What are the grounds for appealing an AI detection finding?

Most institutions accept a narrow list: procedural irregularity, new evidence not available at the time, or a penalty disproportionate to the finding. "The detector was wrong" is usually not itself a ground — it has to be framed as one of those, most often as evidence that was not properly considered. Read your institution's stated grounds before writing anything, because an appeal filed on a ground the regulations do not recognise fails without reaching the substance.

How long do I have to appeal?

Institutions set their own deadlines and they are typically short — often ten to twenty working days from the written decision, sometimes fewer. The exact figure is in your institution's regulations, and it is the single most important thing to establish on the day you receive the outcome. Missing it usually ends the matter regardless of the merits.

Should I get help before filing?

Yes, and it is usually free. Most students' unions and student advice services employ people who handle these cases routinely, know the local regulations, and have seen how the panel in question behaves. They are the single most useful resource available to an accused student, and consistently the most underused.

Can I appeal the penalty rather than the finding?

Usually, and they are separate arguments. Accepting that a rule was breached while arguing the penalty is disproportionate is a legitimate and often stronger route than contesting the finding itself — particularly where the breach was procedural, such as undisclosed but permitted AI use.

Does it help to cite research about detector accuracy?

It can, if it is specific and correctly attributed. A vague claim that detectors are unreliable carries little weight; a named study, with what it tested and what it found, carries more. Take care that the study you cite actually tested the detector in question — the most-quoted false-positive figure comes from research that did not include Turnitin.

What if the appeal fails?

Most systems have a further stage — an internal review, and in some countries an independent external body that reviews process rather than academic judgement. Those bodies generally require the internal route to be exhausted first, which is another reason to file the internal appeal properly and on time even when it feels unlikely to succeed.

Appealing an AI detection finding

Establish your deadline before you do anything else — it is usually short, it starts from the written decision, and missing it ends most appeals more reliably than any argument about detectors. Then find your students' union adviser. Then build the evidence.

Last reviewed 15 August 2026 · The HumanFlow team

This page is about procedure. If you have just been accused and want to know what the number means and what to say first, read falsely accused of using AI — this one picks up after a written decision, when there is a formal process to run.

Regulations differ by institution and by country, so what follows is the shape most systems share rather than any one set of rules. Yours will be published, usually under "academic misconduct", "academic integrity" or "student appeals", and that document beats anything on this page where the two disagree.

Two systems are written up here in full. If you are at a UK university, the stages a UK appeal actually runs through replaces the generic version below, and nobody in UK higher education mandates AI detection — which matters when a panel describes running one as an obligation. If you are in India, the regulation your department is most likely citing is a plagiarism rule: the UGC's similarity bands, and why AI writing sits outside them.

The stages, and what each one decides

1. Informal resolution. A conversation with the marker or module lead. Many cases end here, particularly where the work is clearly yours and the score was the only thing driving suspicion. Bring drafts.

2. Formal investigation or panel. A decision on whether a breach occurred, taken by people rather than by software. This is where evidence about how the work was produced does its real work.

3. Appeal. Not a re-run of stage two. An appeal usually asks a narrow question — was the process correct, is there new evidence, is the penalty proportionate — rather than re-deciding the case from scratch.

4. Final review, then external. An internal completion stage, after which some countries provide an independent body that examines process rather than academic judgement. These almost always require the internal route to be exhausted first.

The evidence pack

Assemble this before you write the appeal, because what you have determines what you can argue. In rough order of weight:

Version history. Google Docs, Word or a repository showing the document growing over time. This is the strongest single item, because fabricating it after the fact is impractical — which is exactly why it persuades.

Drafts, notes and dead ends. The paragraph you cut, the outline you abandoned, the source you decided against. Machine-produced work does not usually come with a trail of discarded thinking.

Your other writing. Earlier submissions establish what your prose normally looks like. That comparison has a name — stylometry — and it is a stronger form of evidence than any detection score, because it compares your work against you rather than against writing in general.

The report itself. Ask for the full originality report, not a summary. Three of its rows point at evidence you can open; the AI row does not, and establishing that in writing is useful.

