Essay Flagged as AI? What to Do and How to Check First

Flagged by Turnitin, GPTZero or a teacher's detector for work you wrote yourself? Here is what to gather today, how to self-check with a free detector that shows its reasoning and does not report to your school, how to answer the accusation without sounding defensive, and how a ScanForAI class code works if your teacher uses one.

You wrote it, and a detector says a machine did. That happens to a lot of people, for reasons that have nothing to do with honesty, and it is survivable if you act in the right order: gather evidence first, self-check second, respond third. This page is the student side. Our partner ScanForAI publishes a student page with a line worth knowing before you start: it does not report to institutions, full stop.

First, what the number is and is not

A detector score is not a percentage of your words written by AI, and it is not proof. Detectors measure how predictable your text is. Formal register, textbook transitions, a narrow vocabulary and short length all push the number up, which is why non-native English writers and careful, rule-following students are flagged most. If you want the mechanics and the research, read why AI detectors flag human writing. For now, the practical point is that the score is the start of a conversation, not the end of one, and your job is to bring better evidence than a number.

Gather your evidence today

Do this before anything else, because some of it disappears.

  • Version history. In Google Docs, File, then Version history, then See version history. In Word, open the file from OneDrive and use Version History. Screenshot the timeline showing the essay growing over days, not appearing in one paste. If you wrote offline, file timestamps and autosave versions still help.
  • Notes, outlines and earlier drafts. Paper notes count. Photograph them.
  • Sources and search history. Export the browser history for the days you researched. Keep the PDFs and library records you used.
  • Messages. Texts to a friend about the essay, questions to the teacher, a tutor's comments, anything dated.
  • AI chat logs, if you used an assistant. Using ChatGPT or Gemini to brainstorm or check grammar is allowed in many courses when disclosed. The log shows what you asked for and what you wrote yourself. Our AI chat export cleaner turns a messy log into a clean PDF you can attach.
  • A before-and-after. If you revised after a self-check, our text diff tool shows exactly what changed.

Self-check with a detector that shows its work

Run your own scan, for two reasons: you want to know which sentences are being doubted, and you want your own report in hand before the meeting. Pick a tool that shows line-level results and the signals behind them rather than a bare percentage. ScanForAI does this without an account, for up to 20,000 characters, and does not store anonymous scans.

  1. Paste the whole essay or upload the file. Fragments score unreliably; ScanForAI says that under about 300 words most of its signals abstain or wobble.
  2. Set the Text type to Student essay. Detectors judge an essay and a blog post by different standards, and this tells the tool which one it is reading.
  3. If English is not your first language, tick English isn't my first language. It relaxes the formality signals that cause most false positives for non-native writers.
  4. Read the line view. The doubted sentences are usually the generic ones: the topic sentence that could open any essay on the subject, the transition every textbook uses. The bands are Low under 25, Moderate 25 to 49, Elevated 50 to 71 and High 72 and over; ScanForAI's own guide calls Moderate normal for edited or formal writing and tells readers to treat Elevated and High as a reason to ask, never a verdict.
  5. Save the PDF report. No account is needed for that. With a free account you also get a share link and a year of history, useful if you want to show the same report later.

If you are still revising, replace the doubted sentences with your specific examples, your data and your opinion, and vary sentence length. That is ordinary editing and it makes the essay better. Do not run it through a paraphraser or a so-called humanizer: it makes the text less yours, it rarely changes the score reliably, and ScanForAI itself says it is not a tool for hiding AI use.

If your teacher uses a ScanForAI class code

Some teachers hand out a ScanForAI class link or a six-character code. The design is consent-first: you scan your own essay, see the result before anyone else does, then decide whether to share the full report or only the number. Nothing leaves your screen until you press Send, and the report you keep is identical to the one your teacher receives. What the teacher can and cannot see is covered in our guide for teachers. A teacher who uses this is treating the score as feedback rather than surveillance, which is good news for you.

