← Back to blog

ThesisDraft vs. Turnitin AI Detection: The One You Cannot Run Yourself

|8 min read

Check your writing for AI text — free

The first 1,500 words are free, with no sign-up. Every verdict shows how often it is wrong about verified human writing — a figure no other detector publishes.

Need a researched draft to work from instead? From €4.99, anonymous, one-time payment — see pricing.

What Turnitin Is, and Why You Cannot Use It

Turnitin is the plagiarism and AI-detection system most European universities have already bought. It is wired into the submission workflow: you upload your thesis, and the report goes to your department. It carries an enormous similarity database built from decades of student submissions, journals and web content, and since 2023 it has added an AI-writing indicator on top.

The relevant fact for a student is structural rather than technical. Turnitin is not sold to you. Its customer is the institution, its report is addressed to your examiner, and in most setups the AI indicator is not visible to you at any point. You find out what it said when someone asks you about it.

ThesisDraft’s AI check is the opposite arrangement. It is sold to the student, it runs before submission, the first 1,500 words are free with no account, and every verdict is printed next to the measured rate at which that verdict is wrong about a human.

Side by Side

CriterionTurnitinThesisDraft AI check
Who can run itYour institutionYou, before submitting
When you see the resultOften never, or after submissionImmediately
Cost to the studentNot purchasable individuallyFree to 1,500 words · €4.99 full document
Published false-positive rateStated, corpora not releasedPublished with corpora and scripts
Similarity / plagiarism databaseVery large, institutionalNo database — phrasing check only
Sentence-level evidenceHighlighting in the reportSentence-level, with the reasoning shown
AnonymityTied to your student identityNo name, no account, no university
Refuses to give a verdictSuppresses low scoresYes — German under 400 words

No Detector Predicts Another

This deserves to be said plainly, because plenty of sites imply otherwise: running your thesis through our check does not tell you what Turnitin will report.Different model, different training data, different thresholds. Anyone selling you a “Turnitin score preview” is selling a number they cannot produce.

What a pre-submission check does give you is real, and worth the fifteen minutes:

  • Which passages in your own writing read as statistically machine-like — uniform sentence length, low variation, stock transitions. Those are the properties every detector keys on, so a passage that looks that way here has a decent chance of looking that way anywhere.
  • Whether the problem is concentrated (one chapter you drafted in a hurry) or spread across the document.
  • A written record, in the form of a report you can download, of having checked your own work before submitting it.

The False-Positive Rate

Turnitin has stated that its document-level false-positive rate is low, and after early criticism it stopped showing scores under a threshold rather than defend them. That is a more responsible response than most vendors managed. What has not happened is publication of the corpora, so nobody outside the company can reproduce the figure, and independent researchers have reported higher error rates — particularly for writers whose first language is not English.

Ours, measured 9 August 2026 on documents published before ChatGPT existed: about 3 false positives in 100 for English over 400 words, 2 in 100 for English under 400, 4 in 100 for German over 400 words, and no high-probability verdict at all for German under 400 words — because at that length no useful threshold costs fewer than roughly 7 false accusations in 100, and one innocent student in fifteen is not a trade we will make.

The caveats are published beside the numbers on our accuracy page, including the fact that our long-form figures are in-sample and our AI corpora are small. You can re-run every script.

What Turnitin Does Better

  • The similarity database. Nothing we do competes with it. Decades of student submissions and licensed journal content is a genuine moat, and for actual plagiarism detection it is the serious tool.
  • Institutional standing. Its report is the one your department acts on. Ours is informational.
  • Scale of validation. Whatever the criticisms, Turnitin has tested on volumes of student writing we cannot match.

If your worry is citation overlap rather than AI, Turnitin is measuring something we do not measure at all. Our phrasing check finds the stock academic sentences that inflate a similarity score; it does not search a database of sources, and we do not pretend it does.

If Turnitin Has Already Flagged You

A detector score is not proof of authorship, and no honest vendor claims otherwise. It is a reason to ask a question, never a reason to conclude an answer. If you wrote the work, the evidence that helps is the process: drafts with timestamps, notes, version history, your reading list, and your ability to talk through your own argument.

Our guide on what to do when a detector flags your thesis sets out how to respond, and our accuracy pageexists partly so that a student in that position has a citable source for the sentence “these tools have measurable error rates, and here is one published by a vendor”.

The Verdict

They are not substitutes, and this comparison is not really a contest. Turnitin is what will be run on your thesis. Our check is what you can run first, on your own terms, without your name attached — and the only one of the two that hands you its error rate along with its verdict.

Run the free AI check · See our measured error rates

Frequently Asked Questions

Can I check my thesis in Turnitin before submitting?

Usually not. Turnitin sells to institutions, not to students. Some universities offer a practice submission through their learning system, but many do not, and the AI indicator is often hidden from students even when the similarity report is shown. In most cases the first time a Turnitin AI score exists for your thesis is after you have submitted it.

Will your AI check tell me what Turnitin will say?

No, and no tool can. Detectors are different models trained on different data with different thresholds, so their scores are not interchangeable. What a check here tells you is whether your own writing carries the statistical signals detectors look for in general — heavily uniform sentence rhythm, low burstiness, formulaic transitions. Those are what get flagged, whichever tool does the flagging.

How accurate is Turnitin's AI detection?

Turnitin has stated a low false-positive rate at the document level and, in response to criticism, stopped displaying scores below a threshold. Those figures come from internal testing on undisclosed corpora. Independent researchers have reported higher error rates, particularly for non-native English writers. We publish our own false-positive rate with the corpora and scripts behind it, so it can be challenged.

Does using an AI checker before submitting count as cheating?

No. Checking your own writing for statistical properties is not a form of academic misconduct in any code of conduct we are aware of — it is the same thing your examiner does, run earlier by you. What is misconduct is submitting work you did not write. We do not sell tools to disguise that, and we say so when a document is largely machine-written.

Check your writing for AI text — free

The first 1,500 words are free, with no sign-up. Every verdict shows how often it is wrong about verified human writing — a figure no other detector publishes.

Need a researched draft to work from instead? From €4.99, anonymous, one-time payment — see pricing.