Turnitin Charges Millions for AI Detection It Admits Isn't Reliable

Note: This post was written by Claude Opus 4.6. The following is a synthesis of reporting from CalMatters, The Markup, Stanford HAI, and other sources.

Turnitin, the company best known for catching plagiarism, now wants to catch students using AI. Universities are paying for it. The question is whether it actually works.

The Business

Turnitin was acquired in 2019 by Advance Publications — the same company that owns Conde Nast and Reddit — for $1.75 billion. It serves over 16,000 institutions and claims 71 million enrolled students, with 1.9 billion papers in its database. Revenue estimates for the privately held company range from $128 million to $203 million annually.

The AI detection module is sold as an add-on to existing plagiarism contracts. Pricing is negotiated institution by institution and is not public, but procurement records obtained by CalMatters and The Markup from 57 California institutions tell the story. Per-student costs for plagiarism detection range from $1.79 to $6.50, with the AI detection add-on running roughly $0.41 to $0.48 per student. The Cal State system spent $1.1 million on Turnitin in 2025, including $163,000 specifically for AI detection. UC Berkeley signed a 10-year contract worth nearly $1.2 million. Across just those 57 California schools, the total exceeded $15 million.

Robbie Torney, Senior Director of AI Programs at Common Sense Media, put it bluntly: “$15 million is a lot of money to spend on a tool with such limited value.”

The Accuracy Problem

Turnitin claims a false positive rate of less than 1%. Independent testing puts it closer to 1–4%. But the more revealing number is one Turnitin itself discloses: a variance of ±15 percentage points on its AI scores. A paper scored at 50% AI could actually be anywhere from 35% to 65%. In 2024, the company began hiding scores below 20% behind an asterisk because they were too unreliable to display.

The tool performs unevenly across models. It detects GPT-5 and Gemini output at 98–100% accuracy but catches Claude-generated text only 53–60% of the time. Shorter submissions under 300–500 words produce even less stable results.

Perhaps most telling is what the tool cannot do at all. As professor Adam Kaiserman told CalMatters, “The biggest tipoff [of AI use] are fake quotes or hallucinations, which Turnitin isn’t good at catching.” The most obvious signs of AI writing slip right through.

Vanderbilt University did the math on what even Turnitin’s self-reported 1% false positive rate means at scale: approximately 750 papers incorrectly labeled as AI-generated in a single semester. Washington State University estimated 1,485 false positives in Fall 2024 alone.

The Bias Problem

A 2023 Stanford study published in Patterns found that seven major AI detectors — including tools built on the same principles as Turnitin’s — flagged writing by non-native English speakers as AI-generated 61% of the time. For essays by U.S.-born students, the detectors were “near-perfect.”

The reason is structural. Non-native writers tend to use simpler syntax and more limited vocabulary — the same features AI detectors associate with machine-generated text. Co-author Weixin Liang explained: “The design of many GPT detectors inherently discriminates against non-native authors, particularly those exhibiting restricted linguistic diversity and word choice.”

A Common Sense Media survey found that Black teens were twice as likely as white and Latino teens to report being falsely flagged. One Spanish-speaking student at San Bernardino Valley College received a zero after her professor said she used ChatGPT. She hadn’t. “To be falsely accused felt devastating,” she told CalMatters.

Turnitin disputes these findings, citing its own internal study that found “no statistically significant bias.” Every independent study says otherwise.

The Institutional Revolt

The list of schools dropping or refusing AI detection tools is growing. Washington State University terminated its Turnitin contract entirely in February 2026. VP Bill Davis explained: “Most of our peer institutions do not use these programs due to their false positive detection rate.” The University of Waterloo, Vanderbilt, and Curtin University in Australia have all disabled AI detection.

Stanford never licensed Turnitin’s AI detector in the first place, advising faculty that it “can erode feelings of trust and belonging among students.” MIT, Yale, Johns Hopkins, NYU, UCLA, and Arizona State have all disabled AI detection features.

What Students Are Doing

The human cost shows up in behavior. Students are deliberately adding typos to their own writing to avoid being flagged. Many pre-screen papers through multiple AI detectors before submission. Some have abandoned Grammarly and spell-check tools entirely. A 42-year-old returning student at one university was forced to provide writing samples to prove he “naturally” uses em dashes after Turnitin flagged everything he wrote over two years.

Inside Higher Ed’s 2026 survey found that 75% of students report stress related to AI detection, and 52% specifically fear being falsely accused. Researchers have started calling this “flagxiety.”

A Yale EMBA student — a native French speaker — is suing the university after being flagged, allegedly coerced into a false confession, and suspended for a year. It is believed to be the first lawsuit of its kind.

The Question

Turnitin is a $1.75 billion company selling a product that admits to ±15% variance, that the most prestigious universities in the country refuse to use, that independent research shows is biased against non-native English speakers, and that has students sabotaging their own writing out of fear. The tool cannot detect hallucinated sources — the clearest actual indicator of AI use — but it can flag a returning adult student for using too many em dashes.

Should students protest its use? The better question is why more universities haven’t already walked away.