How to respond to the accusation

  1. Ask for specifics in writing. Which tool, what score, which passages, and which policy you are said to have broken. Ask for the appeal procedure at the same time.
  2. Stay factual. Do not argue about whether detectors are fair in general; show your process. Attach the version history, notes and sources.
  3. Explain the false-positive mechanics briefly. One paragraph: detectors flag predictable writing; a 2023 Stanford study found seven popular detectors flagged 61% of essays by non-native English writers on average; OpenAI withdrew its own detector in 2023 for low accuracy. Cite, do not lecture.
  4. Offer to demonstrate. Propose a short meeting where you discuss the argument, or a supervised writing sample. Students who wrote their essay usually welcome this; it is the strongest move you have.
  5. Escalate calmly if needed. If the only evidence is the score and the decision stands, use the appeal procedure, and involve the student union or ombudsman.

A short written response could read:

Thank you for raising this. I wrote the essay myself and I would like to show you how. Attached are the document's version history across [dates], my outline and notes, and the sources I used. I understand the detector produced a score of [n]; detectors are known to flag formal and non-native writing, and the vendors themselves advise against treating a score as proof. I would welcome a short meeting to discuss the essay's argument or to write a passage under supervision. Please also send me the written policy and the appeal procedure.

Before you submit next time

  • Write in a tool with version history, from the first sentence.
  • Keep your sources and notes until the grade is final.
  • If you used an AI assistant, follow the course rule on disclosure and keep the log.
  • Self-check the final draft and revise generic sentences in your own voice.
  • Keep the report. If a question comes later, you already have your evidence.

Check your essay on ScanForAI

Related guides: Why AI detectors flag human writing · For teachers: free Class links and a fair policy · How to use ChatGPT effectively.

Disclosure: ScanForAI is a partner site. Links on this page carry tracking tags so they can see visitors arriving from SimpleToolsHub; we don't earn a commission, and everything described here is free to use. This page is general guidance, not legal advice; your institution's policy governs. We make no accuracy claims for any detector.

Frequently Asked Questions

No. A detector score is a signal, not evidence of how a text was written. Detectors flag plenty of human writing, especially formal and non-native English, and the tools themselves say so. What decides an integrity case is your process evidence: version history, drafts, notes, sources and your ability to discuss the work.

Not that 40% of the words were written by AI. Depending on the tool it is either the model's confidence or the share of sentences that crossed a threshold, and short or formal texts swing wildly between scans. Ask which tool was used and what the vendor itself says the number means; most vendors warn against acting on it alone.

Detectors read a narrower vocabulary and textbook grammar as machine-like. A 2023 Stanford study found seven popular detectors flagged 61% of TOEFL essays on average. Say this plainly in your response and point to the study. When you self-check on ScanForAI, tick English isn't my first language, which relaxes those formality signals.

No. Its student page says it does not report to institutions, full stop, and anonymous scans are not stored. With a class link, nothing is shared until you choose to share it.

No. Scanning your own text and revising generic sentences in your own voice is editing. What ScanForAI explicitly says it is not is a tool for hiding AI use: if you pasted AI-generated text, no amount of scanning makes it your work, and paraphrasing tools make it less yours, not more.

Spelling and grammar fixes rarely change a score. Heavy use of rewrite or paraphrase features can, because they push sentences towards the statistical average that detectors associate with AI. Say which features you used, and keep the original draft to show the difference.

Version history from Google Docs or Word showing the essay growing over time, dated notes and outlines, your sources and browser history, earlier drafts, and your own detector report showing which lines were doubted and your revisions. If you used an AI assistant for brainstorming, the chat log helps too.

Yes. ScanForAI accepts .docx, .odt, .pdf, .txt, .md and .html uploads up to 1 MB. Without an account you can paste up to 20,000 characters; a free account raises that to 50,000 and keeps your history for a year.

Ask for the written policy and the appeal procedure, respond in writing with your evidence attached, and ask for a meeting where you can discuss the essay. If the only evidence against you is a detector score, say so and cite the detector's own caveats. Student unions and ombudsman offices exist for exactly this.