For students

AI detector for students: what to do when your own writing gets flagged

Detectors flag human essays. Ours flagged 9 out of 250 human-written samples in our own testing, and 4 out of 50 written by non-native English speakers. If that has happened to you, the position is defensible. This page is about the evidence that defends it, and how to keep it before you need it.

Free credits on signup, no card required. Up to 50,000 characters per scan, about 8,000 words, on every plan.

Key takeaways

  • A score is not proof of anything you did. It measures how predictable the text is. That is all it can measure.
  • 3.6% of human samples were flagged in our 600-sample study, 9 of 250. False positives are a known, published property of these tools.
  • 8.0% for non-native English writers, 4 of 50. If English is your second language you carry more of this risk than your classmates do.
  • Version history is the strongest defence there is. Draft in Google Docs or Word online and the timeline builds itself.
  • Check before you submit. Free credits on signup cover a scan, up to 50,000 characters, about 8,000 words.
3.6%Human samples wrongly flagged
8.0%Non-native English writers
0Scores that constitute proof
50,000Characters per scan, all plans
Why it happens

Human writing that gets flagged has things in common

A detector compares your text against the statistical shape of machine-written text. The closer the match, the higher the score. Nothing in that process looks at who wrote it, so the writing habits that raise your score are the same ones your rubric usually rewards.

Consistent, even sentence rhythm

Writing where every sentence lands at a similar length reads as low variation, and low variation is one of the two main signals. Students trained toward clean, uniform paragraphs produce exactly this.

Conservative, predictable word choice

Choosing the expected word every time is good academic practice and also the definition of low perplexity. Second-language writers are hit hardest here, because caution with vocabulary is a learned, sensible strategy that happens to look machine-generated to a classifier.

Heavy grammar-tool editing

Accepting most of a grammar tool's rewrite suggestions smooths out precisely the irregularity the detector reads as human. Your voice gets sanded off, and the score goes up.

Formulaic structure

Five-paragraph templates, standard lab-report scaffolding, a rigid intro-body-conclusion shape. Predictable structure plus predictable vocabulary is the combination most likely to produce a false flag. More on this in why your writing gets flagged as AI.

If you have been accused

What to do in the first 48 hours

01

Preserve everything, immediately

Do not tidy up your drive. Export the Google Docs version history, save every draft file, screenshot your browser and library history for sources. Evidence you delete cannot be recovered later.

02

Ask what the claim actually rests on

Ask, politely and in writing, which tool was used, what score it returned, and whether anything other than that score supports the claim. Many cases rest on one number, and one number is contestable.

03

Build the timeline, not the argument

Lay out when you started, when each section appeared, what you cut. A revision history over nine days is the single most persuasive thing you can show, because generated text has no history.

04

Offer to talk it through

Volunteer to discuss your argument, your sources and your choices. If you wrote it, this is easy and it settles the matter faster than any document. Use your institution's appeals process and its student advocate.

Stay calm about the score itself

You are not arguing that the tool malfunctioned. You are pointing out what the tool measures and what it does not. Detectors report a statistical similarity. They have no access to authorship, and their published false positive rates, ours included, are not zero. That is a reasonable, factual position and it does not require you to attack anyone.

The step-by-step version, with wording you can adapt, is in how to prove you wrote your essay and how to prove you did not use AI on an essay.

Prevention

Four habits that make an accusation unlikely

Draft in a tool that keeps history

Google Docs or Word online. The revision timeline builds itself while you work and is the strongest evidence available, at zero effort. Never draft somewhere that leaves no trace.

Do this one

Keep your pre-edit draft

Before you run a grammar tool over a paper, save a copy. If the edited version scores high, the unedited one shows how much of the smoothing was not yours.

Two minutes

Put something only you could write in it

A specific seminar exchange, your own worked example, the source you rejected and why. Specificity lowers the score and improves the essay for the same reason: a model cannot produce it.

Helps the grade too

Scan it yourself first

If a passage scores high, read it. Usually it is the section where you were tired and wrote generically. Rewriting it in your own voice is the fix, not adding noise.

Before you submit
Common questions

What students ask after a flag

Because detectors measure how predictable your writing is, not who typed it. Clear, well-structured, conservatively worded prose produces the same statistical signature as model output. In our own 600-sample study 9 of 250 human-written samples were flagged as AI, and 4 of the 50 by non-native English speakers. A flag is a property of the text, not a finding about you.
Process evidence. A Google Docs or Word version history showing the paper built over days, intermediate drafts, outlines, browser or library history for your sources, and notes. Nothing beats a revision timeline, because it is the one thing a paste cannot produce. If you have it, lead with it.
It can raise your score. Grammar tools push writing toward the smooth, regular constructions that detectors read as machine-like, and heavy rewrite suggestions push it further. That is not the same as AI-generated, but a detector cannot tell the difference. Keep your pre-edit draft so you can show what changed.
Yes, and it is worth doing. Sign up free, your starting credits cover a scan, and one credit covers one word of analysis. Each scan handles up to 50,000 characters, about 8,000 words, on every plan including Free. Seeing which passages score high tells you where your own voice went missing, which usually improves the essay as well as the score.
Only if the rewriting makes it more yours. Chasing a number by adding deliberate errors or odd phrasing makes the writing worse and is not a defence if you are asked about it. The version of this that actually helps is adding specifics only you would know: the seminar argument, your own example, the source you rejected and why.

If this is happening to you

The pieces students actually come back to.

Check your essay before someone else does

Sign up free and your starting credits cover a scan. You get a score, the sentences behind it, and a clear read on which passages lost your voice.

Scan your essay free

Free credits on signup · No card required · Up to 50,000 characters per scan, about 8,000 words