Turnitin says my essay is 100% AI. What can I do?

What the percentage is

Turnitin’s AI writing indicator lives inside Feedback Studio and reports the share of a submission it estimates was written by AI. It does not know who typed the paper. It reads the finished prose and compares its patterns with what it has learned generated text looks like. That is why the number can be 100 percent on a paper you wrote alone at your kitchen table: the tool is scoring the text, not you.

Turnitin publishes a page titled “Understanding false positives within our AI writing detection capabilities” and states a document-level false positive rate under 1 percent for documents with more than 20 percent AI writing. Read that carefully. It is Turnitin’s own statement that it flags papers that were not written by AI. On a campus that runs 20,000 submissions a term through the tool, 1 percent is 200 papers.

Source: turnitin.com false positives page; checked 2026-09-08.

Why a number is not a finding

Three published results explain why a score cannot stand on its own. Stanford researchers (Liang, Yuksekgonul, Mao, Wu and Zou, Patterns, 2023) tested GPT detectors on TOEFL essays written by non-native English speakers and found 61.3 percent flagged as AI-generated. Common Sense Media (2024) reported that 20 percent of Black teens said they had been falsely accused of AI use, against 7 percent of white teens. A March 2026 report in Nouse found neurodivergent students disproportionately flagged. Detectors are drawn to prose that is formal, regular and grammatically careful, which is exactly what students are taught to produce.

Institutions have noticed. Vanderbilt turned off Turnitin’s AI indicator in August 2023 with a public statement. The University of Waterloo turned it off in September 2025 after an internal test flagged human-written text as 100 percent AI. Curtin University switched it off on January 1, 2026. Australian Catholic University, according to an ABC report in October 2025, had about 6,000 allegations, roughly 90 percent AI-related, with a large share dismissed. Northwestern, Yale and Johns Hopkins turned the indicator off across 2023 and 2024. Public trackers count 60-plus institutions. The full list has the dates and sources.

Source: Liang et al., Patterns, 2023; Common Sense Media, 2024; Nouse, March 2026; Vanderbilt Brightspace statement; Waterloo announcement; Curtin notice; ABC News Australia; GradPilot tracker; checked 2026-09-08.

A court has said it too. In Newby v. Adelphi University (New York Supreme Court, Nassau County, 2026 NY Slip Op 26021, early 2026) a student’s paper was flagged by Turnitin at 100 percent. Grammarly and ZeroGPT read it as human. The court found the university’s finding “without valid basis” and ordered the record expunged. Kato v. Palo Alto Unified School District, filed May 5, 2026 over a 76 percent flag, asks a federal court to bar detector scores from being treated as dispositive evidence.

Source: Justia; N.D. Cal. docket; checked 2026-09-08.

The next 48 hours

1. Reply briefly and calmly. A short note is enough for now: you received the message, you wrote the paper, and you would like to show your instructor how it came together.
2. Save everything. Drafts with their file timestamps, outlines, notes, annotated readings, the Google Doc or Word file with its version history untouched, and messages where you discussed the topic. Copy it all to a second location.
3. Ask what prompted the concern. Politely ask which passages were highlighted and which course or school policy applies. Knowing the passages lets you bring the draft or notes behind each one.
4. Offer to meet. Say what materials you have and ask for a time to walk through the paper together. Keep it short and friendly. The evidence page has sample wording.
5. Prepare to explain the paper out loud. Thesis, structure, the sources and why each is there, the paragraph you rewrote, what you would change now. You can also offer to write a short response on the same topic in front of your instructor. This is the evidence no tool can supply and the one instructors trust most.
6. Know the next step, just in case. If the conversation does not settle it, your school has an academic integrity process. A student advocacy office or the dean of students can explain how it works. The step-by-step page covers it.

What a writing record would show

If your course had you write in an editor that records the session, ask for that record to be reviewed first. A keystroke-level record shows what neither a detector nor a Docs history can: whether each passage was typed or pasted, the pauses where you stopped to think, the rewrites, and the order in which the argument came together. Its limit is just as clear. It shows how text entered the editor, not whether the ideas were yours, and it cannot be made after the fact. It is evidence for a conversation with your instructor, and that conversation is what you are working toward.

What not to do

• Do not lead with screenshots from other detectors. They show the tools disagree, not who wrote the paper. Your drafts and your explanation do that.
• Do not run your paper through a humanizer or replay it with an auto typer to build a history. That turns a false accusation into a true one.
• Do not edit, rename or delete any file. An incomplete record is normal; an edited one is damaging.
• Do not skip the meeting. It is where this gets resolved.

Talk to your instructor

Your instructor is not your opponent here. Instructors who use the indicator generally know its limits, and a conversation about the paper is often what they are hoping for. Bring the drafts, the history and the notes, be honest about what each shows and what it does not, and then explain the essay the way you would to a friend who asked what it was about. That is what proves you wrote it.