AI code screening for teachers who grade real code

The grading workflow, end to end

Most integrity questions do not arrive one at a time. They arrive as a deadline, a pile of submissions, and a feeling that something in the pile is off. Here is the workflow this tool is built around:

1. Collect the set. Take submissions the way you already do — LMS export, shared folder, a zip from your TA. If it is a folder of source files, it scans.
2. Scan the batch. Upload the set as a ZIP. Each file gets its own AI-likelihood score, so the pile comes back sorted — the files that deserve attention rise to the top instead of hiding in the middle.
3. Read the report, not just the score. Every flagged file comes with highlighted snippets: the specific passages the model found unusual. Read those yourself before you read anything into the number. A score sorts the pile; the snippets tell you why.
4. Follow up where it matters. For the files that still concern you after your own read, move to a conversation — a walkthrough, a quick viva, a "tell me how this works" in office hours. The report shows where to spend that time; it never makes the call.

Same loop whether you grade 30 files a week or 300 a semester: sort, read, follow up, document.

Batch scanning a class set in practice

A few practical notes from how the batch flow works:

• One ZIP is enough. Zip the assignment folder and upload it as-is. Files are scanned independently and reported per file, so one student's work never contaminates another's score.
• Free covers a first look. Run single files on the free tier and see exactly what a report looks like. Batch scanning — up to 50 files in one run — is the Pro workflow, sized for a full section.
• Every file gets the same treatment. Each submission is scored on its own, with its own highlighted evidence. The batch view exists to sort, not to compare students against each other.
• Reports export. Download the report as PDF or CSV and file it with your grading records. If a review conversation happens weeks later — or a grade appeal lands next semester — you have the same document you worked from on grading night.
• Nothing is auto-sent. No student, no integrity office, no department chair gets notified. The report is yours to use inside your own process.

If you want to see the output before scanning your own files, there is a sample report from a full assignment set in the demo library — same layout, synthetic code.

When one suspicious file becomes forty

The hard part is not spotting one odd submission. The hard part is doing it across a full class set while still being fair to the student who wrote unusual but honest code. A screening tool can sort the pile. It cannot tell you intent.

Most intro courses live in a couple of languages — see what the tool flags when screening Python submissions or Java assignments.

A flag is not an accusation

AI code detectors can produce false positives and false negatives. That is why the report is built around review: what was flagged, which snippets mattered, and what a human should check next. If a student can explain the code, that explanation matters more than a score.

See how reports work →

A scan is step two; step one is the signals you can read without any tool.

How to talk to a student about a flagged file

The conversation is the part no tool can do for you — and the part most likely to go wrong. Handled well, a flagged file becomes the same kind of code walkthrough good courses already run.

Before the meeting

• Read the flagged snippets yourself, slowly. If you cannot say what specifically looked unusual, you are not ready for the conversation.
• Pull the student's earlier work. A sudden shift in style is a question; a consistent style is usually an answer.
• Check the assignment itself. Starter code, required libraries, and very short programs can all push scores up on honest work.

In the meeting

• Open with the work, not the tool: "I wanted to walk through your submission — can you explain how this part works?"
• Ask questions only the author could answer: why this approach, what this variable does, what they tried first that did not work.
• Let them talk. A student who wrote the code usually lights up; a student who cannot explain it tells you more than any score.
• Keep the detector in the background. If it comes up at all, it is "one thing that made me look twice," never the verdict.

After

• Write down what you reviewed and what the student explained — your notes are the record, not the tool output.
• If concern remains, route it through your institution's normal integrity process. The screening report is input to your judgment, not evidence that stands on its own.

For the full step-by-step, see How to check student code for AI use →

Fitting screening into your academic integrity policy

Most institutions already have an academic integrity process. This tool is built to sit inside it, not around it:

• It produces screening signals, not findings. A score is a reason to look closer, nothing more. Authorship, intent, and sanctions stay with instructors and your institution's review process.
• It documents your review, not just its output. The exportable report shows what was flagged and why — your file shows a human reviewed the work before any conversation happened.
• It respects disclosure norms. If your policy requires telling students their work may be screened, the syllabus note below gives you the language. Transparency up front is easier to defend than a surprise later.
• It never escalates on its own. Nothing auto-files, auto-emails, or auto-accuses. Every next step is a human decision inside your normal process.

If your department is still writing its AI policy, the limits page lays out what these tools can and cannot tell you, in plain terms. Read the limits →

What this tool should not do

It should not auto-accuse a student, auto-email an integrity office, or turn a score into a discipline recommendation. It should help a teacher find the files that deserve human attention.

See the limits page

Add a clear note before you scan

If your course uses AI screening, say so early. Tell students what you review, what counts as follow-up, and how they can explain or revise flagged work. A clear policy lowers fear and makes the review process easier to defend.

Get the report template

Scan the next assignment before you grade the whole pile

Run a free check on one file, then move to batch review when you need per-student reports.