How Teachers Handle AI Scores on Resubmitted Work

You flagged a piece of Year 11 coursework a fortnight ago, had the conversation, agreed the student would redraft it, and now the resubmission is sitting in your marking pile. You run it through GradeOrbit again and the AI likelihood score comes back at 64%. Down from 91%, but still not low. So what does that number actually tell you, and what is the fair next step? This post is about the second time round — the resubmission — which is a different and trickier moment than the first flag, and one that existing guidance rarely covers.

The temptation on a resubmission is to treat the score as a pass/fail gate: high means they cheated again, low means they fixed it. Neither is true. A score on a redraft has to be read against the history of the piece, the conversation you already had, and your school's academic integrity policy on second attempts. The number is one input. The story around it is everything.

Why a Resubmission Score Reads Differently

When a student redrafts flagged work, the linguistic fingerprint of the original often survives into the new version. If they kept their structure, reused phrases, or worked from the same plan, the detection model will still see patterns it associates with AI-generated text — even if the student wrote every word of the redraft themselves this time. A score that drops from 91% to 64% can mean genuine rewriting that still carries echoes of the original, not a half-hearted patch job.

The reverse also happens. A student who panics after the first conversation might over-correct, deliberately roughening their prose, adding errors, breaking up sentences — and the score drops not because the work became more authentic but because it became less polished. A lower number is not automatically a cleaner conscience. This is exactly why a resubmission score must never be treated as a verdict on its own. It is a prompt to compare, not a conclusion to act on.

Compare the Draft and the Resubmission Side by Side

The single most useful thing you can do on a resubmission is read the two versions together. Open the original flagged piece and the redraft next to each other. Look at what actually changed. Did the student rework the argument, or did they swap a few words and reorder paragraphs? Are the sentences that carried the original's AI-like rhythm still there verbatim? Did the weakest parts — the ones a student usually struggles with — get genuinely harder and more personal, or smoother and more generic?

A student who has done the real work of redrafting can show you the difference. The new version reflects the conversation you had: a point you queried is now developed, an example you said felt thin is now specific. That continuity between your feedback and their changes is far stronger evidence of authentic effort than any movement in the likelihood score. A student who regenerated the piece a second time tends to produce a redraft that is different but not responsive — it changed, but not in the ways you asked for.

The Second Conversation Is About Process, Not the Number

If the resubmission still raises concerns, the second conversation should be calmer and more specific than the first. You are not re-litigating whether the original was AI-assisted — you have already had that talk. You are asking how they approached the redraft. "Talk me through what you changed and why" is the question that does the work. A student who genuinely rewrote can connect their changes to your feedback. A student who did not will struggle to explain why a particular paragraph reads the way it does.

Keep the score out of the opening. As with the first flag, lead with the work, not the percentage. The same principles that govern the first student conversation apply here, but with an added dimension: you are also assessing whether the student took the agreed process seriously. That is a behaviour question as much as an integrity one, and your tone should reflect that you are giving them a fair second hearing, not setting a trap.

Aligning a Resubmission Outcome With School Policy

Your school's academic integrity policy almost certainly distinguishes between a first instance and a repeat. Read the relevant section before the second conversation, not after. Some policies treat a flagged-then-corrected piece as resolved if the redraft satisfies you; others require any repeat concern on the same piece to escalate to the head of department regardless of the score; others have specific rules for NEA and coursework heading to an exam board, where a second instance may need formal reporting.

The likelihood score does not decide which path you take — the type of work and the policy do. What the score gives you is context to record. A note such as "resubmission scored 64% (down from 91%); redraft responsive to feedback; voice consistent with prior work; satisfied this is the student's own" is the kind of contemporaneous record that protects you if the outcome is ever questioned. So is the opposite: "resubmission scored 64%; changes superficial; student unable to explain key paragraph; escalated to HoD per policy." Either way, the decision is documented and defensible.

For coursework heading into moderation, log both scores in the pack. A moderator reviewing a borderline piece benefits from seeing that it was flagged, discussed, redrafted, and re-scored — that is a complete and honest audit trail. The moderation cycle is where this kind of context earns its place, as one input among many rather than a standalone judgement.

When the Redraft Clears the Concern

Often the second conversation ends with you satisfied. The redraft is responsive, the student talks fluently about their changes, and whatever happened the first time, this version is theirs. Say so plainly. "I asked you to redraft this and you've clearly done the work — I'm happy with it" closes the loop and signals that the process was fair, not punitive. A student who has been through a flag, a conversation, and a redraft has earned a clean resolution when they have actually responded to it.

Mid-range scores cause the most second-guessing here, so it is worth being clear-eyed about them. A 64% on a responsive, well-explained redraft is not a reason to keep the case open. As covered in our guide on interpreting mid-range likelihood scores, the middle of the range is exactly where professional judgement, not the number, has to lead.

How GradeOrbit Supports the Resubmission Workflow

GradeOrbit's AI detection tool returns a likelihood score between 0% and 100% on every run, so you can re-score a redraft as easily as the original. There are two model choices: a 1-credit model for quick triage on lower-stakes pieces, and a 3-credit model for higher-confidence analysis on coursework or work heading for moderation. Running the same piece twice — original and resubmission — gives you the comparison this post is built around.

We never save student work. Both the original and the resubmission are processed in memory and discarded after the score is returned, so there is no stored history of a flagged student on a server. Teachers redact personal information before uploading — the canvas tool burns black boxes into the image — and students are anonymous to the model, referred to only as "Student 1", "Student 2", and so on. The record of the case lives in your markbook and moderation pack, where it belongs.

Try GradeOrbit for Fair AI Detection

If you want an AI detection tool that gives you a clear likelihood score on every run — first draft and resubmission alike — and trusts you to run a fair process around the number, GradeOrbit is built for exactly that. New accounts get a small allocation of free credits to try the detection and marking workflows on real student work.

Visit our homepage to learn more, sign up, and run your first detection. The tool exists to support your professional judgment across the whole life of a piece of work — including the second time round.