ThesisDraft vs. Scribbr: Two Different Products Under One Trusted Name

What Scribbr and ThesisDraft Are

Scribbris, for a great many German students, the name attached to getting a thesis finished properly. Its citation generator, its reference guides and its human proofreading service are genuinely good, widely used, and deserve their reputation. Its plagiarism check runs on Turnitin’s database, which makes it one of the few consumer-facing similarity checks with real institutional data behind it.

Scribbr also offers an AI detector. That is a different product from the ones it is famous for, and it should be evaluated separately from the brand it arrives with.

ThesisDraft’s AI check does only that one thing: it tells a student what a detector sees in their own writing, before anyone else runs one, and it publishes the measured rate at which its verdict is wrong about a human. Free for the first 1,500 words, no account; €4.99 once for a full document.

Side by Side

Trusting the Name Is Not the Same as Trusting the Detector

This is the whole argument of the page, and it is not a criticism of Scribbr’s other work. A service can be outstanding at citation formatting and human editing — both of which are labour and expertise problems — while its AI detector is a statistical model with the same limitations every statistical model has.

Those limitations are severe enough that the industry has largely stopped talking about them. Detectors flag formal, uniform prose. Academic writing is formal and uniform by convention. Published research has found detectors falsely flagging a majority of non-native English writers, which describes most students writing an English-language thesis at a German university.

None of that is fixed by a trusted brand. It is only made visible by measurement, and measurement only helps the reader if the result is published.

The False-Positive Rate

Measured 9 August 2026 on documents published before ChatGPT existed:

• English over 400 words — about 3 false positives in 100.
• English under 400 words — about 2 in 100.
• German over 400 words — about 4 in 100.
• German under 400 words — no high-probability verdict is issued at all.

With the limits stated in the same breath: long-form results are in-sample and therefore optimistic; the AI corpora are 20–25 documents from two model families; the German long-form corpus is Wikipedia prose rather than thesis prose. Every corpus and script is listed on the accuracy page, and we will help anyone who wants to reproduce or challenge a figure.

We publish this because we once shipped a detector with an AUC of 0.184 — worse than a coin flip — while it returned confident percentages. The failure was invisible from the output and obvious the moment it met a labelled corpus.

German Is Not a Translation of English

Most detectors treat language as a setting. It is not: a threshold calibrated on English text is simply wrong on German, and applying it anyway is how a German document gets a verdict its evidence does not support.

We measure the two separately, and they came out differently enough that they use different cutoffs. The consequence is a limitation we would rather state than hide: short German text never receives a high-probability verdict here. Under 400 words, no threshold strong enough to catch AI reliably costs fewer than about 7 false accusations in 100. One innocent student in fifteen is not an acceptable price, so at that length we say what we can support and no more.

What Scribbr Does Better

• Citation generation and reference guides. Free, thorough, and better than anything we offer, because we offer nothing here.
• Human proofreading. A real editor reading your thesis is a different category of help from any automated check.
• Plagiarism detection with Turnitin’s database. If similarity is your worry, that is the serious option.
• Breadth of student support — structure guides, methodology explainers, worked examples.

If you are choosing where to spend money on a thesis, Scribbr’s proofreading is a defensible purchase. Our AI check costs €4.99 and answers one narrow question the rest of that toolkit does not.

The Verdict

Use Scribbr for citations, editing, and similarity checking against a real database.

Use our AI check for the specific question of what an AI detector sees in your writing — with the error rate attached, thresholds measured separately for German, and no name required. The first 1,500 words are free.

Run the free AI check · See our measured error rates