Blog

AI Detector False Positive: What to Do If You're Flagged

Why human writing gets flagged, what to do first, and how to show the work is yours.

By the AI Checker Humanizer team. Published . Updated .

What a false positive is, and how often it happens

A false positive is when a detector says human writing was produced by AI. It's a mistake by the tool, and every detector makes some.

How often depends on the detector and the text. Vendors usually publish low rates measured on long, unedited documents. Turnitin's AI writing detection FAQ, for example, says that fewer than 1% of fully human documents get labeled 20% AI or more, and a 2023 Turnitin post put the share of highlighted sentences that were actually human-written at about 4%. Independent studies on short pieces and on writing by non-native English speakers have found much higher rates for some tools: a 2023 Stanford study by Liang and colleagues, “GPT detectors are biased against non-native English writers” (Patterns), found seven detectors wrongly flagged 61% of TOEFL essays on average. (Sources checked November 11, 2026.)

These are the results for our own detector on English texts it never saw while we tuned it:

False positives and detection on our English test set
TextsHow manyResult95% bound
Human-written texts1,0961 flagged as likely AI (0.09%)At most 0.52%
Essays by non-native English writers6000 flagged as likely AIAt most 0.64%
AI-generated texts from 6 model families278270 caught as likely AI (97.1%)At least 94.4%
English, November 2026, detection engine version w0p5-38c6d591.

A low rate still means some people get flagged. If you're one of them, the rest of this article is for you.

Why human writing gets flagged

  • Formulaic assignments. Five-paragraph essays, lab reports and reflective templates have a fixed shape. AI produces that shape too, so the writing looks alike.
  • Short pieces. A discussion post or a 300-word response gives a detector little to work with. Ours doesn't give a verdict under 150 words (200 in Spanish, Portuguese and French).
  • Writing in a second language. Careful writers often stick to common words and safe sentence structures, which some AI detectors read as machine-like.
  • Different writing habits. Some people write in a very regular, literal style for all sorts of reasons, including how they process language. A 2025 paper by Chambers and Kelley, “The Misclassification of Autistic Writing as AI-Generated” (AIED 2025), found an older detector flagged Reddit posts by likely autistic writers more often than others, though the rates were low for both groups (checked November 11, 2026). The research here is still thin.
  • Heavy editing with tools. If a grammar or rewriting tool reworded whole sentences, those sentences can read like AI because, in part, they are.

What to do in the first 24 hours

  1. Don't reply in a hurry. Read the message carefully and note what you're being asked and by when.
  2. Ask for the report: which tool was used, the score, and which parts were flagged.
  3. Find your school's academic integrity policy and read the section on AI and on how cases are handled.
  4. Start collecting evidence of how you wrote the piece (see the next section). Don't edit or delete anything.
  5. Don't run the flagged work through a humanizer or paraphraser. Keep the submitted version unchanged so you can compare it with your drafts and explain how you wrote it.

How to prove you didn't use AI

Nobody can prove a negative perfectly, but you can show how the work came together. The strongest evidence is whatever was created while you were writing.

  • Version history. Google Docs (File → Version history) and Word on OneDrive or SharePoint keep earlier versions. A document that grew over several sessions, with sentences added, cut and moved, looks very different from one pasted in at once.
  • Drafts, outlines and notes. Paper notes count. So do photos of a whiteboard, messages to a classmate about the topic, or an outline in another file.
  • Your sources. Browser history, library loans, PDFs with your highlights, and the search terms you used.
  • If there's no version history. Maybe you wrote in a notes app, on paper, or pasted from another file. Say so plainly, then lean on what you do have: older files on your computer with earlier dates, emails to yourself, a photo of handwritten notes, or a classmate or tutor who saw you working on it.
  • Your understanding. Offer to explain your argument, why you chose your sources, and what you'd change. Being able to talk through your own work is often the most convincing evidence of all.

Talking to your teacher or integrity office

Go in calm and specific. Something like this works:

“I wrote this essay myself. I've brought the version history and my notes. Could you show me which parts were flagged so I can explain how I wrote them?”

Questions worth asking:

  • Which detector was used, and what exactly did it report?
  • Is there evidence other than the detector score?
  • What is the process from here, and can I bring someone with me?
  • Can I submit a statement or a writing sample done under supervision?

Take notes during the meeting, or follow up with a short email summarizing what was said and what happens next. If the outcome goes against you, look up your school's appeal process and its deadline straight away. A student advocate, ombuds office or writing center may be able to help you prepare.

Many schools say in their own rules that a detector score isn't enough on its own. St. Cloud State University's policy says “the use of AI detection software as the sole or primary basis for alleging an act of academic dishonesty is prohibited.” Cal State Fullerton tells faculty that “artificial intelligence detection reports alone are not sufficient to support a referral.” Turnitin itself says its AI score “should not be used as the sole basis for adverse actions against a student.” If your school has similar wording, point to it. (Quotes checked November 11, 2026.)

Why detectors disagree on the same text

It's common to paste the same essay into three AI detectors and get three different numbers. Each tool uses a different model, was trained on different writing, splits text differently, and sets its own threshold for calling something AI. Some report the share of text that looks like AI; others report the chance that the whole document is AI.

For example, Turnitin's percentage is the share of sentences its model labeled AI, while the score on our free AI detector is the chance that most of the text is AI-written. A 40% on one and a 40 on the other don't mean the same thing.

If you're accused, ask what each percentage measures and which passages were flagged. Different scores alone don't establish who wrote the essay. Our article on whether AI detectors are accurate compares what independent studies found.

Protect your next assignment

  • Write in a tool that keeps version history, and write there from the start rather than pasting in at the end.
  • Keep your outline, notes and research in one folder until the grade is final.
  • If you use a grammar tool, accept small fixes rather than whole-sentence rewrites, and note which tools you used if your course asks.
  • Before you submit, you can run your draft through our AI essay checker to see which sentences read like AI and why. Use it to spot passages that sound generic, not to chase a number.
  • Don't rewrite your own work until it no longer sounds like you just to lower a score. Your voice is part of your evidence.

If you're a teacher reading this, our AI detector for teachers page covers how to raise a flag with a student fairly.

0 / 2,000 words

Frequently asked questions

Can Turnitin be wrong about AI?

Yes. Turnitin itself says its AI score can be wrong and shouldn't be the only basis for action against a student. See how Turnitin's AI detection works.

Can I get in trouble from an AI detector alone?

It depends on your school, but many policies say a detector score isn't enough on its own. Ask what other evidence is being considered and what the process is.

Can Grammarly cause an AI flag?

It can, especially its generative features that rewrite or draft text. Turnitin says basic spelling and grammar edits weren't flagged in most of its own tests. If you used a tool, say which features you used and keep the original draft.

Are non-native English writers flagged more often?

Research on several popular detectors found they were. Results vary a lot by tool. On our English tests, none of 600 essays by non-native English writers was flagged as likely AI.

How do I show my writing process?

Share the document's version history, your outline and notes, earlier drafts and the sources you read. Offer to talk through how you built the argument.

Should I use a humanizer if I was falsely flagged?

No. Running work you're being asked about through a rewriting tool makes your process harder to show, not easier. Keep the original exactly as you submitted it and gather your drafts instead.