GPTZero vs Turnitin.

Lervan makes an AI detector and published this guide. Lervan is not ranked in this comparison; the comparison is limited to what GPTZero and Turnitin publish about access, visibility, limits, claims, and interpretation.

GPTZero offers consumer self-serve purchase. Turnitin's AI-writing detection requires an institutional Turnitin Originality license. The vendor pages and help-center statements here were retrieved August 5, 2026.

Individuals can buy iThenticate credits, but Turnitin also says, “AI Writing Detection is not available for individual user packages, including individual credits or 3-credit packages.” Turnitin does not publish its pricing.

That makes the products structurally different for a student. GPTZero is a self-serve consumer product. Turnitin's indicator belongs to an instructor or administrator view, not a student preview.

Turnitin suppresses low numerical scores: “An asterisk (*%) appears when AI detection is below 20%, where results are less reliable and false positives are more likely.” It does not display a specific numerical score or source highlights in the 1% to 19% range.

Turnitin says its model list changes often, so this comparison does not freeze a list of models.

• GPTZero claims a “99% accuracy rate when spotting AI-generated text vs. human writing.” GPTZero claims “96.5% accuracy rate” on mixed human and AI documents. GPTZero claims ESL false positives fell “to 1.1%” on TOEFL texts.
• Turnitin claims a false positive rate “under 1% for documents with over 20% of AI writing”.

Weber-Wulff and colleagues concluded: “This paper exposes serious limitations of the state-of-the-art AI-generated text detection tools and their unsuitability for use as evidence of academic misconduct.” They also wrote that the tested tools “are neither accurate nor reliable (all scored below 80% of accuracy and only 5 over 70%).”

The study covered 14 tools and 54 documents, used GPT-3.5-era output, and ended in May 2023. Turnitin reached 81% on the logarithmic method alone, so the result cannot be turned into a blanket ceiling for every scoring method or current product.

Liang and colleagues found that the tested systems “incorrectly labeled more than half of the TOEFL essays as 'AI-generated' (average false-positive rate: 61.3%).” The seven detectors are never named, so that figure cannot be assigned to either product in this comparison.

That study used 91 TOEFL essays from a Chinese forum and is evidence about Chinese-L1 writers specifically. Enriching vocabulary lowered the measured false positives from 61.3% to 11.6%. It is also a 2023 study, not a current product benchmark.

Turnitin also states a content limit: “The model does not reliably detect AI-generated text in the form of non-prose, or code, nor does it detect short-form/unconventional writing”.

GPTZero says, “No AI detector is 100% accurate, and AI itself is changing constantly.” It also says, “We encourage using it as a conversation starter, and not as the final verdict.”

If you want a student-facing check before submission, use the AI detector. That pointer does not make its result equivalent to the hidden Turnitin verdict and does not turn either result into a finding of misconduct.