ThesisDraft vs. Copyleaks: A Detector Built for the Other Side of the Desk

What Copyleaks and ThesisDraft Are

Copyleaks is an enterprise text-integrity platform: AI detection and plagiarism scanning in one product, sold mainly to universities, publishers and companies. It integrates with learning management systems, exposes an API, supports a long list of languages, and is metered in pages or credits across team plans.

ThesisDraft’s AI check is a single-purpose tool for one person: a student, in the last weeks of a thesis, who wants to know what a detector sees in their own writing. Free for the first 1,500 words with no account, €4.99 once for a full document, and every verdict printed beside the measured rate at which it is wrong about a human.

Side by Side

A 99% Accuracy Claim Is Not an Error Rate

Detector marketing runs on a single number — 99%, 99.9%, sometimes higher — and that number is close to useless for deciding whether to trust a verdict about your thesis. Accuracy on a balanced test set can be dominated by how easily the tool spots obvious machine text. It says almost nothing about the case you care about: verified human writing, formal register, written by someone whose first language is not English.

That is a specific, measurable quantity — the false-positive rate — and it is the one detectors do not publish. Published research has found that detectors falsely flag a majority of non-native English writers, which in a German university is not an edge case but a description of the typical submission.

A single accuracy figure also hides the fact that accuracy is not one number. It changes with language and with document length. We report four profiles rather than an average, because an average would conceal the weak ones.

The False-Positive Rate

Measured 9 August 2026, on human documents published before ChatGPT existed:

• English over 400 words — about 3 false positives in 100 (40 human / 25 AI documents).
• English under 400 words — about 2 in 100 (236 human / 20 AI).
• German over 400 words — about 4 in 100 (38 human / 25 AI).
• German under 400 words — no high-probability verdict at all, because nothing strong enough to catch AI at that length costs fewer than roughly 7 false accusations in 100.

Those samples are small, and we say so on the page that carries them. The long-form weights were fitted on the corpus they were then measured against, which makes the separation optimistic. Every corpus is named, dated, and produced by a script anyone can re-run — see the accuracy page. We publish it in that form specifically so it can be challenged.

What Copyleaks Does Better

• Plagiarism, properly. A real source index, matched against the web and licensed content. We do not do this and do not claim to.
• Language breadth. Dozens of languages against our two. If you are writing in Spanish or Turkish, we have measured nothing for you and will not pretend otherwise.
• Institutional plumbing. LMS integration, API, bulk scanning, team management — everything an examiner or a publisher needs, and nothing a student does.
• Both checks in one report. Convenient if you need both and are willing to pay institutional prices.

Buying a Tool Built for Your Examiner

The mismatch is not quality, it is design intent. An enterprise detector is built so an institution can screen many documents from people it does not know. Its interface, its pricing and its output are all shaped by that job.

A student has the opposite problem: one document, which they wrote, and a need to know whether it will be misread. That job wants a free first look, no account, no name attached to the text, a verdict in plain language, and — above all — an honest statement of how often the verdict is wrong.

It is also why we keep the two sides of our own house separate. Payment identity and document content live in different systems, joined only by a derived order code. We never ask for your name, university, supervisor or matriculation number, because those re-identify a person faster than a name does.

The Verdict

Choose Copyleaksif you need plagiarism detection against a real source index, many languages, or integration into an institution’s workflow.

Choose our AI check if you are the student holding the thesis, and what you need is a fast, anonymous verdict with its error rate attached. The first 1,500 words cost nothing.

Run the free AI check · See our measured error rates