best AI detectors for students, ranked by use case.

Lervan makes an AI detector and published this guide. We place Lervan in one use case only, name the competitors that win the others, and do not create a “most accurate” category from vendor self-claims.

Prices, plan limits, access rules, and vendor statements were retrieved from the vendors' own pages on August 5, 2026. The research section is kept separate because a vendor's claim and a peer-reviewed test answer different questions.

• Copyleaks: Personal is $16.99 monthly or $13.99 per month billed annually. One credit covers up to 250 words or one image. Copyleaks does not publish its free allowance or its maximum scan size.
• ZeroGPT: Free allows 15,000 characters per detection. PRO is $12.99 monthly or $9.99 per month billed annually.
• Winston AI: the 14-day Free trial lists 2,000 credits. Essential is $18 monthly or $10 per month billed annually. For AI detection, one credit equals one word.
• GPTZero: Premium is $23.99 monthly or $12.99 per month billed annually and lists 300,000 words per month. Professional is $45.99 monthly or $24.99 per month billed annually and lists 500,000 words per month.
• Turnitin: public pricing is not published. Its AI-writing submission must contain 300 to 30,000 words of long-form prose and be smaller than 100 MB.

Do not compare vendor credit totals as if they were one currency. Copyleaks defines one credit as up to 250 words or one image. Winston defines one AI detection credit as one word, two plagiarism credits as one word, and 300 credits as one image.

• 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 also 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”.
• Copyleaks claims “99% accuracy backed by independent third-party studies”.
• ZeroGPT claims it is “pushing toward >98% on internal evaluations.”
• Winston AI claims, “The only AI detector with a 99,98% accuracy rate”.
• Scribbr claims its own study named its premium detector the best available. Scribbr claims it “identified 84% of our texts correctly”. That roundup ranks Scribbr first and corporate sibling QuillBot second without disclosing the relationship on that page.

Weber-Wulff and colleagues wrote: “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: “Detection tools for AI-generated text do fail, they are neither accurate nor reliable (all scored below 80% of accuracy and only 5 over 70%).”

Read that result in its frame: the study tested 14 tools on 54 documents, used GPT-3.5-era output, and ended testing in May 2023. Turnitin reached 81% on the logarithmic scoring method alone, so the paper does not support rewriting the result as a universal ceiling.

Liang and colleagues wrote that the tested systems “incorrectly labeled more than half of the TOEFL essays as 'AI-generated' (average false-positive rate: 61.3%).” The study's seven detectors are never named. Do not attach that figure to any product.

That study used 91 TOEFL essays from a Chinese forum, so its evidence is specifically about Chinese-L1 writers. It also found that enriching vocabulary reduced false positives from 61.3% to 11.6%. Both studies are from 2023 and do not measure current products.

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

Turnitin says, “Turnitin does not make a determination of misconduct; rather, it provides data for the educators to make an informed decision based on their academic and institutional policies.” Scribbr says its result “should not be relied upon to provide absolute proof” and warns, “Never rely on AI detection alone to make decisions that could impact someone's career or academic standing.”

No disclaimer was found on the Copyleaks pricing or detector pages reviewed. No disclaimer was found on the ZeroGPT page reviewed. ZeroGPT instead markets PDF reports as “used as a proof of AI-Free plagiarism”; no other reviewed vendor makes that claim.

Pick the use case first, keep word and character limits in their original units, and read a result with the vendor's own disclaimer beside it. If your use case is checking your own draft before submission, open the AI detector. A detector result remains a signal to inspect, not a determination about authorship or misconduct.