# Turnitin's AI Score: What It Can Prove and What It Cannot

Source: https://examfence.com/blog/2026-11-10-turnitin-ai-score-what-it-proves.html

[Home](../index.md) / [Blog](index.md)


**Short answer:** Turnitin's AI writing indicator scores each prose sentence, labels those above a threshold as AI, and shows the instructor the share as a percentage, from 20 percent up. The score is probabilistic, not proof of authorship, and Turnitin says it should not be used as the sole basis for action. Eight universities have switched it off since 2023, and the marker who treats the number as a verdict is making the same mistake the tool warns against.

## How the score is made

Turnitin's AI writing detection is part of the Similarity Report and needs a Turnitin Originality license. It reads the submission's prose sentence by sentence, assigns each sentence a score between 0 and 1 based on the sentence in context with the surrounding text, labels sentences above a threshold as AI, and reports the share as a percentage. Scores below 20 percent are not displayed: 1 to 19 percent shows as an asterisk, and under 300 words of prose gets no score at all. Lists, headers, references, data tables, and formulas are removed before scoring, so the percentage covers evaluated prose only. In October 2026, Turnitin updated the model architecture for English submissions to improve accuracy at transition points where text changes between human and likely AI-generated content.

The score is separate from the similarity score, and the two measure different things. The Similarity Report identifies matching text and source overlap. The AI Writing Report evaluates qualifying prose for text the model considers likely AI-generated or AI-modified. A submission can show 0 percent similarity and still be flagged as AI-written, or the reverse. Turnitin launched the indicator in April 2023, and the model can also identify text modified by AI paraphraser or humanizer tools designed to evade detection. [Contract cheating in 2026 runs on hidden AI](https://examfence.com/blog/2026-11-09-contract-cheating-proxies-hidden-ai.md), and the AI score is the tool most markers reach for first, which is why its limits matter more than its headline number.

## What the number cannot prove

Turnitin targets under 1 percent false positives on documents with over 20 percent AI writing, and says the score should not be used as the sole basis for action or a definitive grading measure. The reason is that the score is probabilistic and the model's performance varies with language, document type, text length, and mixed human and AI writing. False positives are possible, and some genuinely human writing reads as predictable: non-native English speakers, formulaic technical and scientific prose, and students writing in a second language are over-flagged more often than native speakers writing loosely. Several universities, including Yale, Vanderbilt, and Johns Hopkins, paused or de-emphasized the tool over exactly this concern, and at least eight universities have switched the AI score off since 2023.

The score is also invisible to the student by default. Only instructors and administrators see the indicator, and most institutions disable the student view for AI reports, so a student can receive a failing grade or an email from the academic integrity office without ever seeing the number that triggered it. That asymmetry is the fairness problem, and it is why the universities that kept the tool insist on a human process around it. The score is a prompt for further investigation, not a finding. [Take-home exams designed so bots cannot game them](https://examfence.com/blog/2026-11-08-take-home-exams-assessments-bots-cannot-game.md) work on the same principle: the assessment has to be robust to the tool, not dependent on detecting it.

## What a marker should do instead

Use the score as a lead, then ask the student to explain the essay and the argument: how they developed it, what sources they used, how they reached the conclusion. Ask for drafts or notes of early versions. A student who wrote the essay can usually talk through the argument and show the working. A student who did not will struggle with both, and that struggle is better evidence than any percentage. If the assessment permits AI use, the usage must be acknowledged and cited under the institution's policy. If it does not permit AI use, or the student used it without acknowledgement, that is academic misconduct, but the misconduct has to be established by the process, not by the score.

The other move is to design the assessment so the score matters less. Oral defenses, staged drafts, in-class writing, and questions that require the student's own data or fieldwork all reduce the value of generated text, because the generated text cannot supply the student's experience. Detection will keep improving and so will evasion, and the marker who depends on the detector will always be one model behind. The assessment that survives is the one that asks for something only the student can produce. [Cheating rings sell answers as a subscription](https://examfence.com/blog/2026-11-07-cheating-rings-subscription-answers.md), and the subscription model works because the assessment is the same every year. Change the assessment, and the subscription has nothing to sell.

## Frequently asked questions

### Does Turnitin detect ChatGPT?

Yes, if the institution licenses it and has it switched on. Turnitin scores each prose sentence and reports the share that looks AI-generated or AI-modified. The result is probabilistic, not proof of authorship.

### What does an asterisk mean on a Turnitin AI report?

It means the AI writing score is between 1 and 19 percent. Turnitin does not display scores below 20 percent. Under 300 words of prose gets no score at all.

### Can Turnitin's AI score be wrong?

Yes. Turnitin targets under 1 percent false positives on documents with over 20 percent AI writing, but non-native English speakers and formulaic technical prose are over-flagged more often. Eight universities have switched the tool off since 2023 over false positive concerns.

### Should a marker fail a student based on the AI score alone?

No. Turnitin says the score should not be used as the sole basis for action. Ask the student to explain the argument and show drafts or notes. The score is a prompt for investigation, not a finding of misconduct.

[Get a free bot-traffic audit](mailto:hello@examfence.com?subject=Free%20free%20bot-traffic%20audit)