A conversation, if one is offered. A viva voce — being asked to talk through your own work — examines what you understand rather than what a percentage implies, and it is usually the fairest route available to you. Ask for one if it is not offered.

Alongside the evidence, check the procedure. Due process is the set of steps owed to you before a decision is taken — the allegation in writing, the evidence shown, a response invited before rather than after a decision is recorded, and an appeal route with a deadline. Those steps either happened or they did not, which makes them firmer ground than arguing about a probability. Where an institution has moved to supervised assessment instead, proctoring at least produces a recording you can ask to see, which a detection score never does.

What actually carries weight

The strongest available argument is usually not that detectors are unreliable in general. It is that this finding rests on a measurement the vendor itself says should not carry it. Turnitin states its AI indicator is not intended as the sole basis for an academic misconduct finding — a sentence worth quoting exactly, with the source.

The second strongest is institutional precedent. Vanderbilt disabled Turnitin's AI detection and published its reasoning; Washington State University's provost cancelled its use. Neither binds your institution, and both demonstrate that treating these scores cautiously is a mainstream position rather than a student's excuse.

Be precise with research. The widely-quoted finding that seven detectors flagged 61.22% of essays by non-native English speakers is real, peer-reviewed, and did not include Turnitin — so quoting it against a Turnitin finding invites a correction that costs you credibility at the moment you need it. Cite what tested the tool in front of you, or cite the general point and say so.

Where we stand, since we sell one of these tools

We make an AI humanizer and an AI detector, so read this page knowing that. It exists because the same distress that sends people looking for a bypass tool also sends them looking for help with a process nobody explained to them, and the second need is the one we can answer honestly.

Nothing we sell helps with an appeal. If you used AI in a way your course prohibited, this page will not change that, and the advice above assumes the work is yours.

Common questions

What are the grounds for appealing an AI detection finding?: Most institutions accept a narrow list: procedural irregularity, new evidence not available at the time, or a penalty disproportionate to the finding. "The detector was wrong" is usually not itself a ground — it has to be framed as one of those, most often as evidence that was not properly considered. Read your institution's stated grounds before writing anything, because an appeal filed on a ground the regulations do not recognise fails without reaching the substance.
How long do I have to appeal?: Institutions set their own deadlines and they are typically short — often ten to twenty working days from the written decision, sometimes fewer. The exact figure is in your institution's regulations, and it is the single most important thing to establish on the day you receive the outcome. Missing it usually ends the matter regardless of the merits.
Should I get help before filing?: Yes, and it is usually free. Most students' unions and student advice services employ people who handle these cases routinely, know the local regulations, and have seen how the panel in question behaves. They are the single most useful resource available to an accused student, and consistently the most underused.
Can I appeal the penalty rather than the finding?: Usually, and they are separate arguments. Accepting that a rule was breached while arguing the penalty is disproportionate is a legitimate and often stronger route than contesting the finding itself — particularly where the breach was procedural, such as undisclosed but permitted AI use.
Does it help to cite research about detector accuracy?: It can, if it is specific and correctly attributed. A vague claim that detectors are unreliable carries little weight; a named study, with what it tested and what it found, carries more. Take care that the study you cite actually tested the detector in question — the most-quoted false-positive figure comes from research that did not include Turnitin.
What if the appeal fails?: Most systems have a further stage — an internal review, and in some countries an independent external body that reviews process rather than academic judgement. Those bodies generally require the internal route to be exhausted first, which is another reason to file the internal appeal properly and on time even when it feels unlikely to succeed.

Sources

1.Turnitin's AI writing detection capabilities FAQs — Turnitin Guides, 2026
2.Guidance on AI detection and why we're disabling Turnitin's AI detector — Vanderbilt University, Brightspace blog, 2023
3.Cancellation of Turnitin AI Detection Software (memo to instructors) — Washington State University, Office of the Provost, 2026
4.GPT detectors are biased against non-native English writers — Liang, Yuksekgonul, Mao, Wu & Zou — Patterns (Cell Press), 2023

Last reviewed 15 August 2026.