ScribeLens shows which sentences triggered the signal and why, so a conversation with a student can start from specifics instead of an accusation.
A number like "82% AI" tells you nothing you can act on. It does not say which paragraph looked unusual, what pattern caused it, or how confident the analysis actually is.
ScribeLens reports the same overall interpretation, but attaches sentence-level signals and reason keywords underneath it. If only the introduction and conclusion are flagged while the body reads as human, that is a very different situation from a uniformly flagged document — and you can see the difference immediately.
No AI detector is proof on its own, so every flag comes with its reason, ready for a conversation rather than an accusation.
The hardest part of suspected AI use is the discussion that follows. Sentence-level evidence changes that discussion from "a tool says you cheated" to "these three sentences show patterns worth talking about — walk me through how you wrote them."
That framing is fairer to students and far more defensible for you.
A result does not tell you a student cheated. AI detection is pattern analysis, not proof, so ScribeLens shows the sentence-level evidence and reason keywords behind every result. Detector v4.0 was retrained on real student essays and on essays by non-native English writers, and in our own testing it is markedly better at recognising such writing as human. Every flag still shows its sentence and its reason, so you can check it yourself.
Treat the output as one input alongside draft history, prior submissions, and a conversation with the student. A policy built on a score alone, with no evidence attached, is hard to defend and unfair to the student it's applied to.
Be confident about what detection actually contributes, because it's real: AI writing leaves measurable statistical patterns — uniform sentence rhythm, predictable word choice, formulaic paragraph structure — and ScribeLens is built to surface exactly those patterns, sentence by sentence, with reason keywords naming what triggered each one.
Be equally clear about what it doesn't contribute. It doesn't identify which AI tool was used, doesn't check plagiarism, and a result is a probability for you to weigh, not a finding that settles a case on its own. Treat it the way you'd treat any other piece of evidence: alongside drafts, prior submissions, and the student's own account, feeding into a decision that's still yours to make.
Start with the graded verdict, then open the sentence view before you treat the headline as the whole story — that's where the actual evidence lives.
You don't need a formal policy overhaul to use a detector fairly — a short, consistent sequence applied to every result is enough:
With Pro, batch upload checks up to 30 essays in one go (PDF, DOCX or TXT) and gives each one a full report with the same sentence-level evidence. The results are listed in upload order, never ranked, and you can download them as a CSV.
Save each set as a class assignment in My classes and come back to it later. Each assignment has a short class summary: how many reports are worth a closer look, how many are mixed and how many lean human, with every report one click away.
Saved classes stay in your own browser, on your own computer, so students' work is not sent to ScribeLens for saving. Schools can try all of this free for a term.
The difference between a fair process and an unfair one usually isn't the tool — it's how the result gets used once you have it.
Not every scan lands on a clear verdict. ScribeLens reports graded outcomes — including "mixed signals" and "blend of human and AI writing" — for exactly this reason: some drafts genuinely don't show a uniform pattern, and forcing that into a single confident label would be less honest, not more.
An uncertain result is neither exoneration nor confirmation. Look at whether any individual sentences carry flags even inside an otherwise mixed document, and weigh the same context that matters for any result, such as formal register or heavy editing. Treat an uncertain result as a reason to rely more on the conversation and less on the score, not as a reason to escalate.
However a result comes back, the conversation is the part that actually matters. Lead with the specific sentences and reason keywords, not the headline number, and ask the student to explain how they wrote those passages before you draw any conclusion.
For a full workflow — reading the evidence, weighing drafts and context, and having the conversation itself — see how to use an AI detector fairly with students.
Every sentence in a submission is classified individually as human-like, mixed, or AI-like, shown against a graded overall verdict. Flagged sentences carry reason keywords naming the specific pattern behind them, so you can see what actually triggered a result instead of working from a single number.
No. Use it as supporting evidence only. Final decisions should include revision history, sources, prior work, and a conversation with the student.
Detector v4.0 was retrained on real student essays and on essays by non-native English writers. In our own testing it is markedly better at recognising such writing as human. Every flag still shows its sentence and its reason, so read that evidence, not just the overall number, before drawing a conclusion.
No. Without an account you can check up to 2,000 words per scan. A free account gives you 35,000 words a month and 9 PDF/DOCX report exports, and it scans a whole draft in one go.
Yes, with Pro. Batch upload takes up to 30 files (PDF, DOCX or TXT), checks them one after another and gives each a full report, using your normal monthly words. You can save the set as a class assignment in My classes to see a class summary later, and download the results as a CSV. The list is never ranked.
Yes. Reference lists, citations, and publisher boilerplate are set aside automatically so they do not distort the verdict.
Yes. This page and the review workflow above are built around a classroom use case: reading sentence-level evidence on student drafts, not only a score. With Pro, teachers can also check up to 30 essays at once with batch upload and keep each set as a class assignment in My classes, and schools can pilot it free for a term. The underlying detector is the same one students use to self-check their own work before submitting.
Paste any student draft and see the sentence-level signals before you draw a conclusion. Without an account a scan covers 2,000 words; a free account scans the whole draft.