Why AI Detectors Flag Writing You Actually Wrote

You wrote it yourself. You remember writing it. A detector says it is 80 percent AI, and now you are trying to prove a negative to someone who has already made up their mind. This happens constantly, and it is not because you did anything wrong.

Detectors do not detect AI. They detect predictability.

No detector can see where text came from. There is no watermark and no signature. What they measure is how surprising each word is, given the words before it, plus how much your sentence lengths vary. Machine-written text tends to be smooth and even. So detectors learned a rule: smooth and even means machine.

The problem is that plenty of human writing is smooth and even too, and the humans who write that way are not randomly distributed.

Who gets flagged most

Stanford researchers tested detectors on essays written by non-native English speakers and found that more than half were wrongly flagged as AI, while essays by native speakers were classified almost perfectly. The finding was published in Patterns in 2023.

The reason is mechanical rather than mysterious. Writing carefully in a second language means reaching for vocabulary you are sure of and sentence patterns you trust. That produces exactly the evenness detectors were trained to punish. The better your discipline, the worse your score.

• Non-native English speakers, for the reason above
• Anyone taught to write in a formal academic register
• Technical and scientific writing, where precision means repetition
• Anyone who edits heavily, because editing smooths out the bumps

Universities noticed

Vanderbilt University turned off Turnitin AI detection in 2023 and published its reasoning: the false positive rate was unacceptable given the consequences for students, and the tool could not explain its own scores. Several other universities followed.

That is worth knowing if you are ever accused. Institutions have looked at these tools closely and some concluded they are not reliable enough to act on alone.

What to do if you are accused

• Keep your version history. Google Docs and Word both record it automatically, and it shows the document being built over time.
• Keep your notes, outlines and sources. A draft with mess in it is strong evidence.
• Ask which detector was used and what score it gave. Ask what that number actually means.
• Point out that detectors report a probability, not a finding, and that the vendors themselves say so.
• Offer to discuss your argument in person. Someone who wrote a piece can explain why they made each choice.

Where we stand on this

We built a detector, and we will tell you exactly what it is: evidence, not proof. It shows you a reading and the specific signals behind it, so you can judge it yourself rather than take a number on faith. No detector, ours included, can prove who wrote something. Anyone claiming otherwise is selling you a certainty that does not exist.