ThesisDraft vs. Winston AI: A Subscription for a Two-Week Problem

What Winston AI and ThesisDraft Are

Winston AI is a subscription AI detector aimed at educators, publishers and writers. It has a polished interface, supports several languages, will read scanned documents and handwriting through OCR, offers plagiarism checking on higher tiers, and markets a headline accuracy figure of 99.98%.

ThesisDraft’s AI check does less, for one kind of person. A student, one thesis, in English or German: free for the first 1,500 words with no account, €4.99 once for the full document with 14 days of re-checks, and the measured false-positive rate shown next to every verdict.

Side by Side

A Subscription for a Two-Week Problem

Subscriptions are the correct shape for a teacher marking coursework all semester. They are the wrong shape for the thing a thesis student is actually doing.

The realistic pattern is this: you finish the draft, you check it, you dislike two sections, you rewrite them, you check again, you submit. That is a fortnight, not a year — and a monthly plan bought in that fortnight either gets cancelled in a hurry or quietly renews for months after your defence.

So the paid tier here is a single €4.99 payment that unlocks 14 days of unlimited re-checks. Fourteen days because that is the shape of the problem: long enough to fix and re-run as often as you like, short enough that we are not selling you a year of something you needed for two weeks. There is no account to close and nothing to cancel.

99.98% Is a Marketing Number

Not necessarily a false one — but it answers a question you did not ask. Headline accuracy is typically measured on a balanced set of clearly-human and clearly-machine documents chosen by the vendor. Machine text is easy to spot, so that number is dominated by the easy half.

Your question is narrower and harder: how often does this tool call verified human academic writing machine-written? That is the false-positive rate, it is measured on human documents only, and it is the number that decides whether a tool is safe to point at a student. Published research has found detectors falsely flagging a majority of non-native English writers — in a German university, that is not a corner case.

It is also not one number. It changes with language and with document length, which is why we publish four profiles instead of an average that would hide the weak ones.

The False-Positive Rate

Measured 9 August 2026 against human 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 at all. At that length nothing strong enough to catch AI costs fewer than roughly 7 false accusations in 100, and one innocent student in fifteen is not a trade we will make.

We also publish what those figures are not. Long-form results are in-sample. Zero observed false positives in 40 documents is still reported as 3 in 100, because the rule of three puts the upper bound near 7% at that sample size, and printing “0%” would be exactly the overclaiming we are objecting to here. Corpora and scripts: the accuracy page.

What Winston AI Does Better

• OCR and handwriting. If you need to check a scanned or photographed document, Winston reads it and we do not.
• Plagiarism scanning against a source index on the higher tiers.
• More languages than the two we have measured. We do not issue verdicts for languages we have not tested.
• Continuous-use tooling — scan history, team access, a dashboard built for repeated checking.

The Verdict

Choose Winston AI if you check documents every month, need OCR, or want plagiarism scanning in the same subscription.

Choose our AI check if you have one thesis, two weeks, and no interest in a recurring payment — and you want the error rate printed next to the verdict instead of a headline percentage on a pricing page.

Run the free AI check · See our measured error rates