ThesisDraft vs. ZeroGPT: Free Is Not the Problem

What ZeroGPT and ThesisDraft Are

ZeroGPT is one of the most-used free AI detectors on the web. Paste text, press a button, get a percentage, no account needed. It is fast, it costs nothing, and it is supported by advertising and paid upgrades. For a great many students it is the first and only AI detector they ever use.

ThesisDraft’s AI check is also free to start — 1,500 words, no account, no name — and then €4.99 once for a full document with 14 days of re-checks. The difference is not the price. It is that every verdict we give arrives with the measured rate at which that verdict is wrong about a human, and the corpora and scripts behind that rate are published.

Side by Side

Free Is Not the Problem — Unaccountable Is

We have no objection to a free detector. Ours is free for the first 1,500 words precisely because a student should be able to see how a tool behaves on their own writing before paying anyone anything.

The objection is to a confident percentage with nothing behind it. When a tool publishes no error rate, no corpus and no method, there is no way for anyone — including the people who built it — to know whether it is working. That is not a hypothetical worry:

An early version of our own detector scored an AUC of 0.184 against 236 pre-2022 arXiv abstracts. A coin flip is 0.50. It was ranking machine-written text as more human than human-written text; inverting its output would have improved it. It had been returning confident percentages the whole time, and the failure was completely invisible from the output. We only found it because we ran it against a labelled corpus.

The cause was a family of lexical-richness features — Yule’s K, MTLD, the Zipf exponent — all running backwards. They score dense human academic prose as machine-like and lexically varied machine prose as human. That is the same mechanism behind the published finding that detectors falsely flag a majority of non-native English writers. Those features now carry zero weight here. A detector that has never run that test has no idea whether it carries the same defect.

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 is issued at all.

And the limits, published in the same place: the long-form weights were fitted on the corpus they were measured against, so those figures are optimistic; the AI corpora are 20–25 documents from two model families; the German long-form corpus is expository Wikipedia prose rather than thesis prose. Zero observed false positives in 40 documents is still published as 3 in 100, because the rule of three puts the upper bound near 7% at that sample size and printing “0%” would be the overclaiming we object to. It is all on the accuracy page.

What ZeroGPT Does Better

• Zero friction. No word ceiling at the free tier, no payment ever, no decision to make. For a quick look at a paragraph, that is genuinely convenient.
• Language coverage. It will accept text in far more languages than we have measured. We only publish error rates for English and German, and we do not offer a verdict we cannot stand behind in a language we have not tested.
• Familiarity. If your supervisor has heard of one free detector, it is probably this one.

Our Free Tier, and Where It Stops

1,500 words, free, no account, no name. That is an abstract, an introduction, or a section you are worried about — enough to see the tool work on your own writing and decide whether you trust it.

It is not enough to check a thesis, and we do not pretend it is. A full document is €4.99 once, with 14 days of unlimited re-checks so you can fix what you found and check again. We would rather tell you where the free tier stops than quietly truncate your document and score the part we read.

The Verdict

Use ZeroGPT for a free glance at a paragraph, in a language we have not measured, with the score read as a prompt rather than a finding.

Use our AI check when the document is your thesis and being wrongly flagged has a cost — because it is the one that tells you how often it is wrong, and shows you the corpus it learned that from.

Run the free AI check · See our measured error rates