. This release of EduGears AI inside Moodle, Canvas and other LTI 1.3 platforms makes short-answer marking something you can see, set and correct — and stops the one outcome nobody should ever get: a student marked wrong because the AI simply wasn't available.

The mark nobody can explain

Short free-text answers are where automatic marking earns its keep and where it gets argued with. Multiple choice is unambiguous; essays go to a human anyway. Short answers sit in between: there is a model answer, but students phrase things their own way, and whether a paraphrase deserves the mark is a judgement. When that judgement is made invisibly, every disputed mark turns into a small investigation — and when the teacher cannot overturn it cleanly, into a workaround in the gradebook that the tool then knows nothing about.

Three things fix that: knowing how a mark was decided, choosing the standard it was decided against, and being able to change it in the place you mark. This release adds all three.

See how every mark was decided

Every short answer now records how its mark was decided:

• By exact match — the answer matched the expected answer.
• By the AI — the AI judged whether the answer means the same as the model answer.
• Without AI — the AI was unavailable when the answer was submitted.

Quiz Submissions shows this on the attempt, instead of failing silently. So when a student asks why question 7 went the way it did, the first half of the answer — how it was marked — is on the screen in front of you.

Set how strictly the AI marks

“Means the same as” is not one standard. In an introductory course, “it makes the cell's energy” may be exactly the understanding you are looking for. In a pharmacology or legal course, the precise term is often the point of the question. So each course's Curriculum Intelligence page gains an AI marking setting for short free-text answers:

Starting a new term from last term's course with “Copy setup from an existing course” brings the AI marking setting along too, so a course you set to strict stays strict.

Correct any mark

Whatever the setting, the AI will sometimes disagree with you, and your judgement is the one that counts. Each question on Quiz Submissions now has an Override:

1. Set the mark you believe is right.
2. Add a note if you want one — only instructors see it, so it can record your reasoning for a colleague or a moderation check.
3. Save. The attempt is rescored and sent to your gradebook again.

The student sees the corrected mark with your name on it — not an unexplained change, but a decision someone made and owns. Where the AI left a marking suggestion, it shows alongside the Override, so you can see what it thought before you decide. Back to the student with the phone: you can now show them how the answer was marked, and if you agree with them, change it on the spot, gradebook included.

When the AI can't check, it waits for you

Until now, a short answer the AI could not check on a graded quiz was recorded as not correct — conservative, but it meant a student could lose a mark for no reason other than timing. That has changed. When the AI can't check a short answer, it no longer marks it incorrect. The answer waits for you instead:

• Your student sees Awaiting teacher review rather than a wrong mark.
• Your gradebook shows the attempt as waiting.
• Quiz Submissions highlights each waiting answer — for example, “2 answers are awaiting your review — the AI was unavailable when they were submitted.”

You mark it with the same Override you already use, and the attempt is rescored and sent to your gradebook. No student is silently marked wrong just because the AI was unavailable at the moment they submitted, and no waiting answer gets lost, because the screen you mark on tells you it is there.

Did the score reach the gradebook?

The other half of most mark disputes is the gradebook itself: the student sees one number in the tool and another, or none, in Moodle or Canvas. Quiz Submissions now ends each attempt row with a small status:

Until now that information was stored and never shown, so the only way to check was to open the LMS gradebook and compare by eye. The status is read-only and deliberately offers no retry button — this is the screen you read while marking, not a control panel. Where a launch has no gradebook connection at all, the cell simply stays empty. And on such a launch the result banner no longer says “Time expired”; it says “Not connected to a gradebook — score not sent.”

Spotting who needs help, early

Marks are one signal among several. Your Dashboard now has an At-risk students card that flags learners who have gone inactive, are scoring low, are behind pace, or are having most of their AI tutor questions declined — and shows the reason each one was flagged. It opens the At-Risk Learners report for the full picture, which now includes each learner's tutor questions and decline rate.

The thresholds are yours. Change the filters on the report, choose “Save as course thresholds”, and from then on the Dashboard card and Cog use your figures for that course — so “scoring low” means what it means in your subject, not a number someone else picked.

FAQ