AI detection in high school

The same tools, on shorter assignments, applied to writers still being taught the habits that raise a score. That combination produces more wrong flags, not fewer.

School assignments run through the same detectors universities use, usually inside the same submission platforms. What changes is the writing, and every difference points the same way.

Shorter work is scored less reliably

School assignments are short. A 500-word response gives a classifier a fraction of what a 3,000-word essay gives it.

Turnitin's own figures show the effect: 4% wrong on sentences against under 1% on documents. The smaller the unit, the worse the reliability, and short assignments sit near the small end. See why detectors are unreliable on short text.

Bar chart comparing wrong flags on genuine human academic writing. unslop 0.4 percent of documents against Turnitin 4 percent, a ten-fold difference.

Ours is 0.4%, which is 99.6% specificity, measured on 15,900 documents of real academic writing.

School writing is taught to be regular

This is the part that matters most, and it is nobody's fault.

The five-paragraph structure. Introduction, three body paragraphs, conclusion. Taught explicitly, produced by every student in the class, and structurally identical from one essay to the next. Repetition of shape reads as regularity.

Signposting is graded. "Firstly", "Furthermore", "Moreover", "In conclusion" are taught as markers of good structure and rewarded in rubrics. They open 2.4% of human sentences in our corpus and 5.8% of generated ones.

Bar chart comparing how often sentences open with a connective such as moreover or furthermore. Human writing 2.4 percent, generated writing 5.8 percent.

Topic sentences and tie-backs. Start each paragraph with its claim, end by restating it. This is exactly the paragraph template generated text produces.

Formal register is the safe choice. Students unsure of the expected tone write more formally, and formal is conventional, and conventional is predictable.

A student who follows the rubric closely produces the profile a classifier scores highest. A student who ignores it produces something more irregular and scores lower. That is the uncomfortable shape of it.

Writers in a second language

Published evaluations keep finding elevated wrong-flag rates, because learned English is applied more consistently. In a school with a large number of second-language students, the rate met in practice is above the published average. See why non-native English writing gets flagged.

What the score is

A probability that the writing has properties characteristic of generated text. It matches nothing and finds nothing, unlike the similarity score beside it on the same report, which points at a source you can open. See how Turnitin AI detection works.

What follows from one is set by the school's own procedures, which vary a great deal and which we are not the right people to describe.

What is practically available

Check the assignment before it goes in, look at which paragraphs come back flat, and make the two changes that move a score without touching the argument: vary sentence length, and cut the connective openings the rubric asked for wherever the structure survives without them.